Unconstrained Explanations, Ad Hoc Reasoning, and Divine Psychology
- Introduction: When an Explanation Explains Too Much
- What an Explanation Has to Do: Possibility Is Not Probability
- Scharp's Divine Psychology: The Exemplary Case
- Ad Hoc Explanation: From Philosophical Intuition to Bayesian Formalism
- Divine Psychology as a Problem of Latent Parameters
- Miracles as Agentic Explanations
- Three Case Studies: Resurrection, the Shroud of Turin, and Fatima
- Revelation, Hiddenness, and the Free-Will Auxiliary
- Intelligent Design: Adding an Agent Without Adding a Theory
- The Generalized Pattern: The Unconstrained Agent as Explanatory Wildcard
- Objections, Qualifications, and What Would Count as a Successful Divine Explanation
- Conclusion: From Explanatory Flexibility to Explanatory Discipline
This essay argues that explanatory power depends not merely on whether a hypothesis can accommodate an observed outcome, but on whether it independently constrains the space of possibilities strongly enough to make that outcome more expected than relevant alternatives. Kevin Scharp's problem of divine psychology provides an exemplary case: once divine or intelligent agency is invoked, causal capability alone explains very little unless there is an independently justified account of the agent's motives, preferences, and decision procedures. Without those constraints, appeals to miracles, revelation, Intelligent Design, and similar hypotheses risk turning already-known data into retrospectively fitted stories—achieving narrative intelligibility without corresponding predictive or evidential force.
The central distinction throughout is:
and, more broadly:
Introduction: When an Explanation Explains Too Much
Explanations do more than make events intelligible. A good explanation should tell us why one outcome occurred rather than another, and in doing so it should place constraints on what we would have expected to observe. This requirement is easy to overlook because explanation in ordinary language is often satisfied by the construction of a coherent story. If an event occurs and we can subsequently describe some circumstance, motive, or mechanism under which the event would make sense, we are tempted to say that it has been explained. But there is an important difference between showing that an observation can be accommodated by a hypothesis and showing that the hypothesis gave us reason to expect that observation in advance. The former requires compatibility; the latter requires constraint.
This distinction becomes especially important when explanations appeal to intelligent agency. In ordinary contexts, explanations in terms of persons are often highly informative. If someone is accused of committing a crime, for example, it matters whether that person had the physical ability to perform the act, whether they were present, whether they possessed the required knowledge or means, whether they had an intelligible motive, and whether the alleged action is consistent with what is otherwise known about their behavior. Human actions are not perfectly predictable, but they are constrained by a vast body of background knowledge about psychology, physiology, incentives, social relations, geography, technology, and circumstance. The attribution of an intention therefore narrows the relevant possibilities rather than merely adding another entity to the story. Similar constraints operate in scientific explanation: physical, biological, psychological, and social mechanisms make some outcomes more likely and others less likely.
The situation changes considerably when an explanation introduces an agent whose capacities are extremely broad while whose motives remain poorly specified. If an agent can perform almost any logically possible action, then establishing that the agent could have produced an observed event accomplishes very little. The central explanatory problem instead becomes one of selection: why would this agent choose this action, at this time, in these circumstances, directed toward these individuals, rather than one of the indefinitely many alternatives available? An agent with unlimited causal powers but an unknown decision procedure may be maximally powerful as a cause while being minimally informative as an explanation.
Kevin Scharp's notion of divine psychology provides an exemplary case of this problem. Arguments for divine action regularly contain judgments about what God would want, prefer, permit, reveal, create, or choose. These judgments are often implicit. A phenomenon is said to be evidence for God because it is described as something a loving, rational, creative, or purposive deity would plausibly bring about. Yet the inference from divine attributes to particular divine choices is not automatic. The proposition that God is capable of producing some outcome does not establish that God would be disposed to produce it, and the fact that one can imagine a coherent divine reason for an outcome does not establish that the outcome was probable under theism.
This problem is especially vivid in miracle and revelation claims. A miracle is not simply an unusual occurrence. To identify an anomalous event as a miracle is already to move from description to explanation: some agent has intervened, that agent is divine, the intervention was intentional, and in many religious contexts the event is further interpreted as communicating or authenticating a particular theological message. The historical or observational evidence may establish that something unusual occurred while leaving these further intentional claims radically underdetermined. The explanatory burden therefore cannot be discharged merely by showing that ordinary mechanisms have difficulty accounting for every detail. A miraculous explanation must itself specify what God is alleged to have done, why God would have done it, and why the particular action observed should be regarded as a reasonable means of realizing the attributed purpose. Treating miraculous and non-miraculous explanations symmetrically requires applying the same general demands of explanatory scope, plausibility, background knowledge, and causal or intentional fit to both.
The difficulty is compounded when the motives attributed to the agent are supplied only after the evidence is known. A theory can always be made easier to reconcile with an observation by introducing additional assumptions tailored to that observation. But an improvement in retrospective fit is not automatically an improvement in evidential support. An auxiliary hypothesis introduced principally to protect or accommodate a theory, without independent evidence or broader theoretical motivation, is paradigmatically ad hoc. Such modifications can make the past easier to explain while simultaneously weakening the theory's ability to predict the future, because greater flexibility allows a wider range of possible outcomes to be absorbed after the fact. This distinction can be stated compactly:
and, more specifically,
These distinctions provide the organizing framework for what follows. Divine psychology will serve as the clearest theological example, but the underlying problem is much broader. The same structure appears in appeals to miracles, resurrection claims, Marian apparitions, the Shroud of Turin, theories of revelation and divine hiddenness, and arguments from intelligent design. In each case an intelligent agent is introduced to account for already existing data, and in each case the explanatory force of that move depends upon whether there is an independently justified account of what the agent would have been likely to do. Without such constraints, the ability to invent a motive consistent with the observation risks producing narrative intelligibility without genuine predictive discrimination.
The central question of this essay is therefore not whether divine or intelligent agency is metaphysically possible, nor whether an agent could in principle produce the phenomena under consideration. It is more fundamental: what must be known about an agent before invoking that agent adds explanatory value? If explanation requires distinguishing the actual world from relevant alternatives, then an unconstrained agent presents a peculiar difficulty. The more actions available to the agent, and the less we know about how the agent chooses among them, the easier it becomes to accommodate whatever happens—and the harder it becomes to claim that what happened was genuinely predicted. The result is a general tension between explanatory flexibility and explanatory power, one that becomes especially pronounced when theology or design theory attributes intentions retrospectively to an agent whose psychology is otherwise inaccessible.
What an Explanation Has to Do: Possibility Is Not Probability
A useful way to distinguish strong explanations from weak ones is to ask how much they constrain the space of possible observations. An explanation is not powerful merely because the event in question is compatible with it. It is powerful insofar as the hypothesis gives us reason to expect some outcomes more than others. This point is especially important in debates over unusual or anomalous phenomena, because an event can be difficult to explain without thereby favoring every hypothesis capable of accommodating it.
Suppose the evidence to be explained is $E$, while a number of alternative outcomes were also possible:
A hypothesis has genuine explanatory force when it does not treat all of these possibilities alike. It should narrow the range of outcomes that are plausible under it. The more indiscriminately a hypothesis accommodates possible observations, the less evidential significance any one of those observations has for the hypothesis.
This is why causal explanations grounded in established bodies of knowledge are often powerful even when they remain incomplete. Physical, chemical, biological, psychological, and social processes are constrained by background theory. We may not know every boundary with precision, but we know enough to regard some outcomes as common, some as unusual, and some as effectively impossible under specified conditions. Human behavior is similarly constrained. People are not perfectly predictable, but we possess extensive background knowledge about motives, capacities, incentives, memory, social influence, location, communication, and available means. These constraints allow us to assess whether a proposed explanation is antecedently plausible rather than merely conceivable.
The explanatory value of such constraints can be seen by contrast. If a theory permits virtually every possible outcome, then observing the actual outcome does little to distinguish that theory from its competitors. The problem is not that the theory cannot account for the evidence. It is that it can account for too much. A theory that predicts only a narrow class of outcomes has taken an epistemic risk; a theory that allows almost anything has not.
This is the point behind the distinction between possibility and probability. In many apologetic contexts, these are easily conflated. Consider the claim:
God could have had a reason to bring about $E$.
That establishes, at most, that $E$ is compatible with the hypothesis under consideration. In probabilistic terms, one might say only that:
But evidential support requires something stronger. If $E$ is being offered as evidence for God rather than against a competing explanation, one needs some reason to think that $E$ is comparatively expected under the divine hypothesis:
where $H$ is a relevant alternative hypothesis.
The difference is substantial. A state of affairs can be perfectly compatible with a theory while providing little or no support for it. The mere possibility of constructing a divine rationale for an observation does not establish that the observation was probable given God, still less that it was more probable under theism than under alternative explanations.
This is where ordinary words such as plausible, fitting, natural, unsurprising, and what one might expect can become misleading. These terms are often used loosely enough that they slide between different epistemic claims. “It is plausible that God would do this” may mean nothing more than “I can think of a coherent reason why God might do this.” But “this is evidence for God” requires a comparative probabilistic judgment. It requires some reason to think that the actual observation occupied a privileged place within the hypothesis's prior expectations.
This distinction also clarifies the difference between narrative intelligibility and evidential explanation. Narrative intelligibility asks:
Can I tell a coherent story under which this event makes sense?
Evidential explanation asks:
Did my theory, before I saw the event, assign relatively high probability to something like it?
The first task is substantially easier. Human beings are extremely good at generating coherent stories after outcomes are known. Once an event has occurred, additional assumptions, motives, and contextual details can often be introduced to render it understandable. But retrospective intelligibility should not be confused with predictive success.
This point becomes clearer when considering how explanations are assessed comparatively. A naturalistic explanation may be incomplete. It may invoke an unusual psychological response, an uncommon meteorological event, an improbable social process, or some combination of mechanisms whose exact details are uncertain. None of this automatically disqualifies it. Rare events occur, and our knowledge of ordinary processes is incomplete. The relevant question is whether the proposed account is sufficiently connected to independently established mechanisms and constraints to make the observation intelligible without excessive modification. The miracle material emphasizes precisely this point: an unexplained detail or a low-probability natural process does not automatically favor a divine explanation, because the divine explanation must itself be evaluated for plausibility, scope, and explanatory power rather than treated as the default once natural explanations are imperfect.
This is why the “method of exclusion” is epistemically weak. The reasoning takes a form such as:
But this inference moves from the failure of one class of explanations to the success of another without independently establishing the latter. Eliminating or weakening a naturalistic hypothesis does not by itself show that an agentic or supernatural explanation has the properties required of a good explanation.
The proper comparison therefore requires symmetry. If one explanation is criticized for lacking a mechanism, relying on improbable processes, or leaving details unresolved, the alternative must also be subjected to comparable scrutiny. A miraculous explanation cannot gain an evidential advantage simply by being less specified. The question is not whether the naturalist has answered every possible question, but whether the available natural explanation is stronger or weaker than the rival divine explanation when both are assessed under the same standards.
This gives the essay its first general principle:
An agent's ability to produce an outcome is not yet a reason to expect that outcome. An explanation becomes informative only when some independently grounded account of the agent, mechanism, or theory makes the actual observation stand out among the alternatives.
The same point can be expressed in a second form:
A hypothesis may contain a coherent story in which the evidence makes sense while still assigning very little antecedent probability to that evidence. This distinction will become especially important once divine agency is introduced, because a being with extremely broad causal powers enlarges the set of possible outcomes rather than narrowing it. Unless there is an independently justified account of what such a being would choose, the explanatory problem shifts away from capability and toward intention. That is precisely the problem captured by divine psychology.
Scharp's Divine Psychology: The Exemplary Case
Kevin Scharp's idea of divine psychology captures the problem in a particularly direct form. The phrase refers to the attribution of motives, intentions, preferences, and decision-making tendencies to God. Whenever an argument claims that some feature of the world is evidence for theism because it is the kind of thing God would want, value, create, reveal, permit, or bring about, it relies on a substantive model of divine psychology.
This is easy to miss because theological arguments often move rapidly from very general divine attributes to highly specific expectations. One begins with claims such as:
God is perfectly good.
God is rational.
God is loving.
God is omniscient.
God is omnipotent.
Then, often without much further argument, one arrives at claims such as:
Therefore God would likely create embodied moral agents.
God would likely create a life-permitting universe.
God would likely reveal himself.
God would likely authenticate a prophet.
God would likely intervene to vindicate a religious message.
But the inferential gap between these two sets of claims is enormous. General attributes do not automatically determine particular choices. A perfectly good being might have many possible goods available, many ways of realizing them, and perhaps many reasons not to instantiate some of them at all. Omnipotence expands the range of available actions, while omniscience eliminates ignorance about their consequences. Neither attribute by itself tells us which action would be selected.
The core question of divine psychology is therefore not:
Can God produce $E$?
but:
Why should we expect God to prefer $E$ to the relevant alternatives?
This difference is critical. Let:
If an observed fact $E$ is offered as evidence for $G$, the relevant quantity is something like:
Yet the existence of God by itself does not determine this probability. Once one begins to say what God would or would not choose, the hypothesis has been enriched by further assumptions about divine intention.
A more realistic structure is:
where:
- $G$ represents the existence and general attributes of God;
- $I$ represents divine intentions or preferences;
- $C$ represents the relevant circumstances;
- $A$ is the action selected by God;
- $E$ is the resulting observation.
The probability relevant to the specific explanation is therefore closer to:
than simply:
That distinction matters because the intentional component $I$ is frequently doing most of the explanatory work. If someone says that a miracle occurred because God wanted to vindicate a revelation, it is the attributed desire to vindicate that revelation that makes the action intelligible. The bare proposition that God exists contributes causal capacity, but not much information about why this event should occur.
This is precisely where the problem of constraint reappears. Human psychology is imperfectly understood, but it is richly constrained. We know that humans experience hunger, fear, jealousy, affection, ambition, status-seeking, loyalty, grief, resentment, and many other motives. We possess large bodies of knowledge from ordinary social experience, psychology, sociology, neuroscience, history, and law that help us estimate what kinds of behavior are plausible in particular circumstances. By contrast, the available theory of divine motivation is far thinner. The miracle analysis emphasizes that while naturalistic explanations can appeal to extensive background knowledge about mechanisms and human behavior, divine explanations often have little to work with beyond whatever theological characterization is supplied for the specific case.
This does not mean that divine psychology is impossible in principle. A sufficiently detailed and independently justified theology might constrain expectations about divine action. One could imagine a religious theory specifying divine values, priorities, preferred forms of communication, willingness to intervene, attitudes toward ambiguity, and other relevant features in enough detail to generate genuine predictions. The difficulty is that such a theory must be justified independently of the very observations it is later invoked to explain. Otherwise, the reasoning becomes circular or ad hoc.
Suppose, for example, that one observes a highly localized religious event and then infers:
God prefers localized revelation because it preserves faith.
If the only reason for attributing that preference to God is that the observed revelation was localized, then the explanation has been constructed from the data. One cannot then turn around and claim that the localization strongly confirms the divine hypothesis because it is exactly what a God with that preference would do. The preference was inferred from the same event now being used as evidence.
The issue is therefore not simply that we lack access to God's inner life. The deeper problem is that an explanatory appeal to agency requires a theory of selection. An agent explanation does not merely say that an agent possessed the ability to produce an outcome. It says that the agent chose that outcome for some reason.
This creates a contrastive demand:
Why this action rather than another?
If the action is a miracle:
Why this miracle rather than another miracle?
If the action is revelation:
Why this revelation rather than a clearer or more universal one?
If the action is creation:
Why this universe rather than another?
If the action is resurrection:
Why these appearances rather than broader appearances?
If the action is biological design:
Why this structure rather than another?
The more extensive the agent's capacities, the more important this contrastive question becomes. A limited human engineer may choose a suboptimal method because of cost, available materials, ignorance, time pressure, or technological limitations. An omnipotent and omniscient agent cannot be explained in the same way. If the proposed divine action appears oddly inefficient relative to its alleged purpose, one cannot appeal to ordinary constraints without adding further theological assumptions.
This produces what might be called the inscrutability dilemma.
On one side, theological argument sometimes appeals confidently to divine psychology:
God is loving, therefore this is something we should expect.
God wants relationship, therefore revelation is expected.
