Epistemic Parallax

State-Modified Abduction and a Distinctly Human Epistemology

Abstract

Human reasoning is often evaluated against an implicit ideal of the stateless reasoner: a cognitive agent presented with a fixed evidentiary record and a determinate set of hypotheses who then selects the hypothesis best supported by the evidence. Actual human inquiry begins earlier and under more severe constraints. Before hypotheses can be compared, they must be generated; yet the space of possible explanatory models is effectively intractable for a finite reasoner. Moreover, human cognition is state-dependent. Affect, arousal, motivation, bodily condition, memory accessibility, and attentional orientation systematically alter what information becomes salient and which explanatory possibilities become cognitively available.

This article develops epistemic parallax, a model according to which some forms of state-dependent cognitive bias may contribute positively to human knowledge because differently configured states generate systematically different partial representations of the same underlying situation. The analogy is binocular vision: the two eyes do not provide identical retinal representations, and depth perception depends partly on the disparity between them. Binocular disparity is a well-established basis of stereoscopic depth perception. (ScienceDirect⁠) Analogously, differently biased cognitive states may generate discrepant explanatory models whose divergence supplies information unavailable from either perspective alone.

The theory modifies Charles Sanders Peirce’s tripartite account of inquiry. Peircean abduction generates explanatory hypotheses, deduction derives their consequences, and induction subjects those consequences to experience. (Cambridge University Press⁠) We propose that human abduction is state-modified: internal state changes both the region and resolution of hypothesis space sampled by a bounded reasoner. State-dependent biases can therefore function as differentiated abductive proposal systems. Threat-oriented states may preferentially generate models of hostility, deception, or danger; affiliative states may preferentially generate cooperative or benign explanations; exploratory states may broaden the range of candidate models. The epistemic value lies not in treating any state as authoritative, but in retaining complementary models and integrating their disparities through subsequent deduction and induction.

The resulting framework distinguishes local bias from global epistemic competence, hypothesis generation from hypothesis weighting, and action-sufficient from explanatory abduction. It offers a distinctly human epistemology: one in which embodiment, changing state, finite computation, urgency, and systematic bias are not simply deviations from rational cognition but potential components of the architecture by which finite agents construct increasingly adequate models of reality.

Keywords: abduction; Peirce; cognitive bias; emotion; epistemic parallax; bounded rationality; salience; stereopsis; embodied cognition; hypothesis generation; ecological rationality

1. Introduction: The Missing Problem in Theories of Reasoning

Much of epistemology and decision theory begins after one of the hardest cognitive problems has already been solved.

A reasoner is imagined as possessing some evidence E and considering a collection of hypotheses:

H_1,H_2,\ldots,H_n.

The normative question is then how strongly the evidence supports each hypothesis, which inference is logically valid, or which decision maximizes expected utility.

But where did H_1,\ldots,H_n come from?

For sufficiently complex problems, there is no psychologically available set containing every plausible explanatory model. The number of causal structures potentially capable of explaining an ambiguous social interaction, scientific anomaly, medical presentation, legal dispute, or environmental event is enormous. Human beings cannot enumerate those structures before deciding which deserve consideration.

This is a problem of abductive search.

Charles Sanders Peirce identified abduction as the inferential operation through which explanatory hypotheses enter inquiry. Deduction determines what follows from those hypotheses; induction tests them through experience. Later treatments likewise distinguish the generation of explanatory hypotheses from merely choosing among hypotheses already supplied to the reasoner. (Cambridge University Press⁠)

Yet identifying abduction does not solve its computational problem.

If explanatory possibilities are numerous while cognition is finite, then human reasoners require mechanisms for determining which very small portion of explanatory space is worth representing at all.

The present article proposes that affective and physiological state is one such mechanism.

Its central claim is:

State-dependent cognitive variation can be epistemically functional because differently configured states sample different, context-sensitive regions of an otherwise intractable hypothesis space.

The proposal goes beyond the familiar observation that mood affects judgment. Emotion has documented effects on perception, attention, memory, reasoning, and problem solving. (PubMed Central (PMC)⁠) The more specific claim developed here is that those effects can alter which explanatory models are generated before formal evaluation begins.

An even stronger possibility follows. If different states predictably generate different partial models of the same situation, then disagreement among states need not merely represent noise requiring correction. The difference itself may contain information.

We call this epistemic parallax.

2. From Binocular Disparity to Epistemic Parallax

Two human eyes do not receive identical images of the environment. Because they occupy slightly different spatial positions, objects project differently onto the two retinas. The visual system exploits this binocular disparity as an important source of stereoscopic depth information. (ScienceDirect⁠)

The critical point for present purposes is structural.

Neither eye is defective because its representation differs from the other’s. Nor does the visual system determine which eye has produced the uniquely correct two-dimensional image and suppress the alternative. The positional difference between the eyes produces a systematic discrepancy, and the system extracts information from that discrepancy that is unavailable to either retinal representation considered in isolation.

Formally, let an object W be represented from two positions:

R_1=f(W,P_1),

R_2=f(W,P_2).

The representations differ because:

P_1\neq P_2.

The difference

\Delta R=R_1-R_2

is not merely error. Under the appropriate generative model, disparity provides additional information concerning the underlying three-dimensional structure.

The proposed cognitive analogue is:

M_1=A(O,S_1),

M_2=A(O,S_2),

where O is an ambiguous body of observations, S_i is the state of the reasoner, and M_i is an explanatory model generated under that state.

If

S_1\neq S_2,

then it should not be surprising that

M_1\neq M_2.

Conventional bias correction tends to frame such divergence as a problem: which state distorted the evidence?

Epistemic parallax introduces an additional question:

What feature of the underlying situation makes differently tuned cognitive systems resolve it differently?

The discrepancy may sometimes support construction of a model more sophisticated than either initial representation.

A threat-oriented state may produce:

M_1=\text{“This behavior is potentially adversarial.”}

An affiliative state may produce:

M_2=\text{“This behavior may represent attempted connection.”}

The mature model need not choose categorically between them. It may instead discover a structure such as:

M_3=\text{“This is a low-cost social move with multiple favorable branches, including both connection and exploitation of an adverse reaction.”}

The third model contains dimensions made salient separately by the first two.

In this sense, cognitive depth can emerge from perspectival disparity.

3. Peirce and the Architecture of Inquiry

Peirce provides the logical architecture required to make this proposal epistemically disciplined.

In his mature account, abduction is the operation of forming or adopting an explanatory hypothesis. Peirce distinguished it from deduction and induction and treated it as indispensable to scientific discovery. (Cambridge University Press⁠)

The three operations can be represented schematically as:

\textbf{Abduction:}
\quad
\text{What might explain this?}

\textbf{Deduction:}
\quad
\text{If that explanation were correct, what else should follow?}

\textbf{Induction:}
\quad
\text{Do subsequent observations conform to those consequences?}

Their epistemic functions differ.

Abduction is generative.

Deduction is explicative.

Induction is corrective.

This separation becomes particularly important once state-dependent cognition is introduced. A state may legitimately participate in generating a hypothesis without thereby conferring warrant upon it.

This preserves the creative value of affectively modulated cognition without treating felt conviction as evidence.

The modified architecture is:

\boxed{\text{State-Modified Abduction}}

\downarrow

\boxed{\text{Deduction}}

\