God values moral agency, therefore a certain kind of world is expected.
On the other side, when observations seem difficult to reconcile with those expectations, the response is often:
God's reasons may be beyond our understanding.
But these two strategies pull in opposite directions.
If divine motivations are sufficiently knowable to generate strong evidential expectations, then apparently contrary observations should count against those expectations.
If divine motivations are sufficiently inscrutable that almost any observation can be reconciled with unknown higher reasons, then the hypothesis becomes less predictively specific.
Formally, if the divine hypothesis permits a very broad range of possible outcomes:
and there is no principled way to assign substantially greater probability to one rather than another, then observing any particular $E_k$ provides relatively weak confirmation.
This yields an important general principle:
The same flexibility that protects a hypothesis from counterevidence can weaken its ability to claim favorable evidence.
The problem is not unique to theology. Any agent hypothesis becomes less informative when the agent's powers increase while knowledge of its preferences decreases. But theology represents an extreme case because God is commonly characterized as having maximal causal capacity while divine motives are often treated as only partially accessible.
Divine psychology therefore exposes a fundamental distinction between capability and expectation:
The first is a modal claim about possibility.
The second is a probabilistic claim about preference and selection.
Only the second can do substantial work in an evidential argument.
This is why Scharp's insight has broader significance than any particular debate about fine-tuning, miracles, or revelation. It identifies a hidden explanatory burden in personal theism. Whenever an observed feature of the world is said to be evidence for God because it is something God would plausibly produce, the argument depends on an account of divine psychology. That account must explain not only why the observed outcome is coherent with God's purposes, but why it should have been relatively expected among the enormous number of alternatives available to an omnipotent agent.
Without such a constraint, the appeal to divine agency risks becoming less a predictive explanation than a retrospective rationalization. The observation is known first; a fitting divine purpose is supplied afterward; and the resulting coherence is then mistaken for evidential confirmation. The next section develops this problem more formally by examining ad hoc auxiliary hypotheses, Bayesian priors, and the difference between predicting an outcome and constructing a theory around it after the fact.
Ad Hoc Explanation: From Philosophical Intuition to Bayesian Formalism
The problem of divine psychology becomes more precise once it is connected to the logic of ad hoc explanation. The term “ad hoc” is often used loosely as a synonym for implausible, contrived, or convenient. But its epistemic significance is more specific. An adjustment is objectionably ad hoc when it is introduced mainly to protect or fit a theory in response to some particular observation, while lacking independent support or a broader theoretical rationale. The problem is not simply that the modification was proposed after seeing the evidence. Scientific theories are frequently revised in response to new data. The problem is that a hypothesis can be altered so flexibly that it accommodates the very observation that threatened it, without gaining genuine independent support or predictive discipline.
Suppose a theory $H$ makes a relatively definite prediction, but the observed data $D$ conflict with it. One response is to introduce an auxiliary assumption $A$, producing:
If $A$ is chosen appropriately, the modified hypothesis may fit the data much better:
At first glance, this seems like progress. The modified theory explains what the original theory did not. But the improvement in fit does not by itself establish that $H'$ is better supported. The new auxiliary assumption must itself be justified. The modified hypothesis now requires not only that $H$ be approximately correct, but that $A$ also be correct. More importantly, if $A$ was selected specifically because it reconciles the theory with $D$, then the fit between $H'$ and $D$ is partly a product of the way the model was constructed.
This is closely analogous to overfitting. A sufficiently flexible statistical model can be adjusted to reproduce a set of past observations extremely well, including accidental features and noise. Yet the same flexibility can make the model less impressive as a predictor of new data. The issue is therefore not simply how closely a hypothesis can be made to match what has already happened. It is whether the hypothesis had enough independent structure to make the observed result likely before that result was used to construct the model.
Bayesian reasoning gives this distinction a particularly clear form. Bayes' theorem states:
Here:
- $P(H)$ is the prior probability of the hypothesis;
- $P(D\mid H)$ is the likelihood, or how strongly the hypothesis expected the observed data;
- $P(H\mid D)$ is the posterior probability after updating on the data.
The crucial feature is temporal and logical: Bayesian confirmation concerns not merely whether a hypothesis can be made consistent with the data, but how much probability the hypothesis assigned to those data before they were observed.
Suppose $H$ initially predicts something that makes $D$ very unlikely. After observing $D$, one introduces an auxiliary $A$ such that the expanded theory $H+A$ makes $D$ highly probable. The likelihood may rise dramatically. But that does not mean the evidence strongly confirms the expanded theory, because the prior support for that particular auxiliary matters as well. If there was no antecedent reason to expect $A$, then the hypothesis may have gained retrospective fit only by becoming more complex and flexible.
This becomes clearer when there is not merely one possible auxiliary but a very large space of them:
Before seeing the anomaly, perhaps any of these modifications could have been introduced. After seeing the data, however, one searches the space and selects the particular $A_k$ that produces a good fit.
The fact that some member of a vast hypothesis space can accommodate the observation is not nearly as impressive as having specified the correct member beforehand. If the prior probability had to be distributed across a large number of possible adjustments, then the particular adjustment chosen after the fact may have had very little antecedent probability. This is the Bayesian form of the distinction between prediction and post hoc accommodation.
The same issue appears in Bayesian model comparison through the marginal likelihood or prior predictive distribution. Suppose a model $M$ contains an adjustable parameter $\theta$. Bayesian evidence for the model is not evaluated merely at whichever value of $\theta$ happens to fit the data best. Instead:
This asks, in effect:
Across the possibilities the model regarded as plausible before seeing the data, how much probability did it assign to observations resembling what actually occurred?
This introduces a natural complexity penalty. A highly flexible model spreads its prior probability across many outcomes. A more constrained model concentrates probability on a narrower range.
Compare a theory that predicts:
with one that permits:
Suppose the observation is:
Both theories can “fit” the result. But they do not deserve equal evidential credit. The first theory made a much more specific prediction and therefore took a greater epistemic risk. The second would have been compatible with almost any result in a very broad range. Observing $10$ is therefore substantially more confirmatory of the first model than of the second.
This distinction is central to the argument developed here because many weak explanations appear impressive only when they are evaluated at their best-fitting specification. Once the full space of possibilities available to the hypothesis is taken into account, their predictive success may be far less striking.
A simple example makes the problem vivid. Suppose an original model states:
But the observed value is:
One then modifies the model:
After seeing the result, one selects:
The revised model now fits perfectly. But that perfect fit is not itself impressive if, before observing the data, $A$ could have taken a wide range of values. The model has gained flexibility precisely by allowing the auxiliary to absorb the discrepancy. It has exchanged predictive precision for explanatory accommodation.
An even more serious error occurs if one subsequently acts as though $A=7$ had been expected all along. Suppose the data are observed first and the theorist then assigns a prior distribution sharply concentrated around $A=7$. The theory will appear extraordinarily successful. But the value of $7$ came from the same observation now being used to confirm the model. The evidence has effectively been used twice: once to construct the auxiliary and again to test the theory containing it.
This is why antecedent specification matters. There is an important difference between:
Independent evidence already gave us reason to expect $A\approx7$, and then we observed $Y=17$.
and:
We observed $Y=17$, inferred that $A$ must be approximately $7$, and then claimed that the observation strongly confirmed a theory containing $A\approx7$.
The final mathematical structure may be identical, but the epistemic histories are radically different.
This does not mean that every theoretical modification introduced after an anomaly is illegitimate. That would make scientific development nearly impossible. An initially speculative auxiliary can cease to be objectionably ad hoc if it becomes independently supported. A modification becomes increasingly credible when it:
- explains phenomena beyond the original anomaly;
- generates new and risky predictions;
- can be tested independently;
- connects naturally with established background knowledge;
- successfully predicts observations not used in constructing it.
The ideal pattern is not merely:
It is:
The difference is crucial. A hypothesis that merely absorbs the observation that motivated it has demonstrated compatibility. A hypothesis that subsequently succeeds on independent evidence has begun to demonstrate genuine explanatory structure.
The broad lesson can therefore be stated as follows:
Or, equivalently:
This framework is especially important for agentic explanations. If an agent's motives are left unspecified, then those motives can function like adjustable parameters. Once an observation is known, one may select whatever intention makes the observed action appear reasonable. If a different outcome had occurred, a different intention might have been attributed instead.
The resulting explanation may be psychologically or narratively coherent while possessing little prior predictive concentration. In that sense, the unknown intentions of an unconstrained agent can function much like a high-dimensional parameter space: there may always be some assignment of motives under which the observed result “makes sense.”
This is precisely why the connection between ad hoc reasoning and divine psychology is so consequential. The theological question is not merely whether one can identify some divine motive under which an event is intelligible. The relevant evidential question is whether that motive was independently justified and sufficiently constrained before the event was used to motivate it. The next section formalizes this point by treating divine intentions and actions as latent parameters within the broader theistic hypothesis.
Divine Psychology as a Problem of Latent Parameters
The connection between divine psychology and ad hoc reasoning can now be stated more formally. The central problem is that a bare divine hypothesis does not by itself specify a sufficiently narrow distribution over possible divine actions. To generate a prediction, one must supplement the existence of God with further assumptions about what God values, what God intends, what goals God prioritizes, and how God chooses among the many means available for realizing those goals.
Let:
- $G$ = God exists;
- $M$ = a particular divine motive or preference;
- $A$ = a particular divine action;
- $E$ = the observed evidence.
Once a sufficiently specific motive and action are supplied, it may be easy to make the observed evidence highly probable:
Suppose, for example, that $E$ is a localized religious phenomenon. If one defines $M^*$ as God's desire to authenticate a particular revelation through a localized but ambiguous sign, and $A^*$ as God's choice to produce exactly that kind of sign, then the observation may become highly unsurprising under the fully specified hypothesis.
But this does not tell us how predictive theism itself was. The quantity relevant to a general evidential argument is not merely:
but something closer to:
To obtain that quantity, the uncertainty over possible divine motives and actions must be taken into account. Schematically:
The continuous analogue is:
where $\theta$ represents the relevant features of divine psychology: preferences, priorities, intervention policies, communicative aims, tolerances for ambiguity, and other facts governing divine choice.
This way of framing the issue reveals why selecting the best-fitting divine motive after observing the evidence can be misleading. If the space of possible divine motives is large, then showing that there exists some $M^*$ under which $E$ makes sense does not establish that much prior probability was concentrated on $M^*$. The same point applied earlier to flexible statistical models: a model should not be judged only at whichever parameter value happens to reproduce the observation best. Its evidential success depends on how much of its prior possibility space would have produced something resembling what actually occurred.
Divine psychology can therefore function like a set of latent parameters. The parameters are not directly observed, but they determine what actions the agent would be expected to perform. If those parameters are weakly constrained, then the hypothesis acquires enormous flexibility. After observing the world, one can infer whichever combination of divine preferences makes the world appear intelligible.
Suppose the evidence is $E_1$. One may infer motive $M_1$:
Had the evidence instead been $E_2$, a different motive $M_2$ might have been invoked:
And similarly:
If each possible observation can be paired retrospectively with a suitable divine purpose, then the apparent explanatory success of the resulting story may reflect the flexibility of the motive-space rather than strong predictive content in the original hypothesis.
This is the theological analogue of searching through a large family of auxiliary hypotheses after seeing the data. The important evidential question is not whether there exists some divine psychology that fits the observation, but how much probability the theistic hypothesis placed on that psychology before the observation was known.
This distinction also explains why divine omnipotence creates an unusual explanatory problem. Omnipotence is normally treated as increasing the power of a divine explanation. If an event appears physically extraordinary, an omnipotent agent is not limited by ordinary causal mechanisms. The hypothesis therefore has enormous causal reach.
But explanatory power depends on more than causal reach. It also depends on discrimination among possible outcomes.
For an ordinary human agent, physical and practical limitations eliminate huge numbers of possibilities. A person cannot appear simultaneously in thousands of places, alter astronomical objects by choice, resurrect the dead, instantaneously communicate with every person on Earth, or restructure biological history at will. These limitations constrain what actions can plausibly be attributed to them.
For an omnipotent being, most of those constraints disappear. The set of available actions becomes dramatically larger.
Thus there is a crucial difference between:
and:
Increasing the first does not necessarily increase the second.
Indeed:
This produces what at first seems like a paradox:
Omnipotence increases causal scope while potentially decreasing predictive specificity.
A being capable of doing almost anything is trivially capable of producing the observed result. But precisely because so many alternatives were also available, the observation cannot strongly confirm the hypothesis unless there is an independent reason to think the agent was especially disposed to select this one.
The same issue arises with omniscience. Human agents frequently make imperfect choices because they lack information. If someone selects an inefficient method, the explanation may be that they did not know a better one existed or misunderstood the likely consequences. Such explanations are unavailable for an omniscient being. If a divine action appears poorly suited to its proposed end, ignorance cannot explain the mismatch. Some further purpose must be invoked.
Similarly, ordinary agents face constraints of time, resources, geography, coordination, technology, and risk. These can explain why someone uses a suboptimal means to achieve a goal. An omnipotent deity does not face those limitations in the ordinary sense. The removal of such constraints makes the question of divine selection more rather than less important.
This can be represented schematically. For a limited human agent $H$:
may be sharply constrained because circumstance $C$ rules out many alternatives.
For an omnipotent divine agent $G$, ordinary causal limitations contribute much less:
depends much more heavily on the motive $M$ itself.
The miracle analysis makes this point explicitly. Naturalistic explanations draw on a rich body of physical, psychological, social, and historical constraints, whereas appeals to divine intervention largely depend on whatever theology is supplied to specify why God would act in a particular way.
This means that theology must perform the role played elsewhere by mechanism and background theory. If a religious hypothesis is to make strong predictions, it must specify enough about divine preferences to narrow the action-space. But the more detail that is added, the more assumptions require independent justification.
This creates a structural dilemma.
A minimal theology avoids committing itself to detailed claims about divine motives, but then predicts very little:
A highly specific theology may generate stronger expectations:
But now the epistemic burden has shifted to the additional theological claims:
- Where did they come from?
- How independently well supported are they?
- Would they have been accepted had the observation been different?
- Were they genuinely part of the theory beforehand, or were they inferred from the same evidence they now help explain?
The first transcript captures precisely this requirement when it argues that any theology constraining divine action must be available antecedently; it cannot derive its entire justification from the miracle whose probability it is then being used to assess.
This provides a formal way to understand what makes retrospective divine explanations potentially ad hoc. The problem is not that one has added theology to the proposition that God exists. Some theology is unavoidable if divine action is to be discussed at all. The problem arises when the additional psychological assumptions are chosen because they fit the already-known evidence and are then treated as though they had supplied a genuine prior prediction.
The pattern is:
But the inference is epistemically suspect when $M^*$ was itself obtained from $E$.
A stronger evidential structure would instead be:
The temporal order is not the only issue; what matters is the logical independence of the support for the motive.
This yields another central principle:
The distinction will matter repeatedly in the applications that follow. Miracle claims, resurrection arguments, Marian apparitions, theories of revelation, and intelligent design all introduce an agent with broad causal capacities. In each case, the explanatory success of that move depends not merely on whether the agent could produce the observation, but on whether an independently constrained theory of the agent made that particular observation stand out from the alternatives.
Without such a theory, unknown intentions behave like free parameters. They can be adjusted to produce a close retrospective fit while leaving the underlying hypothesis weakly constrained. The result may be an elegant story about why the agent could have chosen what we observe, without corresponding evidence that the agent would have been likely to choose it. That distinction is the bridge from the formal problem of divine psychology to the concrete evaluation of miracle claims.
Miracles as Agentic Explanations
Miracle claims provide one of the clearest applications of the preceding framework because they make the structure of agentic explanation especially explicit. A miracle, in the sense relevant here, is not merely an unusual or poorly understood event. It is an event explained by a deliberate intervention of God into the ordinary course of affairs. The distinguishing feature is therefore not anomaly by itself, but intentional divine action.
This distinction matters because anomalies are common. An event may be rare, puzzling, difficult to reproduce, or resistant to present explanation without thereby being miraculous. Scientists regularly investigate anomalies without inferring agency, still less divine agency. To say that something is anomalous is only to say that it departs from what is ordinary or expected enough to invite further investigation. The further claim that the anomaly is a miracle adds a very different explanatory structure.
The inferential sequence can be separated into several stages:
- Something unusual occurred.
- Presently available natural explanations are incomplete or contested.
- The event was caused by an agent.
- The agent was supernatural.
- The supernatural agent was God.
- God acted intentionally.
- The action had some specific purpose, such as revelation, authentication, or vindication.
- The event therefore supports a particular theological interpretation.
These propositions are often compressed into the single word miracle, but they are not epistemically equivalent. Evidence that strongly supports the first proposition may provide little support for the later ones. Testimony may establish that witnesses experienced something unusual. It does not automatically establish that God caused the event, nor that God intended it as a message, nor that the message authenticates a particular doctrine.
This is why miracle claims must be treated as agent explanations. Their central structure is:
Once the explanation is framed this way, the same questions that apply to other agent explanations become unavoidable.
- Why would the agent have this motive?
- Why would the agent choose this means?
- Why this time?
- Why this location?
- Why these witnesses?
- Why this degree of publicity?
- Why this degree of ambiguity?
- Why not one of the many alternatives available?
The argument is not that divine agency should be excluded from consideration. It is that if divine agency is admitted as an explanatory hypothesis, then it must be assessed as an explanation rather than merely invoked as a label.
Anomaly is not explanation
The first methodological mistake is to move too quickly from anomaly to agency.
Suppose an event $E$ is unusual and its natural explanation is uncertain. That may warrant saying:
is difficult to estimate for some particular naturalistic hypothesis $H_N$, or perhaps that a currently proposed naturalistic model has a low likelihood.
But from this it does not follow that:
is high, where $H_M$ is a miraculous explanation.
The weakness of one explanation is not itself evidence that a rival explanation succeeds. A miraculous hypothesis has its own explanatory burdens.
This is especially important because natural explanations can be incomplete for perfectly ordinary reasons. Rare events occur. Human testimony is noisy. Meteorological, psychological, historical, and social processes are complex. Background data are often missing. The fact that a naturalistic account leaves unresolved details does not establish that no natural process could have produced them.
The relevant comparison is therefore not:
Has naturalism answered every question?
but:
Which available explanation gives the better account of the evidence when both are subjected to comparable standards?
The transcript's methodological argument is explicit on this point: miraculous and non-miraculous explanations should be assessed with the same broad explanatory criteria rather than treating the miraculous hypothesis as exempt from the demands applied to naturalistic competitors.
The method of exclusion
A common pattern in miracle apologetics can be represented as:
therefore:
Call this the method of exclusion.
The problem is that the inference is underdetermined. Even if every presently proposed natural explanation were shown to be inadequate, it would not follow that a particular divine explanation had been established. There could be:
- unknown natural causes;
- misunderstood psychological processes;
- evidential distortions;
- missing historical information;
- non-divine forms of agency;
- multiple supernatural hypotheses;
- or simply insufficient evidence to determine the cause.
The miracle analysis explicitly criticizes the move from exclusion of natural causes to default acceptance of a miraculous explanation.
The stronger criticism, however, is not merely that another natural explanation may someday appear. That response would make the argument depend too heavily on promissory naturalism. The deeper point is comparative:
Even an incomplete natural explanation may be epistemically preferable to a divine explanation that is still less constrained.
Suppose a naturalistic hypothesis $N$ leaves some details uncertain but is connected to known psychological, physical, or historical processes. Suppose a divine hypothesis $G$ explains the event only by saying that God wanted it to occur for some purpose inferred after the fact.
Then it is possible that:
is imperfectly known but still better grounded than:
because the latter depends on unspecified divine motives.
The absence of a complete natural mechanism therefore does not remove the need to explain why the proposed divine agent would produce this particular outcome.
Methodological symmetry
The principle of symmetry is central.
When a naturalistic explanation is proposed, critics properly ask detailed questions.
If the explanation appeals to psychology:
What psychological mechanism is proposed?
Are similar experiences known?
What role could expectation or social influence play?
If it appeals to meteorology:
Could the relevant atmospheric conditions produce the observed effect?
If it appeals to fabrication:
Did the alleged fabricator possess the necessary materials, skills, opportunity, and motive?
If it appeals to historical development:
Is the proposed sequence compatible with what is known about the period?
These are legitimate explanatory questions.
But the miraculous hypothesis must face analogous scrutiny.
If the hypothesis is:
God did it to vindicate a revelation,
then one should ask:
Why would God want to vindicate this revelation?
Why would God choose this event as the means?
Why this degree of evidential clarity?
Why these recipients?
Why not a more direct intervention?
The problem identified in the miracle discussion is that naturalistic hypotheses are often examined at extremely fine resolution while the divine hypothesis remains at the level of a phrase such as “God acted to vindicate his message.”
That is not explanatory symmetry.
A vague supernatural hypothesis should not gain an advantage from its vagueness. If anything, lack of specification is an explanatory liability.
Means and motive
Ordinary agent explanations often rely on two broad categories:
means and motive.
A person is more plausibly responsible for an event if they:
- had the ability to perform the action;
- had access or opportunity;
- possessed the necessary knowledge;
- had some intelligible reason to do it.
The miracle material uses this general structure to explain how agency is assessed in law and ordinary reasoning.
With God, however, the first category loses much of its discriminating force.
If God is omnipotent, omniscient, and omnipresent, then:
for almost any logically possible event.
God can produce a luminous object.
God can resurrect a body.
God can alter perceptions.
God can produce an artifact.
God can appear to one person, ten people, or everyone.
God can act locally or globally.
God can leave permanent evidence or no permanent evidence.
Causal capacity therefore does very little to narrow the possibilities.
This leaves motive carrying almost the entire burden of explanatory discrimination.
But if motive is poorly constrained, then the divine explanation remains weakly predictive.
This is why the question:
“Could God do it?”
is generally the least interesting part of the miracle hypothesis.
The more important question is:
“Why would God choose to do this?”
The contrastive structure of miracle explanation
Intentional explanation is inherently contrastive.
If someone says:
God caused event $E$,
the relevant question is not simply why $E$ occurred.
It is:
For example:
Why a localized apparition rather than a universal one?
Why a temporary luminous phenomenon rather than a permanent astronomical sign?
Why appearances to existing followers rather than hostile or neutral observers?
Why an artifact appearing centuries later rather than contemporaneous public evidence?
Why ambiguous testimony rather than uniform testimony?
Why geographically restricted revelation rather than simultaneous communication?
The point is not that every alternative is obviously better. The point is that an agentic explanation must provide some account of the selection.
If the only answer is:
God had reasons,
then the explanation has not yet specified why the actual event should have been expected.
If the answer is:
God had reason $R$,
then the next question is:
Why believe God had $R$?
And if $R$ was inferred because it makes the observed event fit, the ad hoc problem returns.
Means–end fit
There is another important dimension of agentic explanation: does the selected action actually fit the attributed purpose?
Suppose the proposed motive is:
Then the action $A$ should be evaluated relative to that objective.
If the action is:
- highly localized;
- ephemeral;
- difficult to verify;
- accessible only through disputed testimony;
- open to multiple interpretations;
- invisible to later observers;
then one may ask whether $A$ is well suited to $M$.
This does not require identifying the single best possible action available to God. It only requires the ordinary explanatory judgment that actions should bear some intelligible relation to the motives invoked to explain them.
The miracle discussion illustrates this with the analogy of robbers who allegedly break into a bank because they want money but take only ten dollars despite having access to much more. The point is not that taking ten dollars is logically incompatible with greed. It is that the action fits the attributed motive poorly enough that one would reasonably suspect some additional motive or alternative explanation.
The same logic applies to divine agency.
If God is said to want:
- universal revelation;
- authentication;
- salvation;
- missionary propagation;
- clear communication;
then an extremely limited or ambiguous intervention requires explanation.
The more capable the agent, the less easily poor means–end fit can be blamed on practical limitation.
The asymmetry of explanatory burden
Miracle debates often create a peculiar asymmetry.
The naturalistic hypothesis is required to explain:
- every witness discrepancy;
- every physical detail;
- every historical uncertainty;
- every psychological mechanism;
- every apparent anomaly.
The divine hypothesis is required only to remain logically possible.
But that reverses the proper explanatory burden.
If a naturalistic model is incomplete, that counts against it to some degree.
If a divine model is unspecific, that should count against it as well.
One should not compare:
against:
The relevant comparison is between comparably articulated hypotheses.
If the miracle hypothesis is:
God intentionally produced $E$ in order to achieve $M$,
then the defender of the hypothesis must explain both:
and:
The first concerns divine psychology.
The second concerns the relation between motive and selected action.
Both are necessary.
Rare natural events and unconstrained divine events
There is also an important asymmetry between rarity and lack of constraint.
A critic may object:
The naturalistic explanation is extremely improbable.
But rare events occur within well-characterized processes.
A one-in-a-million natural event may still arise from a mechanism whose probability structure is understood.
By contrast, if the divine hypothesis provides no principled distribution over possible divine actions, then saying that the event is “not unlikely under God” may have little content.
The comparison is not necessarily:
versus:
It may instead be:
versus:
The latter does not automatically win.
Miracles and the general explanatory problem
Miracle reasoning therefore reproduces the general pattern identified earlier.
First, an anomaly is observed:
Then a divine intention is proposed:
Then a divine action is posited:
The resulting hypothesis may make the evidence highly intelligible:
But if $M^*$ and $A^*$ were selected because they fit $E$, the real question remains:
That quantity cannot be inferred simply from the fact that one retrospectively constructed divine story fits the observation.
This gives the general methodological rule for miracle claims:
A miracle explanation should be evaluated not by whether God could have caused the event, but by whether an independently justified account of divine motives and action made something like the event antecedently expected.
The distinction can be summarized as:
and:
and finally:
These principles do not show that miracles are impossible. They show that miracle claims inherit the same explanatory burdens as every other intentional explanation, while divine omnipotence removes many of the ordinary constraints that make agent explanations informative. The burden therefore shifts almost entirely to divine psychology: what God wanted, why God wanted it, and why God selected this means rather than another.
The next section applies this framework to three concrete cases—the resurrection appearances, the Shroud of Turin, and the reported miracle at Fatima—to show how the same explanatory pattern recurs across very different kinds of religious evidence.
Three Case Studies: Resurrection, the Shroud of Turin, and Fatima
The abstract problem of divine psychology becomes clearer when applied to concrete cases. The resurrection appearances, the Shroud of Turin, and the reported miracle at Fatima differ substantially in their historical form, evidential basis, and theological significance, yet the same explanatory structure recurs in each. In every case, there is some body of data that proponents regard as anomalous; an intentional divine explanation is then introduced; and the explanatory force of that hypothesis depends upon claims about what God wanted to accomplish and why God would have selected the particular means observed.
The point of the following analysis is not to settle every historical dispute surrounding these cases. Indeed, the argument can grant a substantial portion of the data for the sake of examining what follows from it. The central question is what additional explanatory work is required to move from:
to:
That move cannot be made by anomaly alone. It requires a theory of divine motive and divine selection.
The resurrection: why these appearances, to these people, in this way?
Consider the resurrection in deliberately generous terms. Suppose, for the sake of argument, that after Jesus' death a number of followers sincerely reported experiences they took to be appearances of the risen Jesus, that these experiences played an important role in the subsequent Christian movement, and that the resulting historical data are sufficiently unusual to require explanation.
None of those concessions yet entails resurrection.
A specifically divine explanation might be formulated as:
God bodily resurrected Jesus and caused him to appear to followers in order to vindicate Jesus' divine mission, authenticate his message, and initiate or strengthen the subsequent Christian proclamation.
The structure is:
Once formulated in this way, however, the explanatory questions multiply.
First, why should we antecedently attribute this motive to God?
The hypothesis does not merely say that God is able to resurrect a person. Given omnipotence, that is comparatively easy. It says that God wanted to vindicate this person, this message, by this particular intervention, at this point in history. Those are substantive theological propositions.
Second, even granting the motive of vindication, why this pattern of appearances?
The miracle discussion emphasizes the restricted social character of the resurrection reports. The appearances are associated primarily with people already connected to Jesus and the nascent Christian community rather than being distributed indiscriminately among the wider population.
This matters because ordinary psychological and sociological processes are naturally constrained in precisely this way. If experiences, memories, reports, interpretations, expectations, and narratives arise through human social processes, one would expect them to cluster among people who:
- knew the relevant individual;
- cared about his fate;
- shared expectations and beliefs;
- communicated with one another;
- occupied overlapping social networks;
- lived within the relevant geographical and cultural setting.
A naturalistic hypothesis therefore has independent reasons to expect localization.
The relevant processes are not omnipresent. Human communication spreads through social networks. Memory depends upon acquaintance. Grief is concentrated among those connected to the deceased. Interpretations are shaped by shared cultural expectations. Testimony spreads geographically and historically through ordinary channels.
Thus, under a broad class of naturalistic hypotheses:
may be substantially higher than:
The naturalistic hypothesis is constrained.
But God, as traditionally conceived, is not subject to those limitations.
If the purpose is divine vindication, God could have produced appearances to:
- existing followers;
- hostile opponents;
- neutral observers;
- political authorities;
- distant populations;
- contemporary historians;
- large public gatherings;
- every relevant individual;
- or some combination of these.
The divine action-space is enormously larger.
Consequently, the mere fact that God could have chosen appearances primarily to followers tells us very little. The relevant question is:
Why should the antecedent divine psychology have concentrated probability on that particular pattern?
The theological answer may be:
God wished to strengthen the followers who would carry the message forward.
But this immediately introduces another auxiliary motive:
Why privilege that motive?
Perhaps the response is that revelation normally operates through human witnesses rather than universal manifestation. But then another theological principle has been introduced:
Why expect that principle?
Perhaps direct universal revelation would somehow undermine faith. Then another auxiliary appears:
At every stage a coherent story may be available. The question is whether these motives were independently part of the theory or are being generated because the actual evidence requires them.
That distinction is crucial. If one begins with the resurrection reports and then constructs a theology according to which God prefers socially mediated, historically localized, non-universal revelation, one cannot subsequently treat the very localization that generated that theology as powerful confirmation of it.
The structure risks becoming:
That is precisely the form of retrospective parameter fitting considered earlier.
The resurrection case also raises the problem of means–end fit. If the proposed divine motive is public vindication of Jesus, the observed means appear oddly limited. The miracle discussion notes the tension between an allegedly sweeping divine objective and appearances concentrated among those already favorably situated to accept the claim.
The argument need not assert that God would necessarily choose the most spectacular imaginable demonstration. The point is more modest. In ordinary agent explanations, if an action seems poorly suited to the motive invoked to explain it, that weakens confidence in the proposed motive unless some additional constraint is independently known.
If I claim that a person desperately wants the entire city to know some fact but deliberately communicates it only to a small group of close associates, the action is not logically incompatible with the stated goal. But the mismatch demands explanation.
With an omnipotent agent the pressure is even greater because ordinary excuses involving lack of reach, resources, opportunity, or technological capacity are unavailable.
The missionary variant of the resurrection argument encounters the same difficulty. If God's purpose was to initiate a missionary movement, one can ask why this particular historical method was chosen, why bodily resurrection was necessary for that purpose, why revelation was concentrated in this region and period, and why the preceding span of human history did not receive an analogous intervention. The original analysis emphasizes precisely this proliferating need for further theological explanation once the initial motive is scrutinized.
The resurrection therefore illustrates the central distinction:
while:
remains poorly constrained.
The specific motive makes the data fit. The difficulty is establishing the motive independently.
The Shroud of Turin: an ambiguous artifact and the problem of divine method
The Shroud provides a different type of case because the relevant evidence is material rather than primarily testimonial. Again, the argument can proceed without resolving every dispute over the object's provenance or formation. Suppose, for the sake of analysis, that the image on the Shroud contains unusual features and that its precise production process is disputed.
A naturalistic explanation might propose deliberate human production, whether as fraud, religious art, devotional object, or some other artifact later interpreted more strongly than originally intended. Such hypotheses can be criticized in ordinary explanatory terms:
- Did the maker possess the required technology?
- What materials were available?
- What motive existed?
- Could the proposed technique produce the relevant features?
- Is the historical setting compatible with the hypothesis?
The transcript explicitly notes this structure: natural explanations concerning fabrication or icon production can be investigated through questions about human motive and available means.
Now compare the divine hypothesis:
God miraculously produced the image on the Shroud in order to leave evidence of Jesus' resurrection.
The explanatory structure is:
Again, causal possibility is trivial relative to the larger question. An omnipotent God could presumably produce such an image.
But why would God choose this artifact as a means of authentication?
If the goal is to provide evidence of the resurrection, then several features become relevant:
- the object's disputed historical trajectory;
- its limited accessibility across much of history;
- the technical complexity involved in evaluating it;
- disagreement concerning its authenticity;
- the ambiguity of its evidential significance even if some aspects of its formation remain unexplained.
The problem is contrastive.
- Why an artifact whose interpretation depends upon later historical preservation and highly technical investigation?
- Why not contemporaneous evidence available to broader populations?
- Why a physical trace that permits continuing dispute over whether it is humanly produced?
- Why should the evidence appear historically late or obscure relative to the event it is allegedly intended to authenticate?
The miracle analysis makes this point directly: even if the divine motive is specified as vindicating the resurrection, the historical timing, restricted reach, and ambiguity of the Shroud sit uneasily with the alleged capabilities and goal of the agent.
One can of course generate theological responses.
Perhaps God values subtle evidence.
Perhaps God wants evidence strong enough to invite belief but weak enough to preserve freedom.
Perhaps God wanted later generations to possess a physical sign.
Perhaps the gradual historical emergence of the Shroud serves some providential purpose.
Every one of these propositions is logically possible.
But again, logical possibility is cheap.
Each explanation adds another component to the divine psychology:
The relevant question is whether those components have antecedent support.
Suppose the Shroud had instead been unambiguously authenticated from the first century and continuously preserved in a transparent chain of custody. A theologian could easily say:
Of course God would preserve clear physical evidence of the resurrection.
Suppose instead, as in the actual evidential situation under discussion, its origin is controversial and its interpretation disputed. The theologian can say:
Of course God would avoid making the evidence coercively obvious.
Both outcomes can be accommodated by changing the attributed divine purpose.
That flexibility is exactly the problem.
A hypothesis that can explain both:
and:
by switching between auxiliary motives acquires very little confirmation from either outcome unless there was antecedent reason to prefer one auxiliary over the other.
The Shroud therefore illustrates a particularly important form of divine-psychology reasoning:
An unexplained mechanism does not automatically establish an intentional purpose.
Even if one conceded that the image's physical formation was not presently explained, it would remain a separate question whether:
- an agent produced it;
- the agent was God;
- God intended resurrection authentication;
- the particular evidential ambiguity of the artifact fits that intention.
Each step requires argument.
The unexplained residue cannot do all of this work by itself.
Fatima: from anomalous spectacle to specific Marian revelation
Fatima provides perhaps the clearest case because the gap between event and interpretation is especially large.
Again, suppose one grants a generous core of the reports. Suppose a substantial crowd witnessed an unusual luminous or solar-like phenomenon associated with prior predictions by children claiming Marian apparitions. Suppose further that the event was sufficiently striking that ordinary dismissal as “nothing happened” is inadequate.
Even these concessions do not yet give us:
God authenticated a revelation from the Virgin Mary.
The observed evidence and the theological interpretation remain distinct.
A simplified divine hypothesis might be:
God caused an extraordinary luminous phenomenon in order to authenticate the children's Marian message.
Formally:
The first divine-psychology question is obvious:
Why should God want to authenticate a Marian apparition?
That already presupposes a substantial theology involving:
- the theological status of Mary;
- the possibility and purpose of Marian apparitions;
- God's willingness to communicate through Mary;
- the importance of the particular message;
- the propriety of miraculous authentication.
None of this follows merely from:
A Jewish, Muslim, Protestant, Catholic, generic theist, or deist framework may assign very different prior probabilities to the proposition that God would authenticate a Marian apparition. The interpretation therefore depends upon theology antecedent to the event.
But even granting Catholic Marian theology, the questions continue.
- Why children?
- Why these particular children?
- Why this location?
- Why this historical moment?
- Why an outdoor luminous phenomenon?
- Why an event whose descriptions vary?
- Why something apparently associated with the sun or sky while not producing the universally observable astronomical consequences that literal solar movement would imply?
- Why an event visible under particular local circumstances rather than a global phenomenon?
- Why no permanent physical trace sufficient to settle the matter for later investigators?
These questions do not prove that the event was non-divine. They expose the amount of additional theory required before the divine explanation becomes predictive.
The original miracle discussion emphasizes precisely these contrastive questions: why Marian apparitions, why children, why this type of authentication, why such a localized and photographically elusive event, and why a solar-like phenomenon rather than an objectively global astronomical occurrence?
This case is especially revealing because the observed event appears subject to exactly the kinds of restrictions expected under ordinary conditions.
Whatever happened occurred:
- at a particular place;
- at a particular time;
- among people gathered for a shared religious expectation;
- under specific environmental circumstances;
- without an independently confirmed global astronomical event.
Those are precisely the sorts of constraints imposed by geography, meteorology, perception, social expectation, and communication.
The divine hypothesis, by contrast, is not naturally limited in those ways.
If God wants to authenticate Mary's message, God could in principle produce:
- identical experiences in every observer;
- a permanent physical sign;
- a globally observable astronomical event;
- simultaneous revelation in multiple continents;
- independently verifiable information inaccessible to the children;
- repeated controlled manifestations;
- direct communication to every person for whom the message is relevant.
The point is not that God must choose any of these alternatives.
The point is that the observed choice occupies one small region of an enormous divine action-space.
We therefore need a reason to think God preferred that region.
At this point the theological auxiliaries typically begin to proliferate.
Perhaps:
God wanted enough evidence to reward faith without eliminating doubt.
Then introduce:
Perhaps:
God favors humble recipients rather than powerful authorities.
Then:
Perhaps:
God ordinarily works through historically situated revelation.
Then:
Perhaps:
The ambiguity itself serves a spiritual purpose.
Then:
Each proposition may be independently defensible. But if they are introduced only because they reconcile Fatima's actual evidential character with divine authorship, then the explanation becomes increasingly fitted to the observation.
The Bayesian structure is revealing.
Suppose the actual evidence is $E_F$.
After observing $E_F$, construct a specific divine psychology $\theta_F$ such that:
is high.
It does not follow that:
was high beforehand.
The latter depends upon how much prior probability should have been assigned to $\theta_F$ relative to all the other possible divine preferences and intervention policies:
Fatima therefore closely resembles the data-dependent auxiliary problem.
The event supplies:
- the locality;
- the type of phenomenon;
- the degree of ambiguity;
- the selection of recipients;
- the temporal context.
A divine motive is then constructed around precisely those properties. Once constructed, the motive is used to say:
This is exactly the sort of event God would produce for these purposes.
But unless the theology independently selected those features beforehand, the inference risks treating retrospective fit as predictive success.
The significance problem
Fatima also illustrates another level of underdetermination absent from simpler causal questions.
Even if one granted that something supernatural occurred, the intended theological significance would remain open.
The inference:
contains additional premises.
Perhaps the cause was supernatural but non-divine.
Perhaps an agent produced the event but its purpose was misunderstood.
Perhaps the event was divine but intended for some purpose other than validating every statement attributed to the apparition.
Perhaps the witnesses' interpretation of the event exceeded what the event itself established.
This demonstrates why evidence for an unusual phenomenon must be distinguished from evidence for a specific intentional interpretation.
The more detailed the theological conclusion becomes, the more background theory is required to connect the observation to it.
What the three cases reveal together
The resurrection, the Shroud, and Fatima differ in many respects, but their explanatory structure is remarkably similar.
In each case there is first some evidential target:
Then a divine purpose is supplied:
Then a particular divine action is interpreted as the means:
The resulting explanation takes the form:
Once $M$ and $A$ are specified to fit the case, the result may appear highly coherent.
But the central evidential question remains:
What independently justified $M$ and $A$?
The resurrection raises the problem of social localization:
Why revelation largely through people already connected to Jesus?
The Shroud raises the problem of historical and evidential ambiguity:
Why authenticate a central religious event through a disputed artifact of limited historical reach?
Fatima raises the problem of localized and equivocal manifestation:
Why authenticate a supposedly important revelation through a temporary phenomenon embedded in a specific social, geographical, and perceptual context?
In all three cases, defenders can supply possible answers.
That fact is not disputed.
The problem is that the ease with which answers can be generated is itself epistemically ambiguous. It may reflect a well-confirmed theory of divine purposes. But it may instead reflect the extraordinary flexibility of attributing motives to an agent whose preferences are weakly constrained.
This is why the recurring question should not be:
Can I imagine a reason God might do this?
It should be:
What independently justified theory of divine psychology made this course of action substantially more expected than the alternatives before the evidence was known?
That question converts miracle discussion from a contest over imaginative possibility into a genuine problem of comparative explanation.
And once stated that way, the three cases exhibit the same general tension:
The power of the objection therefore does not depend upon proving that the resurrection did not occur, that the Shroud has a particular natural origin, or that nothing unusual happened at Fatima. It remains even after substantial evidential concessions. The argument targets the inferential bridge from anomaly to intentional divine explanation.
A fact can be strange without being a miracle.
An event can remain unexplained without thereby becoming divine.
A phenomenon can even be granted as supernatural without fixing its theological meaning.
The central explanatory burden begins precisely where the descriptive evidence ends: with the claim that a particular divine agent chose this particular intervention for this particular purpose. That is where divine psychology enters, and that is where retrospective accommodation must be distinguished from antecedently constrained explanation.
Revelation, Hiddenness, and the Free-Will Auxiliary
The same explanatory problem becomes especially sharp in theories of revelation and divine hiddenness. Here the relevant evidence is not merely that some unusual event occurred, but that purported divine communication is unevenly distributed, historically mediated, geographically concentrated, interpretively disputed, and often ambiguous enough to sustain deep disagreement among sincere inquirers.
If God is said to desire relationship with human beings, communicate morally or salvifically important truths, or make some revelation available, then these features of the evidential situation create an obvious question of means–end fit.
Suppose the theological hypothesis includes:
One might initially expect this motive to favor some degree of divine self-disclosure. Yet the actual evidential situation contains:
- disagreement between religious traditions;
- disagreement within traditions;
- historical dependence on particular texts and witnesses;
- large geographical differences in religious exposure;
- contested miracle claims;
- ambiguous experiences;
- uncertainty regarding which purported revelations are genuine;
- sincere nonbelief and nonresistant doubt.
The explanatory problem is therefore contrastive:
If God wants persons to know enough to enter freely into relationship with God, why is divine communication structured in this particular way?
One common theological response is that greater clarity would interfere with human freedom. God must, on this view, maintain sufficient epistemic distance that belief and relationship remain voluntary.
This can be formulated as an auxiliary principle:
The important issue is not whether this idea is conceivable. It plainly is. The issue is whether it is independently justified and whether the proposed relation between evidential ambiguity and freedom is actually persuasive.
Evidence and coercion are not the same thing
The first problem is conceptual. Strong evidence that a person exists does not ordinarily coerce one's attitudes toward that person.
Knowing that another person exists does not force one to:
- trust them;
- admire them;
- obey them;
- love them;
- forgive them;
- enter into relationship with them.
Knowledge of existence and freedom of response are distinct.
The same distinction applies to God. Even if a person were certain that God existed, substantial questions of response would remain:
Is God worthy of worship?
Should I obey?
Do I trust God's purposes?
Do I want relationship?
How should I respond morally?
Which purported revelation, if any, correctly represents God's will?
Thus:
And therefore:
The inference from freedom to ambiguity requires much more argument than it often receives.
The miracle/revelation discussion develops this point through an analogy with informed consent. If an individual is to make an autonomous decision about an important matter, we normally regard relevant, intelligible, and reliable information as enhancing autonomy rather than undermining it. A person's choice is not more meaningful merely because important information has been withheld.
This analogy does not establish that divine-human relationships are identical to ordinary cases of informed consent. Its force is narrower: it undermines the intuitive assumption that less information naturally produces more freedom.
In many ordinary contexts the opposite is true.
If one wants someone to make a meaningful decision, one normally tries to ensure that they understand:
- what the options are;
- what is at stake;
- who is communicating;
- whether the information is credible.
Accordingly, if God's purpose is to elicit a free and informed response, one might initially expect evidence sufficient to identify the relevant relationship and its terms.
That expectation is at least as intuitive as the contrary claim that ambiguity is required for freedom.
Means–end fit in revelation
The issue can be formalized.
Suppose:
One possible means is:
Another is:
The question is:
To argue that the actual structure of revelation supports theism, one needs some reason to think the second course is especially fitting.
But that reason cannot simply be:
Because revelation is in fact ambiguous, God must value ambiguity.
That would reproduce the same retrospective pattern already identified.
The actual evidential structure would first generate the divine motive, and the motive would then be used to explain the evidential structure.
The reasoning would be:
Without independent support for $M_{\text{epistemic distance}}$, this is structurally similar to fitting an auxiliary parameter after observing a discrepancy.
The original discussion explicitly argues that, if God wishes people freely to respond to revelation, one might instead expect people capable of such response to have access to the relevant information in an intelligible and evidentially credible form. It further argues that stronger evidence would not itself eliminate the freedom to accept, reject, love, obey, or resist.
The escalation of auxiliaries
The problem becomes more serious when one auxiliary is insufficient.
Suppose someone objects:
Why does God not reveal himself more clearly?
A reply might be:
Because God values free will.
But if clear knowledge does not obviously eliminate freedom, another explanation may be added:
God wants persons to develop morally through uncertainty.
Now introduce:
and:
If someone asks why religious evidence is also distributed so unevenly across history and geography, another principle may appear:
If someone asks why sincere people interpret the available evidence so differently:
If someone asks why some apparently sincere seekers remain unconvinced:
Any one of these propositions may have a theological defense. The epistemic concern concerns the cumulative structure.
As more auxiliaries are introduced, the hypothesis becomes increasingly capable of accommodating the observed pattern:
But every auxiliary also requires support.
The expanded hypothesis may fit the data better while increasing uncertainty about which theological additions are actually correct.
This is precisely the phenomenon described earlier:
If the auxiliaries are mainly selected because they neutralize particular objections, they behave like ad hoc repairs.
Defensive and offensive divine psychology
The hiddenness problem also exposes a broader asymmetry in how divine psychology can be used.
Consider two evidential contexts.
Favorable observation
Suppose people report powerful religious experiences.
The theist may argue:
This makes sense because a loving God would seek relationship and reveal himself.
Here divine psychology is used offensively:
Unfavorable observation
Now consider sincere nonbelief, religious ambiguity, or the apparent absence of clear revelation.
The response may become:
God may have reasons we cannot understand for remaining hidden.
Here divine psychology becomes defensive:
But these two strategies are difficult to combine without cost.
If we know enough about God's psychology to infer:
then the absence or uneven distribution of such experience should have evidential consequences.
If, alternatively, God's reasons are sufficiently inscrutable that both revelation and non-revelation are easily accommodated, then the original favorable observation becomes less confirmatory.
The problem can be expressed schematically.
Suppose:
and:
If the hypothesis can accommodate $E$ by saying:
God wants to reveal himself,
and accommodate $\neg E$ by saying:
God wants to preserve distance or has unknown reasons,
then:
has not strongly discriminated between the two observations.
This does not show that either theological explanation is false. It shows that explanatory flexibility should not be mistaken for confirmatory power.
The unfalsifiability concern, stated carefully
This point is sometimes expressed too crudely by saying that theological explanations are “unfalsifiable.” That formulation can obscure the real issue. A hypothesis need not be straightforwardly falsifiable in a Popperian sense to possess evidential content. Probabilistic theories routinely survive observations they regarded as unlikely.
The more precise concern is one of likelihood discrimination.
A theory is evidentially informative insofar as some observations satisfy:
If instead a theory can be supplemented after the fact so that:
is high for nearly every possible $E_i$, then observing one particular $E_i$ does little to confirm the unsupplemented theory.
The issue is therefore not:
Can theology invent a possible explanation for hiddenness?
It clearly can.
The question is:
Did the theological framework independently lead us to expect this pattern of hiddenness rather than substantially different patterns of revelation?
That is the evidential question.
Revelation as an intentional communication problem
The problem becomes even clearer if revelation is treated as communication.
Ordinary communication involves at least:
- a communicator;
- an intended recipient;
- a message;
- a medium;
- a purpose;
- some desired degree of successful transmission.
If communication fails systematically, we ask whether:
- the medium was inadequate;
- the sender lacked knowledge;
- the sender lacked ability;
- the message was deliberately obscure;
- the recipient was incapable of understanding;
- or communication was not actually the sender's objective.
These possibilities constrain ordinary explanations.
Divine revelation removes many ordinary limitations. An omniscient God would know:
- what each person understands;
- what evidence each person possesses;
- what cultural background each person inhabits;
- which ambiguities will predictably arise;
- which messages will be mistranslated or disputed;
- which forms of evidence will persuade without coercing.
An omnipotent God would also possess extraordinarily broad communicative options.
Consequently, if revelation remains historically contingent, geographically uneven, and deeply disputed, the usual explanation cannot be inability.
The explanation must once again appeal to choice.
God chose this communicative structure.
The relevant question therefore becomes:
Why?
That is divine psychology again.
And because the number of available communicative strategies is vast, the theory must explain why God prefers:
over alternatives:
The larger the available action-space, the more important the decision theory becomes.
Hiddenness and the global skeptical pattern
The revelation case therefore reinforces the broader pattern.
Theological reasoning often begins with broad attributes:
From these, relatively specific predictions are drawn when the evidence is favorable.
When the evidential pattern becomes difficult, additional motives are introduced:
The more such principles are available, the easier it becomes to reconcile theism with any observed degree of divine visibility.
But this produces the same Bayesian cost discussed earlier. A theory that spreads probability across many outcomes gains less confirmation from any particular one.
Thus:
while simultaneously:
This is why hiddenness is not merely another objection to one particular religious doctrine. It reveals the methodological tension at the center of divine psychology. A theologian cannot freely alternate between strong claims about what God would probably do and appeals to unknown divine purposes without affecting the evidential force of the original claims.
The broader lesson is therefore the same as in miracle reasoning: conceivability of a divine purpose is not evidence for the probability of that purpose. If free will, soul-making, epistemic distance, or providential hiddenness are to explain the actual structure of revelation, they must be supported by more than their ability to reconcile the theory with the data that generated the problem in the first place.
Intelligent Design: Adding an Agent Without Adding a Theory
Intelligent Design provides one of the clearest non-miraculous applications of the preceding argument. The details differ from claims about resurrection, revelation, or Marian apparitions, but the underlying explanatory structure is strikingly similar. In each case, some already-existing feature of the world is identified as difficult, improbable, highly organized, or otherwise in need of explanation; an intelligent agent is introduced as an additional explanatory factor; and the agent's presumed intention is then inferred from the very feature that is being explained.
The crucial question is therefore not whether an intelligent agent could produce biological complexity. Of course an intelligence with sufficient capacities could produce many biological structures. The question is whether positing such an intelligence adds enough independent structure to make the observed biological facts more probable than they were under competing hypotheses.
The distinction developed throughout this essay applies immediately:
This is especially important because talk of “design” can create the impression of explanation simply by introducing intentional vocabulary. A structure is complex or functionally organized; therefore it is said to have been designed for some purpose. But the attribution of purpose does not automatically explain why the designer would have selected this particular structure rather than one of the enormous number of alternatives available.
The explanatory burden is again one of agent psychology and decision theory.
Design explanations can be legitimate
The objection should first be distinguished from the much stronger and less plausible claim that explanations involving intelligent agents are inherently unscientific or explanatorily defective. They are not.
Agency can provide excellent explanations when the relevant agent is sufficiently constrained by independent background knowledge.
Archaeology offers an obvious example. If an archaeologist encounters a shaped stone object, the hypothesis of human manufacture may explain its properties well. But the explanatory hypothesis is not merely:
Some intelligence produced this object.
It contains a large body of implicit background information about humans:
- humans have bodies with particular physical capacities;
- humans manipulate materials;
- humans use tools;
- humans construct objects for recognizable purposes;
- humans live in particular historical and geographical settings;
- different technologies are available at different periods;
- particular materials permit some manufacturing techniques but not others;
- human behavior is constrained by effort, time, resources, skill, and social circumstances.
Consequently, the human-design hypothesis rules things out.
An archaeologist might regard one set of marks as likely products of deliberate tool use and another as more consistent with erosion because there is a prior theory of what human beings can do, how they tend to do it, and what traces those activities leave.
The same is true in forensic reasoning. If investigators infer that a fire was deliberately set, the hypothesis involves more than the mere logical possibility that a human could ignite something. Investigators ask about:
- motive;
- opportunity;
- means;
- materials;
- behavioral patterns;
- likely points of origin;
- attempts at concealment;
- known human incentives.
Human agency provides explanatory content precisely because it is not unconstrained.
Likewise, an engineer can infer that a machine component was intentionally manufactured because there is extensive background knowledge about engineering objectives, production techniques, tolerances, available materials, and the relation between structure and function.
Thus the general principle is not:
It is:
This qualification matters enormously for evaluating Intelligent Design.
The unspecified designer problem
Let:
Suppose we observe some biological feature $E$, perhaps a highly organized molecular system, a functionally integrated anatomical structure, or some other complex biological arrangement.
The weakest design inference says:
An intelligent designer could have produced $E$.
That establishes only something like:
But, as argued earlier, causal possibility is not enough. If $E$ is offered as evidence for $D$, then we require some reason to think:
is sufficiently high, and ideally higher than under relevant competing hypotheses:
where $N$ represents some naturalistic evolutionary hypothesis.
The problem is that $D$ by itself does not tell us very much about what the designer would produce.
To make the likelihood informative, we need additional propositions about the designer's psychology or objectives. For example:
- Does the designer value complexity?
- Does the designer value simplicity?
- Does it value efficiency?
- Does it value redundancy?
- Does it favor robustness over elegance?
- Does it prefer modularity?
- Does it reuse existing structures?
- Does it value maximal biological diversity?
- Does it primarily value conscious organisms?
- Does it care about suffering?
- Does it optimize reproductive success?
- Does it permit apparently inefficient developmental pathways?
- Does it intervene continuously or only occasionally?
- Does it work through evolutionary mechanisms or independently of them?
- Is it constrained by materials or prior structures?
- Does it deliberately conceal evidence of intervention?
- Does it prefer designs that resemble the products of ordinary descent?
These are not incidental questions. They determine what observations should be expected under the design hypothesis.
If they are left unanswered, then “an intelligent designer exists” may possess enormous causal scope while placing very few restrictions on biological outcomes.
The structure is therefore directly analogous to divine psychology:
where:
- $D$ = designer exists;
- $M$ = designer's motives or objectives;
- $S$ = designer's strategy or constraints;
- $A$ = design intervention;
- $E$ = observed biological structure.
What does the bare designer hypothesis tell us about:
If very little, then it also tells us very little about:
The hidden parameter problem reappears
The same latent-parameter framework developed for divine psychology can therefore be transferred almost directly to Intelligent Design.
Let $\theta$ represent the unknown properties of the designer:
- preferences;
- design goals;
- intervention policies;
- constraints;
- tolerances;
- strategies.
Then:
After observing the biological feature, it may be easy to find a particular value $\theta^*$ such that:
is very high.
For example, suppose the biological system exhibits substantial redundancy. One might propose:
The designer values robustness, so redundancy is exactly what we would expect.
If instead the system were extremely economical and non-redundant, one might say:
The designer values efficiency and elegant simplicity.
If a structure appears to reuse older components:
The designer prefers modular reuse.
If it introduces radically novel architecture:
The designer is innovative and not constrained by previous designs.
If an organism appears beautifully optimized:
Intelligence predicts optimization.
If another structure appears inefficient:
The designer may have been balancing multiple objectives.
If a biological feature appears poorly engineered:
The apparent defect may serve an unknown function.
If no such function is discovered:
The designer's goals may simply differ from ours.
Each of these stories may be possible.
But collectively they illustrate the problem.
The hypothesis is not gaining explanatory precision. The designer's psychology is being adjusted to whichever biological outcome is observed.
The form is:
and so on.
The ability to fit each outcome by choosing a different motive is not the same thing as predicting any particular outcome.
Retrospective designer psychology
This is where the problem of ad hoc reasoning becomes especially important.
Suppose one examines an already-known biological structure $E$ and concludes:
This looks as though it was designed to accomplish function $F$.
One then attributes to the designer the intention:
The structure now appears intelligible:
But notice the direction of inference.
The feature $E$ was observed first.
The function or goal was inferred from $E$.
The designer's motive was then defined in terms of that inferred function.
Finally, the close match between the structure and the inferred motive is presented as evidence of design.
Schematically:
This can generate an illusion of strong explanatory success.
Once we specify that the designer wanted precisely the function implemented by the structure, of course the structure seems comparatively unsurprising.
But the relevant quantity remains:
not merely:
And if $M_F$ was inferred largely from $E$ itself, it cannot simply be treated as an independently given prior feature of the designer hypothesis.
This is exactly the data-dependent-prior problem encountered earlier.
The Dialogue's analysis of ad hoc reasoning emphasizes that a hypothesis may fit observations extremely well when a parameter is chosen after those observations are known, while receiving far less evidential credit once the full range of possible parameter values is considered.
Intelligent Design confronts the same issue if the designer's preferences are effectively free parameters.
The selection-space problem in biological design
The problem becomes clearer if we imagine the number of possible biological systems an unconstrained designer could create.
For some biological function $F$, there may be an enormous set of possible implementations:
A sufficiently powerful designer could perhaps create many of them.
If the design hypothesis does not independently tell us which one the designer prefers, then observing $E_k$ does not strongly favor design merely because $E_k$ performs a function.
One needs a probability distribution:
Without it, the inference can become:
We observed $E_k$.
$E_k$ performs function $F$.
An intelligent designer could have wanted $F$.
Therefore $E_k$ is what we would expect from a designer.
But this does not establish that the designer would have chosen $E_k$ rather than the potentially enormous set of alternatives that also realize $F$.
The deeper question is therefore contrastive:
Why this biological architecture rather than another?
That question becomes increasingly important as the proposed designer becomes more powerful.
A human engineer may be forced to select a particular solution because of:
- material constraints;
- manufacturing costs;
- inherited infrastructure;
- limited computational ability;
- imperfect information;
- time pressure;
- available tools.
Those constraints can explain why the resulting artifact has the form it does.
But an unspecified cosmic designer may have no known equivalent limitations.
The less constrained the agent, the less the actual design follows simply from the existence of the agent.
This gives the same paradox identified earlier:
Indeed, if the designer can implement nearly unlimited solutions while its preferences remain unknown:
Why evolutionary theory is explanatorily different
This comparison becomes particularly important when Intelligent Design is contrasted with evolutionary theory.
A weak characterization of evolutionary explanation would be:
Nature produced the organism.
That would indeed be explanatorily vacuous.
“Nature did it” is no better in principle than “God did it” if nothing is said about mechanisms or constraints. The miracle material explicitly recognizes this symmetry: merely invoking “nature” without specifying a causal account would not constitute an adequate explanation either.
But evolutionary theory is not merely the proposition that nature somehow produced life.
It supplies a structured set of mechanisms and historical constraints, including:
- heredity;
- variation;
- mutation;
- selection;
- drift;
- developmental constraints;
- population structure;
- descent from prior forms;
- environmental pressures;
- historical contingency.
The important philosophical point is not that evolutionary theory provides a complete prediction of every organism or every biological structure.
It plainly does not.
The relevant point is that it restricts the space of expected outcomes.
Evolution cannot simply redesign an organism from scratch at every generation.
New structures arise from inherited developmental systems.
Changes must occur within historical lineages.
Existing anatomy constrains future anatomy.
Mutational processes operate through particular biological mechanisms.
Population dynamics affect which variants spread.
Environmental circumstances alter selective pressures.
Evolutionary history therefore matters.
These constraints mean that the probability of a particular biological outcome depends upon antecedent states:
where the terms may represent mutation, selection, drift, developmental constraint, and other relevant processes.
The theory is therefore historical and path-dependent.
That path dependence can explain why biological structures sometimes appear less like clean-sheet engineering and more like modifications of inherited systems.
The explanatory contrast is important.
A generic designer can, in principle, be assigned nearly any preference after the fact.
Evolutionary mechanisms cannot so easily be adjusted without consequences elsewhere in the theory.
The mechanisms carry commitments.
Those commitments generate explanatory risk.
Predictive constraint rather than perfection
This distinction should not be exaggerated into the claim that evolution predicts biological outcomes with anything approaching exact precision.
That is unnecessary for the argument.
The relevant standard is comparative constraint, not determinism.
A theory can possess substantial explanatory power while allowing many possible outcomes, provided it assigns different expectations to different classes of outcomes.
Thus, the contrast is not:
versus:
It is:
versus:
This also explains why identifying some limitation or unresolved problem in a particular evolutionary account does not automatically establish design.
If an evolutionary explanation is incomplete, then its evidential standing may decrease.
But that does not entail:
The design hypothesis must still establish its own likelihood.
This is the same methodological symmetry demanded in miracle reasoning.
Intelligent Design and the method of exclusion
A design argument can become particularly weak when it takes the form:
therefore:
This reproduces the method of exclusion.
Even if one demonstrates that a particular evolutionary pathway is insufficiently understood, several possibilities remain:
- the mechanism is incomplete;
- relevant historical information is missing;
- another natural process contributed;
- the reconstruction of the pathway is mistaken;
- the phenomenon requires a modification of existing biological theory;
- or the evidence remains genuinely unresolved.
None of these possibilities proves design false.
But their existence shows that the negative case against one explanatory proposal does not automatically become a positive case for another.
The inferential structure:
is valid only if $N$ and $D$ exhaust the possibility space.
They generally do not.
More importantly, even if one were comparing only two broad models, the evidence for $D$ would still depend upon:
And this is exactly what remains difficult to estimate without a theory of the designer.
The same problem occurred in miracle arguments:
Likewise:
unless the design hypothesis independently predicts the feature in question.
The designer as a residual explanatory variable
There is an even subtler problem when Intelligent Design accepts large portions of ordinary evolutionary history while introducing intelligent intervention only at selected points.
The structure then becomes something like:
Whenever evolutionary mechanisms appear sufficient, the designer need not be invoked.
Whenever some structure seems difficult to explain, design is introduced.
This risks turning the designer into a kind of residual explanatory variable:
Such a hypothesis can appear extraordinarily successful because its domain is defined by explanatory difficulty itself.
But this creates an asymmetry.
Ordinary biological mechanisms must specify:
- what they do;
- when they operate;
- under what conditions;
- with what constraints.
The designer is permitted to enter precisely where those mechanisms become uncertain.
If later research explains the structure naturally, the scope of design can simply retreat.
If another anomaly appears, design can be relocated there.
This creates a potentially mobile explanatory boundary:
But a moving residual does not necessarily constitute an independent causal theory.
To function as one, the design hypothesis would need to specify in advance:
Where should intervention occur?
What characteristics should designed structures possess?
What patterns should distinguish design from ordinary evolution?
What outcomes would count against intervention?
What should we discover next if design occurred?
Without such constraints, the designer risks becoming the biological counterpart of an appeal to miracle whenever ordinary causal explanation is incomplete.
Borrowing the explanatory structure of evolution
This gives special significance to the way Intelligent Design can make use of facts already organized by evolutionary biology.
Suppose a biological system has:
- a historical lineage;
- homologous structures;
- developmental dependencies;
- ecological functions;
- interactions with other inherited systems.
Evolutionary theory provides much of the causal and historical framework explaining how these properties relate.
Design may then be introduced to account for some additional feature regarded as insufficiently explained.
But notice what can happen.
Much of the explanation's genuine structure is supplied by the evolutionary framework:
- why the organism is located in this lineage;
- why it shares traits with related organisms;
- why its development proceeds through certain inherited pathways;
- why its structures depend upon earlier biological forms.
The designer is then added at a point of difficulty.
The resulting explanation can seem richer than the design hypothesis actually is because it inherits the surrounding explanatory structure from the theory it supplements.
The important question becomes:
What new predictions arise specifically from the designer component?
If removing $D$ leaves all the independently predictive structure in place while $D$ serves only to explain a presently difficult residual, then the marginal explanatory contribution of $D$ may be small.
This can be represented schematically:
while:
is invoked because of anomaly $E_k$.
For $D$ to provide additional evidence, it must contribute more than:
after $E_k$ has already been identified.
It should independently generate further expectations.
Efficiency, imperfection, and the flexibility of design reasoning
Biological imperfection reveals the problem especially clearly.
Suppose a biological structure appears impressively efficient.
This may be offered as evidence of intelligence:
Intelligent agents produce efficient solutions; therefore efficiency supports design.
But now suppose another structure appears inefficient, fragile, redundant, or historically awkward.
Several responses are available:
The designer had multiple objectives.
What appears inefficient may serve an unknown function.
The designer intentionally used evolutionary processes.
Historical constraints were part of the chosen method.
Robustness required redundancy.
Our standards of good design may not match the designer's.
Again, none of these replies is logically impossible.
The epistemic question is what happens to the theory's predictive content.
If both:
and:
can be equally reconciled with design by changing the attributed goals, then efficiency itself cannot provide strong confirmation without independent reason to expect the relevant design preference.
The same is true of simplicity and complexity.
Observe simplicity:
Elegant design.
Observe complexity:
Sophisticated design.
Observe reuse:
Efficient modular design.
Observe novelty:
Creative design.
Observe robustness:
The designer values reliability.
Observe fragility:
The system may serve other purposes.
The designer hypothesis becomes explanatorily permissive because intentional vocabulary can be fitted to almost any biological outcome.
This is the direct analogue of the theological pattern:
while:
In both cases, opposite observations can be absorbed through different agent psychologies.
The asymmetry between criticizing evolution and specifying design
This produces an important methodological asymmetry in some design arguments.
Evolutionary explanations may be subjected to extremely detailed questions:
What mutation occurred?
What was the ancestral structure?
What selective advantage existed?
What intermediate stages were viable?
What developmental mechanisms were involved?
What population processes permitted fixation?
These are legitimate questions.
But explanatory symmetry requires equally detailed questions of the design hypothesis:
What exactly did the designer alter?
At what historical point?
Why then rather than earlier?
Why this structure?
Why this organism?
Why this developmental pathway?
Why this amount of intervention?
Why leave surrounding structures untouched?
What objective was being optimized?
What constraints governed the design?
What observation would have been surprising if the designer were responsible?
If the answer is:
We cannot know the designer's motives,
then the hypothesis has acquired an important defense against criticism.
But it has paid for that defense by surrendering predictive specificity.
One cannot coherently insist:
The designer's goals are too inscrutable for apparent imperfections to count against design,
while also arguing:
Functional organization is exactly what we should expect an intelligent designer to produce.
The first claim broadens:
across many possible outcomes.
The second requires it to be concentrated around the actual one.
The tension is structurally identical to the divine-psychology problem.
The designer's inscrutability dilemma
The dilemma can be stated directly.
Horn 1: the designer's psychology is knowable
If we possess an independently justified theory of the designer's goals, then design can potentially generate real predictions.
Suppose, for example, independent evidence established that the designer:
- maximizes energetic efficiency;
- never uses redundant structures;
- always minimizes developmental complexity.
Then biological observations could test those claims.
Some findings would support design.
Others would count against it.
The hypothesis would take epistemic risks.
Horn 2: the designer's psychology is inscrutable
If instead we know almost nothing about the designer's goals, then apparent anomalies can always be defended:
The designer may have had some unknown reason.
But now the designer hypothesis makes fewer predictions.
The observation:
cannot strongly confirm $D$ simply because an unknown intelligence could have wanted $E$.
Thus:
This is not a rhetorical disadvantage imposed on design from outside. It follows from the basic logic of confirmation.
The Bayesian comparison
The problem can be summarized with the same formalism used earlier.
Let:
- $D$ = designer exists;
- $\theta$ = designer psychology and strategy;
- $E$ = observed biological feature.
Then:
The design advocate may identify some $\theta^*$ after seeing $E$ such that:
But this does not establish:
If the prior distribution:
ranges across enormously many possible designer preferences, then the design model has spread its predictive probability over an enormous possibility space.
This is precisely what Bayesian model comparison penalizes.
A theory receives less evidential credit for accommodating a datum when the model could also have accommodated many radically different data.
The Dialogue's central principle applies without alteration:
whereas:
The problem of Intelligent Design is therefore not necessarily that “design” has zero explanatory meaning. It is that an unspecified designer can be too flexible to supply the likelihoods required by the argument.
What a genuinely explanatory design theory would need
The criticism becomes stronger, rather than weaker, if one states what would overcome it.
A genuinely predictive design theory would need to tell us substantially more than:
Some intelligence was involved.
It would need an independently motivated account of such questions as:
- Who or what is the designer?
- What capacities does it possess?
- What limitations does it face?
- What objectives does it pursue?
- How does it rank competing objectives?
- When does it intervene?
- When does it allow ordinary evolutionary processes to operate?
- What kinds of structures does it prefer?
- What empirical signatures should intervention leave?
- What observations would be surprising under the hypothesis?
From this, the theory would need to derive novel expectations not selected from the same biological structures originally offered as evidence.
For instance:
followed by independent observation:
The relevant evidential pattern would be:
That would be fundamentally different from:
This parallels the distinction made earlier between a legitimate theoretical development and an objectionably ad hoc auxiliary. A modification introduced after an anomaly can become well supported if it generates new predictions and acquires independent confirmation.
The criticism of Intelligent Design is therefore methodological rather than definitional.
It is not:
Intelligence can never be an explanation.
It is:
An intelligence whose psychology, constraints, and intervention policy are unspecified does not automatically explain whatever structure it is capable of producing.
Intelligent Design as a general illustration of the unconstrained explainer
The deeper importance of Intelligent Design for this essay is that it shows the problem of divine psychology is not confined to explicitly religious miracle claims.
The generalized structure is:
This creates a potential circularity of explanatory construction.
A more epistemically robust structure would be:
That is the difference between an agent functioning as a theory and an agent functioning as an explanatory wildcard.
This point can be compressed into several principles:
and:
Intelligent Design therefore exemplifies the same generalized explanatory pathology seen in miracle reasoning. Existing data are selected because they seem anomalous or difficult. An agent is introduced because the agent possesses the causal ability to produce the feature. A motive is inferred from the feature itself. The resulting agentic story makes the feature retrospectively intelligible. That intelligibility is then treated as though it were the antecedent predictive success of the theory.
But the essential question remains unanswered:
Before observing this biological structure, what independently justified theory of the designer made this structure substantially more expected than the alternatives?
Without an answer, the designer adds metaphysical possibility without necessarily adding explanatory constraint.
The comparison with evolution sharpens the issue. Evolutionary theory is valuable not because the word natural possesses explanatory power, but because the theory specifies mechanisms, histories, and constraints that distribute probability unevenly across possible biological outcomes. An intelligent agent could in principle provide equally powerful or even superior explanation—but only if comparable constraints on that agent were independently supplied.
The generalized lesson is therefore not a preference for impersonal causes over personal ones. It is a preference for constrained explanations over unconstrained ones. Human designers can explain artifacts because much is known about humans. Evolutionary mechanisms can explain biological patterns because much is specified about how those mechanisms operate. A divine or cosmic designer becomes explanatory in the same sense only when enough is known about its preferences and decision procedures to answer the contrastive question that every intentional explanation eventually confronts:
Until that question receives an independently constrained answer, “intelligent design” risks naming a possible source of the phenomenon without supplying the theory necessary to explain why that source would have produced the world we actually observe.
The Generalized Pattern: The Unconstrained Agent as Explanatory Wildcard
The preceding cases suggest that divine psychology is not an isolated difficulty peculiar to theology. It is an instance of a more general epistemic pattern. The same structure appears whenever an agent is invoked to explain an already-known observation while the agent's preferences, constraints, and decision procedures remain sufficiently unspecified that a suitable motive can be supplied after the fact.
The generalized problem can be stated simply. First, some observation $E$ is encountered. An agent $A$ is then proposed as its cause. Because the agent's psychology is poorly constrained, an intention $I_E$ is inferred that would make production of $E$ intelligible:
The resulting story may be perfectly coherent. But the direction of epistemic dependence is important. The evidence is doing much of the work of specifying the hypothesis that is subsequently invoked to explain the evidence.
A stronger explanatory structure would run in the opposite direction:
In the first case, the intention is selected because the observation requires it. In the second, the intention is independently constrained and generates expectations before the observation is used as confirmation.
This distinction unifies the earlier discussion of divine psychology, miracle claims, revelation, hiddenness, and Intelligent Design. The central issue in each case is not agency itself. It is the difference between an agent functioning as a genuinely predictive theory and an agent functioning as an explanatory wildcard.
Causal possibility is cheap
The first principle of the generalized pattern is that establishing causal possibility is relatively weak evidence.
For any hypothesis $H$, showing that:
does not establish that $E$ strongly confirms $H$.
The point is obvious in ordinary reasoning. Suppose a suspect in a criminal investigation was physically capable of committing the crime. That fact alone provides little explanatory force if thousands of other people were equally capable. What matters is whether other independently established facts narrow the possibilities:
- motive;
- access;
- location;
- opportunity;
- prior behavior;
- possession of relevant means.
Likewise, an intelligent designer's ability to produce a biological structure does not establish that the structure was probable under design. God's ability to produce a miracle does not establish that this miracle was likely under theism. A supernatural agent's ability to cause an apparition does not establish that the apparition should be interpreted as a specific revelation.
The explanatory question is always contrastive:
Why this outcome rather than the alternatives?
This is why the original miracle analysis places so much emphasis on constraint. A hypothesis gains explanatory force by narrowing the range of expected outcomes; a hypothesis capable of accommodating almost anything gains comparatively little from the fact that the actual outcome falls somewhere within that enormous range.
Thus:
Agent explanations require something like a decision theory
A causal mechanism can explain an event by specifying how antecedent conditions produce subsequent outcomes. An agent explanation requires something additional: a theory, however informal, of choice.
For an agent $A$, circumstances $C$, and possible actions $a_1,\ldots,a_n$, what matters is something like:
That distribution depends upon facts about:
- desires;
- values;
- beliefs;
- priorities;
- risk tolerance;
- capabilities;
- constraints;
- information;
- preferred means.
This is why ordinary human-agent explanations are possible despite imperfect predictability. Human beings are not deterministic machines, but neither are they blank causal capacities. Extensive background knowledge allows us to make differential judgments about what a person is more or less likely to do.
An unconstrained agent lacks that structure.
If nothing is known about the agent's ranking of possible outcomes, then the assertion:
An intelligence chose $E$
does little more than relocate the mystery.
The question:
Why did $E$ occur?
has become:
Why did the agent choose $E$?
Unless the second question is better constrained than the first, the explanatory gain may be minimal.
This yields a general principle:
Scharp's divine psychology is therefore one special case of a broader need for agent decision theory.
Power and explanation can move in opposite directions
An especially important implication follows when the proposed agent is extremely powerful.
Intuitively, greater power may seem to improve an agent explanation because it removes doubts about causal capability. But removing constraints also increases the number of actions available to the agent.
Let the set of possible actions available to an agent be:
For an ordinary human agent, $n$ may be limited by physical capacities, resources, geography, knowledge, and time.
For an extremely powerful or omnipotent agent, the relevant action-space becomes vastly larger.
If the agent's preferences are independently well specified, that may not matter. A clear motive can still concentrate probability on a relatively narrow subset of actions.
But if preferences remain unknown, then increasing the action-space increases underdetermination.
Schematically:
This produces one of the central reversals of the argument:
The removal of causal constraints can weaken rather than strengthen explanatory discrimination.
An omnipotent agent solves the problem of whether the event could happen while worsening the problem of why this event should have happened.
This is why:
Retrospective specificity can conceal antecedent flexibility
Ad hoc explanations often possess a misleading feature: they look extremely specific once they have been constructed.
A theologian may eventually provide a detailed account:
God chose these recipients, at this time, through this sign, because revelation had to remain sufficiently indirect to preserve freedom while still authenticating a particular message.
That is a specific story.
Likewise, a design theorist may say:
The designer reused this molecular system because modular reuse promotes robustness while minimizing developmental disruption.
Again, highly specific.
But specificity after the evidence has been examined is not the same thing as predictive specificity beforehand.
The relevant question is how many alternative stories could have been generated if the evidence had looked different.
Suppose a general agent hypothesis allows a large family of possible motives:
After observing $E$, one selects:
because:
is high.
If $I_k$ had very little independent prior support, the resulting specificity is partly an artifact of selection.
The ad-hoc reasoning discussed earlier captures precisely this phenomenon: finding one assumption from a large space that fits an anomaly is much less evidentially impressive than specifying that assumption before the anomaly was known.
Thus:
A theory can look highly detailed while remaining deeply flexible at the level that matters for confirmation.
The explanation can become parasitic on the explanandum
This produces a particularly important form of circularity.
Ordinarily, we expect explanatory structure to exist at least partly independently of what it explains.
But in the unconstrained-agent pattern, the observation itself may determine:
- the proposed intention;
- the proposed preference hierarchy;
- the proposed intervention strategy;
- the interpretation of the observed event.
The resulting structure can look like:
The explanandum has helped construct the explanans.
This does not make the explanation formally circular in the simple deductive sense. It is better described as epistemically dependent or accommodative. The problem is that the evidence is being asked simultaneously to supply the parameters of the hypothesis and to confirm the hypothesis once those parameters have been fitted.
This is why data-dependent priors are so useful as an analogy. Once a parameter value has been obtained from the observation, one cannot legitimately evaluate the resulting model as though that value had been fixed independently in advance. The Dialogue emphasizes that doing so effectively counts the same evidence twice.
The generalized lesson is:
An observation should not receive the same confirmatory credit when it was substantially used to construct the version of the hypothesis under which it becomes probable.
Explanatory flexibility creates an asymmetry between defense and confirmation
The unconstrained-agent pattern becomes especially visible when a theory alternates between two modes.
Confirmatory mode
When evidence appears favorable:
This is what an intelligent agent would plausibly do.
The hypothesis claims explanatory specificity.
Defensive mode
When evidence appears unfavorable:
The agent may have had reasons we do not understand.
The hypothesis claims psychological flexibility.
These modes cannot be combined without cost.
Suppose the observed outcome is $E$.
A defender says:
because an agent of this type would likely produce $E$.
Now suppose instead that $\neg E$ occurs.
The defender says:
because the agent could have unknown reasons for producing something else.
But the more outcomes treated as unsurprising under the hypothesis, the less probability can be concentrated on any particular one.
In probabilistic terms, if:
then expanding the range of outcomes assigned substantial probability necessarily limits how strongly the distribution can privilege a particular $E_k$.
Thus:
This is the generalized form of the tension between divine psychology and skeptical appeals to inscrutable purposes.
“Unknown reasons” are not neutral
This point deserves emphasis because appeals to unknown motives often appear epistemically harmless.
Suppose a critic argues that an observation is surprising under hypothesis $H$.
The response:
There may be some unknown reason why $H$ would produce this observation,
establishes only possibility.
It does not establish a probability.
Ignorance about the distribution is not the same thing as a distribution favorable to the hypothesis.
One cannot infer:
merely because one does not know whether $E$ or $\neg E$ is more likely.
Nor can one infer that $E$ is unsurprising merely because some unknown auxiliary might explain it.
This is another form of the earlier warning:
The space of imaginable explanations can be enormous. The epistemic issue is how probability should be distributed across them.
Ad hoc reasoning is not simply “changing your mind”
The generalized argument should not be misunderstood as a prohibition against theoretical revision.
A theory can encounter unexpected evidence, develop a new auxiliary assumption, and become substantially better as a result.
The critical difference is what happens afterward.
The Dialogue identifies several features that reduce ad-hocness:
- explanation of additional phenomena;
- independent testing;
- integration with established knowledge;
- novel risky predictions;
- successful prediction of data not used to generate the adjustment.
Thus, if a theological theory developed a new account of divine action after encountering an anomaly and that account subsequently generated successful independent predictions, its original post-hoc origin would become much less troubling.
Likewise, if a design hypothesis specified a particular intervention policy and then successfully predicted previously unknown biological structures, the designer component would gain independent evidential support.
The problem is not:
It is:
without subsequently acquiring independent constraint.
This gives a useful distinction:
The former is essential to inquiry.
The latter can preserve fit at the cost of explanatory discipline.
The prediction–accommodation distinction
The generalized structure can now be represented very clearly.
Predictive success
A theory independently specifies:
From $H$, one derives a relatively narrow expectation:
Then one observes:
The evidence confirms the theory because the theory had taken a risk.
Accommodation
One observes:
Then one adjusts the hypothesis:
such that:
The second procedure may produce excellent fit.
But its evidential force depends heavily upon what independent reason exists for $A_E$.
This is the difference between saying:
“My theory predicted the Queen of Hearts before the card was drawn”
and:
“My system contains a page explaining the Queen of Hearts now that I know which card appeared.”
A hypothesis that contains enough possible pages will always contain the right one.
The important question is how much probability it placed on that page before the outcome was known. The Dialogue's card analogy captures precisely this distinction between genuine prediction and a system capable of selecting the appropriate explanation only afterward.
A hierarchy of explanatory strength
The generalized framework suggests a useful hierarchy.
Level 1: Mere compatibility
The hypothesis allows the observation.
Level 2: Retrospective intelligibility
After observing $E$, some coherent auxiliary $A$ can be added such that:
is reasonably high.
Level 3: Independent explanatory constraint
There is independent reason for $A$, and therefore $E$ is genuinely more expected under the expanded theory.
Level 4: Comparative explanatory advantage
The hypothesis not only makes $E$ probable, but does so more strongly than relevant rivals:
Level 5: Novel predictive success
The theory generates a risky prediction about evidence not used in constructing it, and the prediction succeeds.
Confusion frequently arises because an explanation at Level 1 or 2 is spoken of as though it had reached Level 4 or 5.
This is particularly easy in agentic explanations because intentions provide immediate narrative coherence.
Once one hears:
God wanted to authenticate the message,
or:
the designer wanted to optimize the function,
the outcome may feel explained.
But the important evidential work has scarcely begun.
The generalized explanatory failure
The pathology can therefore be summarized as follows.
-
An observation is identified as anomalous, complex, significant, or otherwise in need of explanation.
-
An agent with broad causal capacities is introduced.
-
The agent's intentions are not independently specified in sufficient detail to generate strong expectations.
-
A motive is inferred from the observed outcome itself.
-
The resulting motive makes the outcome retrospectively intelligible.
-
That retrospective intelligibility is treated as evidence that the agent hypothesis predicted the observation.
-
When contrary observations arise, alternative motives or unknown purposes are introduced.
-
The resulting flexibility protects the theory from disconfirmation while reducing its predictive specificity.
This is the generalized pattern underlying the cases examined throughout the essay.
It can be represented compactly:
The problem occurs in the transition from retrospective fit to confirmation.
The deeper methodological principle
The generalized argument is therefore not fundamentally anti-theological, anti-design, or even anti-agentic.
Its governing principle is methodological:
An explanation gains evidential force by imposing independently justified constraints on what should be observed.
This principle applies equally to naturalistic hypotheses.
“Nature did it” is not an explanation.
“Evolution did it” is not an explanation if nothing is specified about heredity, mutation, selection, development, history, or population dynamics.
“Psychology did it” is not an explanation without some account of the relevant psychological process.
“Society caused it” is similarly empty unless social mechanisms are specified.
The standard is symmetrical.
What distinguishes stronger explanations is not whether they invoke personal or impersonal causes, natural or supernatural entities. It is whether they contain enough independently supported structure to discriminate among possible outcomes.
This can be stated in its most general form:
The relation should not be taken as a literal numerical equation, but as a methodological principle: as a theory becomes less constrained and more capable of absorbing arbitrary outcomes, the mere fact that it can accommodate the actual observation becomes less evidentially impressive.
The unconstrained explainer
This is the sense in which the cases examined in this essay instantiate the figure of the unconstrained explainer.
The unconstrained explainer is an entity or process introduced with enough causal power to account for the observation but without enough independently specified structure to explain why that observation rather than another should have resulted.
In theology:
God could have a reason.
In miracle apologetics:
God could have chosen this sign.
In revelation:
God could value precisely this degree of ambiguity.
In Intelligent Design:
the designer could have wanted precisely this structure.
Each claim may be true.
But explanatory possibility is not yet explanatory evidence.
The central demand remains:
What did the hypothesis make probable before the observed result was used to specify the hypothesis?
That question is the common thread running through divine psychology, ad hoc reasoning, miracle explanation, revelation, and design.
It also reveals why the distinction between explanation and accommodation matters so deeply. A flexible system can generate a coherent account of nearly any outcome. Indeed, the more flexible it becomes, the easier retrospective explanation becomes. But the very flexibility that guarantees a story after the fact can deprive the theory of the ability to tell us what should have happened beforehand.
Thus the generalized tension can finally be expressed as:
while, absent additional independent constraints,
The epistemic virtue of an explanation therefore lies not in its capacity to tell a story about whatever world happens to exist, but in its ability to distinguish that world from worlds that, according to the theory, should have looked different. That is the generalized methodological lesson of divine psychology.
Objections, Qualifications, and What Would Count as a Successful Divine Explanation
The preceding argument is intentionally demanding, but it should not be stronger than the epistemic principles that motivate it. Several objections naturally arise. Some challenge the requirement that agent explanations be probabilistically constrained; others contend that theological background knowledge already supplies the relevant constraints; still others argue that the standards being imposed would undermine ordinary scientific reasoning as well. These objections are important because the central claim is not that divine explanations are impossible, nor that every theological auxiliary is illegitimate. It is that appeals to divine agency acquire evidential force only to the extent that they satisfy the same broad requirements that distinguish explanation from retrospective accommodation elsewhere.
The strongest version of the argument is therefore conditional:
If an observation is offered as evidence for intentional divine action, then the evidential force of that observation depends upon an independently warranted account of what the divine agent would be comparatively likely to do.
The objections below help clarify what this requirement does and does not entail.
Objection: agents are not laws and need not generate precise predictions
A natural response is that the argument appears to demand too much from intentional explanation. Human beings are not deterministic systems. We routinely explain actions without possessing numerical probability distributions over everything an agent might do. If someone enters a restaurant because they are hungry, we do not calculate:
before accepting hunger as part of the explanation.
Why, then, demand a precise “divine decision theory” before allowing God's intentions to explain anything?
The objection is correct about one thing: exact numerical probabilities are unnecessary. Nothing in the argument requires assigning a number such as:
Explanatory reasoning frequently proceeds through qualitative comparisons. We may be justified in judging:
without knowing either probability numerically.
The relevant requirement is therefore not exact prediction but differential expectation.
Ordinary agent explanations work because we can discriminate among possibilities. Hunger does not uniquely predict entering a particular restaurant, but it raises the probability of some actions relative to others. Other facts further constrain the prediction:
- the person is near the restaurant;
- they have money;
- they know food is available;
- they have eaten there before;
- they do not strongly dislike the cuisine.
A richer background model yields a more informative intentional explanation.
The same applies in forensic reasoning. Knowing that a suspect desired money does not uniquely predict bank robbery. Yet motive becomes explanatory when combined with opportunity, means, prior behavior, circumstance, and other background information.
Thus the relevant contrast is not:
It is:
The argument against unconstrained divine explanation does not depend on expecting God to behave mechanically. It requires only that an evidential claim such as:
“This is more likely if God exists”
have some principled basis for the word more.
Without that, “God might do this” remains a possibility claim.
The methodological requirement can therefore be stated modestly:
Objection: God's goodness already constrains divine psychology
A more substantive theological objection is that divine psychology is not unconstrained. Classical theism attributes significant properties to God:
- perfect goodness;
- perfect rationality;
- omniscience;
- love;
- justice;
- wisdom.
These properties may appear to provide exactly the antecedent theory that the argument claims is missing.
For example:
A good God would value conscious beings.
A loving God would desire relationship.
A rational God would create an ordered universe.
A just God would care about moral agency.
If these claims are independently entailed by theism, then perhaps the divine action-space is already significantly narrowed.
This is probably the strongest response to the divine-psychology objection, and it should be granted as far as it goes. General divine attributes can constrain divine action. A conception of God as perfectly good presumably excludes some motives. A perfectly rational God would not act from confusion. Omniscience eliminates decisions based on factual ignorance.
But the important question is how much these general attributes constrain particular choices.
Consider:
Suppose one wants to infer:
The required bridge is:
But why?
Perhaps embodied conscious life is valuable.
Yet presumably many other goods are possible:
Even if creating embodied agents is good, it does not follow that a perfectly good God would probably instantiate that particular good, especially if God could instantiate innumerable alternatives or combinations of goods.
The same problem intensifies with specific miracle claims.
From:
it is a long inferential distance to:
or:
or:
Each transition requires increasingly specific theological premises.
The argument therefore need not deny that goodness constrains divine psychology. It asks proponents to identify and justify the intermediate premises.
For example:
The evidential burden increases as the conclusion becomes more specific.
The objection therefore succeeds against an excessively strong claim that divine psychology is wholly unknowable. It does not solve the central problem:
How does one move from abstract divine attributes to sufficiently discriminating expectations about concrete divine choices?
The bridge cannot be assumed merely because both ends of the inference are theologically intelligible.
Objection: theology itself supplies the needed background knowledge
A related response is that the argument artificially isolates the bare proposition “God exists.” Actual religious believers rarely reason from bare theism. Christianity, Judaism, Islam, and other traditions contain detailed theological claims about:
- God's purposes;
- human nature;
- revelation;
- providence;
- salvation;
- covenant;
- divine communication;
- historical action.
Thus a Christian need not ask merely:
The relevant hypothesis might instead be:
Then perhaps:
is genuinely high.
This objection is important, and in one sense it is entirely correct. A richer hypothesis can make stronger predictions than bare theism.
Indeed, this is exactly what the argument predicts:
The problem is not that such enrichment is illegitimate.
The problem is where the enriching theology receives its evidential support.
If the Christian theological framework is independently justified by evidence not substantially dependent upon the resurrection evidence under consideration, then conditioning on that theology may be legitimate.
But suppose much of the justification for the theology comes from:
- the resurrection;
- Christian scripture interpreted through the resurrection;
- doctrines whose credibility depends upon the resurrection;
- traditions whose divine authority is partly inferred from the resurrection.
Then using that same theology to assign a high likelihood to the resurrection risks circularity.
The structure becomes:
The issue is not deductive circularity but evidential dependence.
The same point applies to Fatima.
Suppose one says:
Catholic theology gives independent reason to think God may authenticate Marian apparitions.
That claim has force only insofar as the relevant Catholic theology is warranted independently of the particular Marian evidence being evaluated.
Otherwise, the tradition teaches that such events are theologically fitting partly because the tradition contains those events, and the events are then used to confirm the tradition because they fit its theology.
This does not show that no theological background knowledge can ever be used. It establishes a bookkeeping requirement:
Evidence used to construct or justify the divine psychology cannot simply be counted again as though the psychology independently predicted that same evidence.
A well-developed theology may therefore answer the divine-psychology objection, but only to the degree that its relevant motivational claims possess independent warrant.
Objection: prediction is not the only form of explanation
Another important objection is that the argument places excessive weight on prediction. Many legitimate explanations are developed only after the event being explained.
Historical explanations are obvious examples. Historians explain revolutions, wars, political decisions, and institutional developments after they occur. Evolutionary biologists explain features of organisms that were not predicted before discovery. Geologists reconstruct past events from present traces. None of these explanations becomes illegitimate merely because the explanandum was known before the explanation was formulated.
This is correct.
The contrast between prediction and accommodation must therefore not be interpreted as:
and:
A post hoc explanation can be excellent.
The deeper distinction is between:
an explanation developed after the fact using independently constrained principles
and:
an explanation made successful principally by tailoring unconstrained auxiliaries to the fact itself.
Consider a historical explanation proposed after an event. It may rely upon independently established facts concerning:
- institutional structures;
- economic incentives;
- military capacities;
- known personal motives;
- documents;
- geographical constraints;
- causal mechanisms.
The explanation was formulated later, but its components were not freely invented merely to fit the outcome.
Similarly, evolutionary explanations of previously unknown features can gain credibility by integrating those features into an independently established framework of inheritance, phylogeny, mutation, selection, development, and population dynamics.
Moreover, a post hoc hypothesis can strengthen itself through independent tests.
The Dialogue captures this important qualification: an initially ad hoc-looking adjustment becomes progressively more credible when it explains additional phenomena, is independently measurable, generates new predictions, and successfully predicts observations not used in its construction.
Thus the argument is not:
It is:
The difference is fundamental.
Objection: every theory contains auxiliary assumptions
A related objection comes from philosophy of science. Scientific hypotheses are never tested in complete isolation. Predictions depend upon auxiliary assumptions concerning measurement, initial conditions, background theories, instrument reliability, and other factors. When predictions fail, scientists frequently revise these auxiliaries rather than abandon the central theory.
If theology is criticized whenever it supplements theism with additional assumptions, does the same argument not undermine science?
No. The point is not that auxiliary assumptions are suspect merely because they are auxiliary.
The distinction concerns how constrained they are.
Consider two cases.
Case 1: independently motivated auxiliary
A physical theory predicts $D$, provided some background condition $A$ obtains. Independent evidence supports $A$.
Then:
There is nothing objectionably ad hoc about the structure.
Case 2: freely adjustable rescue assumption
The prediction fails. One invents $A_D$ specifically so that:
after $D$ is already known.
The new auxiliary has:
- no independent evidence;
- no additional explanatory consequences;
- no novel predictions;
- no independent means of testing it.
The two cases are epistemically very different.
Scientific theories frequently survive anomalies because the revised auxiliary hypotheses subsequently earn independent support. That is precisely why the mere presence of auxiliaries cannot be the objection.
The problem with weak divine-psychology explanations is that the relevant auxiliaries—God wanted ambiguity, God preferred these witnesses, God chose this historical means, the designer values this particular type of architecture—can sometimes be difficult to test independently of the very phenomena for which they were introduced.
The correct standard is therefore symmetrical:
The “if and only if” should not be interpreted too rigidly, but it captures the central methodological point.
Objection: Bayesian priors are subjective, so the critique proves little
Since the argument repeatedly invokes priors and prior predictive distributions, one might object that Bayesian priors are subjective. Different investigators can assign different values to:
or:
If there is no uniquely correct prior distribution over divine motives, perhaps the formalism simply redescribes disagreement rather than resolving it.
This objection identifies a real limitation of the Bayesian presentation, but it does not undermine the underlying argument.
First, the problem does not depend upon there being one uniquely correct numerical prior.
The issue is comparative and structural.
Suppose there are one million possible auxiliary motives and no antecedent reason to privilege one particular motive. After observing $E$, someone selects the one motive that makes $E$ highly likely.
Even if reasonable agents disagree over exact priors, they can still agree that the selected motive should not retrospectively be treated as though all prior probability had been concentrated upon it.
The Bayesian point concerns probability mass and selection, not merely numerical subjectivity.
Second, priors need not be arbitrary. They can be constrained by:
- background evidence;
- symmetry considerations;
- theoretical simplicity;
- previous observations;
- independently justified theological commitments.
The argument is precisely that such constraints must be identified rather than hidden.
Third, even a critic unwilling to adopt Bayesian epistemology can accept the nonnumerical version:
A hypothesis that could easily have been modified to accommodate many contrary outcomes receives less evidential credit from fitting the actual outcome than a hypothesis that independently singled that outcome out.
The Bayesian formalism makes the principle explicit, but does not create it.
Thus:
Objection: perhaps divine reasons really are beyond human comprehension
Perhaps the most serious theological response is simply to accept inscrutability.
God is not a human agent. Divine wisdom may vastly exceed human understanding. It may therefore be unreasonable to demand that finite human beings understand why God selects one action rather than another.
On this view, apparent mismatches between divine action and human expectations should not count heavily against theism because:
does not entail:
This is correct as a matter of logic.
From:
“I cannot identify God's reason for permitting or producing $E$,”
one cannot infer:
“There is no such reason.”
The argument developed here should not make that inference.
But the epistemic consequence of inscrutability cuts both ways.
If we are poorly positioned to know what God would choose, then we are correspondingly poorly positioned to make strong likelihood claims such as:
This is exactly what we would expect God to do.
The tension can be formulated:
This may protect theism from some forms of disconfirmation.
But it simultaneously weakens arguments from:
- miracles;
- beauty;
- fine-tuning;
- religious experience;
- providence;
- revelation;
whenever those arguments depend on judgments about what God would probably choose.
The objection therefore does not refute the divine-psychology problem. It sharpens the dilemma.
One may choose a relatively thick divine psychology, permitting strong predictions but also exposing the theory to contrary evidence.
Or one may choose a relatively thin divine psychology, protecting the theory with inscrutability but weakening its predictive content.
What is difficult is to use both opportunistically:
confident psychology when the observation favors theism;
inscrutable psychology when it does not.
Thus:
Objection: a low-probability miracle could still provide strong evidence
Another objection is that the preceding argument may be conflating absolute probability with evidential support.
Evidence need not be highly probable under a hypothesis to support it. What matters is the likelihood ratio.
An observation $E$ can favor $G$ whenever:
even if both likelihoods are small.
For example:
and:
would make $E$ strong evidence for $G$, despite $E$ being rare under $G$.
This objection is entirely correct.
The argument should therefore not say:
Unless theism made the miracle highly probable, it cannot be evidence.
That would be false.
The proper claim is comparative:
To argue that $E$ favors theism, one needs a reason to think the likelihood under theism exceeds the likelihood under relevant alternatives.
The divine-psychology problem remains because estimating:
still requires an account of divine preferences.
A naturalistic event could be exceedingly rare and yet a miraculous event even rarer under a poorly constrained divine model—or vice versa. The relevant issue cannot be settled merely by calling one event “unlikely naturally.”
Thus the stronger formal statement is:
But this comparison cannot be performed responsibly if the numerator is inflated by selecting a convenient divine motive after observing $E$.
This objection therefore improves the argument by making its probabilistic target more precise.
Objection: naturalistic explanations can also be highly flexible
A defender of miracle or design explanations may reasonably respond that naturalistic theorizing is hardly immune to ad hoc reasoning.
Naturalistic explanations can appeal to:
- unknown psychological mechanisms;
- undocumented social processes;
- rare physical events;
- speculative evolutionary pathways;
- unobserved historical circumstances.
If critics insist on strict explanatory discipline for divine agency, they should apply the same standards to naturalism.
This objection should be accepted.
Indeed, methodological symmetry is central to the argument.
A claim such as:
“Some unknown natural process caused it”
has no automatic explanatory superiority over:
“God caused it.”
Both are underspecified.
Similarly:
“Evolution somehow produced this”
is not a satisfactory explanation merely because evolution is naturalistic.
The relevant naturalistic hypothesis must identify enough mechanism and constraint to possess explanatory content.
The miracle material explicitly recognizes this point: “nature did it” is vacuous if it is not supplemented by an actual causal account.
The argument therefore does not establish a metaphysical privilege for naturalism.
Rather, it establishes a methodological preference for better constrained hypotheses.
If, in a particular case, the naturalistic hypothesis is more speculative, more ad hoc, and less independently supported than a well-specified divine explanation, then the divine explanation may legitimately be preferable.
That possibility must remain open if the methodology is genuinely symmetrical.
The claim developed throughout this essay is narrower: in many familiar arguments, the divine explanation is not subjected to the same degree of specification demanded of its naturalistic competitors.
Correcting that asymmetry is not naturalistic bias. It is explanatory consistency.
Objection: extraordinary singular events cannot make “risky predictions”
Miracles and historical revelations are often one-time events. A resurrection cannot simply be repeated under laboratory conditions. Fatima occurred in a particular historical setting. Perhaps demanding independent prediction is inappropriate for singular historical claims.
Again, the objection identifies an important qualification.
A hypothesis need not predict a repeated instance of the exact same phenomenon to acquire independent support.
A historical hypothesis can generate consequences concerning:
- independent documents;
- geographical patterns;
- artifacts;
- chronological relationships;
- testimony from sources not used in constructing the hypothesis;
- previously unknown details;
- expected absence of certain evidence.
Likewise, a theological theory constraining divine action might generate predictions about other cases.
For example, if one proposes a general principle:
God authenticates genuine revelation by manifestations satisfying conditions $C_1,C_2,C_3$,
then this principle can be tested against religious claims not used to construct it.
Similarly, if a theory says:
God provides revelation in a way that preserves condition $R$,
then one can ask whether independent cases of purported revelation consistently exhibit $R$.
The requirement is therefore not laboratory repetition.
It is independent evidential reach.
A proposed explanatory principle becomes stronger when it successfully applies beyond the case that motivated it.
Thus:
Historical and theological hypotheses can take epistemic risks by making claims whose truth is not guaranteed by the evidence that generated them.
Objection: perhaps the evidence supports a more specific hypothesis, not bare theism
Another sophisticated response is to accept the critique of:
but argue that this is irrelevant.
Perhaps the evidence is not supposed to support “God exists” in isolation. It supports:
where $T$ is a substantive theology and $M$ a particular account of divine purposes.
This is legitimate Bayesian reasoning. Evidence can certainly confirm a specific composite hypothesis.
The issue then becomes one of hypothesis accounting.
If the actual explanation is:
then one should not rhetorically present the evidence as straightforward confirmation of bare $G$.
The full explanatory package includes:
- God;
- theological doctrines;
- motivational assumptions;
- intervention policies.
The evidential comparison must evaluate that package.
This actually sharpens the argument rather than defeating it.
It reveals that many claims superficially described as:
“Evidence for God”
are more accurately:
“Evidence for God plus a highly specific theology of what God intends and how God acts.”
The probability of the conjunction matters.
In Bayesian terms:
is not automatically equivalent to:
A highly detailed hypothesis may obtain an excellent likelihood:
while paying an epistemic price if the added components had weak antecedent support.
This is exactly why detailed post-hoc hypotheses can fit extraordinarily well without receiving proportionately high posterior probability.
The proper response is not to forbid specific theology.
It is to make the specificity visible.
Objection: the criterion risks privileging prediction over understanding
Someone might argue that an explanation has value even if it does not significantly raise probabilities. A theological story may deepen understanding by embedding an event within a larger metaphysical narrative. Perhaps this sort of explanation is not trying to compete with scientific prediction at all.
This is a reasonable distinction.
There are multiple senses of “explanation.”
A theological interpretation may provide:
- existential meaning;
- narrative coherence;
- symbolic significance;
- moral interpretation;
- integration within a worldview.
Nothing in the present argument shows that these functions are worthless.
The mistake occurs only when narrative or existential intelligibility is converted into evidential confirmation without further argument.
One may legitimately say:
Within Christianity, the resurrection narrative makes theological sense as vindication and salvation.
That is different from:
The fact that the resurrection would make theological sense is strong evidence that it occurred.
The second is an evidential claim and therefore inherits the likelihood problem.
The distinction should therefore be preserved:
A worldview may interpret an event richly without thereby demonstrating that the worldview made the event antecedently probable.
Does the argument establish a global defeater for theism?
The breadth of the pattern may invite a stronger conclusion. If divine psychology affects fine-tuning, miracles, revelation, providence, religious experience, hiddenness, and design, perhaps it defeats theism itself.
That conclusion would be too quick.
The argument developed here is best understood as a global undercutting challenge to a class of evidential arguments, not automatically as a direct refutation of theism.
It targets inferences of the form:
when the required divine psychology is weakly constrained.
A person might possess other reasons for theism:
- metaphysical arguments;
- ontological commitments;
- independent religious experience;
- philosophical arguments not relying upon particular divine choices.
Whether those arguments succeed is a separate matter.
The present argument says that one cannot easily turn observed features of the world into confirmation of intentional divine action merely by supplying a possible divine reason after the fact.
Thus the conclusion should be carefully stated:
rather:
Its scope is nevertheless broad because so many evidential arguments for personal theism involve precisely such claims.
What would a successful divine explanation look like?
The objections above reveal that the argument should ultimately be constructive. If divine explanation is not impossible in principle, what would count as a strong one?
At least five conditions appear important.
1. Independently motivated divine psychology
The proposed motive should not be inferred solely from the event it explains.
There should be independent reason to believe:
The support could come from philosophical theology, other evidence, or a broader theory—but its dependence upon the target evidence must be made explicit.
2. Genuine contrastive constraint
The motive should make some actions more expected than others.
It should answer:
Why this rather than something else?
A theology that says only:
God has reasons,
does not satisfy this requirement.
3. Means–end coherence
Given the attributed motive, the observed action should be a plausible means of realizing it.
For:
and:
one needs some reason to regard:
as comparatively substantial.
This is especially important for omnipotent agency because ordinary constraints on means are absent.
4. Independent evidential reach
The theological principle should explain or predict phenomena beyond the original datum.
An initially speculative principle gains credibility if it successfully accounts for cases not used in its construction.
The ideal pattern remains:
5. Exposure to evidential risk
A successful explanatory theory should tell us what would count against it.
This need not mean strict falsification.
But there should be conceivable observations such that:
Without such differences, the hypothesis has little predictive content.
The qualified conclusion
The objections therefore refine rather than overturn the central argument.
Agent explanations need not be deterministic.
Probabilities need not be numerical.
Post hoc explanations can be excellent.
Auxiliary assumptions are unavoidable.
Theology can provide genuine background knowledge.
Naturalistic explanations can themselves become ad hoc.
Singular historical events can be legitimately explained.
None of these concessions threatens the central distinction.
What matters is whether the explanatory structure is independently constrained enough to make the actual observation differentially expected.
The most defensible general principle is therefore not:
“Divine explanations are inherently bad.”
Nor:
“Any explanation constructed after seeing the evidence is ad hoc.”
Nor:
“Only hypotheses that make precise numerical predictions are explanatory.”
It is:
An intentional explanation earns evidential force only insofar as independently warranted information about the agent's motives, constraints, and decision procedures makes the observed action more expected than relevant alternatives.
Where that information is unavailable, modesty is required.
One may conclude:
God could have had a reason.
One may conclude:
This event can be interpreted coherently within this theology.
One may even conclude:
The divine hypothesis remains possible.
But none of those conclusions by themselves establishes:
much less:
That is the burden divine psychology reveals.
The challenge is not to eliminate intelligent agency from explanation, but to demand that an agent earn its explanatory role in the same way any other explanatory posit does: by constraining expectations rather than merely expanding the range of stories that can be told after the evidence is already known.
Conclusion: From Explanatory Flexibility to Explanatory Discipline
The argument developed throughout this essay began with a relatively simple question: what must an explanation do in order to count as genuinely evidential rather than merely compatible with what we already know? The answer is that explanation requires constraint. A hypothesis earns explanatory force not simply by being capable of producing the observed outcome, but by giving us some antecedent reason to expect that outcome rather than the relevant alternatives.
This distinction becomes especially important when explanation proceeds through intelligent agency. Ordinary agent explanations work because agents are not treated as bare causal powers. Human behavior is interpreted against extensive background knowledge about motives, capabilities, circumstances, limitations, psychology, and available means. That information constrains what actions are reasonable to expect. The fact that an agent could have produced an outcome is only the beginning of an explanation; the substantial work is done by a theory of why the agent would have selected that outcome.
Kevin Scharp's problem of divine psychology reveals what happens when this structure is applied to God. Arguments from divine action routinely move from general theological claims to highly specific conclusions about what God would create, reveal, permit, or cause. Yet the transition from:
to:
requires assumptions about divine intentions that are not contained in the bare existence claim itself. The actual explanatory hypothesis therefore has a richer form:
where $M$ represents divine motives and $A$ the selected divine action.
Once those variables are made explicit, the central problem becomes visible. An omnipotent agent possesses an extraordinarily large action-space. Causal capacity therefore does not narrow the possibilities in the way ordinary physical or practical constraints do. If divine preferences are also weakly specified, then almost any observed state of affairs may be reconciled with some possible divine purpose.
This yields one of the central conclusions of the essay:
A being capable of producing almost anything is trivially capable of producing what actually occurred. What matters evidentially is why this outcome should have been selected.
The analysis of ad hoc reasoning makes the epistemic consequence more precise. A theory may improve its fit with existing evidence by introducing additional assumptions:
But if $A$ was selected specifically because it makes the observed data $D$ unsurprising, then the relevant question is not merely whether:
One must also ask what reason existed to privilege $A$ independently of $D$. A flexible hypothesis space makes it easy to find some auxiliary assumption that fits an already-known observation. Bayesian model comparison captures this by evaluating the entire prior predictive distribution rather than merely the best-fitting parameterization:
The corresponding theological problem is therefore:
where $\theta$ represents possible divine motives, priorities, and intervention strategies. Showing that there exists some retrospectively selected $\theta^*$ for which:
does not establish that theism itself strongly predicted $E$.
This distinction between prediction and accommodation is the thread connecting every case considered.
In miracle reasoning, an anomaly does not become a miracle merely because ordinary explanations remain incomplete. The inference to divine intervention requires an account of why God would produce the particular event, for the particular recipients, at the particular time, and by the particular means observed. The resurrection, the Shroud of Turin, and Fatima each illustrate how quickly the explanatory burden shifts from divine capability to divine choice. The more specific the theological conclusion becomes, the more substantial the required divine psychology becomes as well.
The same pattern appears in revelation and hiddenness. Clear revelation may be explained by saying that a loving God wishes to be known, while ambiguous revelation may be explained by saying that God values freedom, epistemic distance, soul-making, or some unknown providential purpose. Each response may be conceivable. But a theory capable of accommodating both clarity and obscurity through interchangeable motives cannot claim the full evidential benefit of whichever outcome occurs. The flexibility that protects a hypothesis from disconfirmation necessarily limits how strongly a particular observation can confirm it.
Intelligent Design demonstrates that the problem is broader than theology in the narrow sense. Agency itself is not explanatorily defective. Human design inference works precisely because human agents are independently characterized. By contrast, an unspecified designer whose goals, constraints, and intervention policy remain unknown may add causal possibility without adding predictive structure. Biological efficiency can be attributed to a preference for efficiency; inefficiency to competing design objectives; redundancy to robustness; novelty to creativity; reuse to economy. Unless these preferences are independently constrained, the designer risks becoming a residual variable assigned whatever psychology the data require.
The generalized pattern is therefore:
The epistemic mistake occurs in the final transition.
Retrospective coherence is not worthless. A theory may provide genuine narrative, historical, theological, or conceptual understanding even when it was formulated after the event. Nor is every auxiliary hypothesis introduced in response to evidence illegitimate. The crucial issue is whether the explanation subsequently acquires independent support, generates new consequences, integrates with established knowledge, and exposes itself to evidential risk. The distinction is not between theories that change and theories that do not. It is between revision disciplined by independent evidence and flexibility used merely to preserve fit.
The argument therefore does not establish that divine action is impossible, that miracles cannot occur, that intelligent agency can never explain biological structures, or even that theism itself is false. Its conclusion is narrower but far-reaching. It undercuts a broad class of evidential arguments whenever those arguments rely upon weakly constrained assumptions about what a divine or intelligent agent would probably choose.
The governing principle can be stated simply:
and:
A possible divine motive can make an event intelligible without making it probable. An unspecified designer can account for a structure in the sense of being capable of producing it without explaining why that structure should have been expected. An unknown reason can preserve logical compatibility without generating confirmatory likelihood.
What matters is whether the explanation discriminates.
A strong explanatory theory tells us not merely why the actual world can be fitted into its framework, but why substantially different worlds would have been less expected. It thereby takes an epistemic risk. A theory that can accommodate almost any outcome avoids that risk, but precisely for that reason receives less evidential credit when one particular outcome occurs.
The central tension can therefore be summarized:
while, absent additional independent constraint,
This is the broader significance of divine psychology. Scharp's insight identifies not merely a localized difficulty for one apologetic argument, but an exemplary instance of a general problem in explanatory reasoning. Whenever an agent is introduced whose powers are broad, whose motives are weakly constrained, and whose intentions are inferred largely from the event already under consideration, the possibility of explanation threatens to become indistinguishable from the ability to construct a fitting story.
The appropriate response is not to prohibit agent explanations, supernatural explanations, or theological explanations in advance. It is to demand explanatory discipline. What did the hypothesis independently commit itself to? What alternatives did it render less likely? What motives were justified before the observation was used to select them? What would the theory have found surprising? And what new evidence follows from it that was not already used in its construction?
Those questions mark the boundary between a theory that genuinely explains and one that merely accommodates. The decisive issue is not whether one can tell a coherent story about why the world we observe might have been chosen. It is whether the theory gave us reason, independently of already knowing the answer, to expect a world like this one rather than something else.
Comments
Post a Comment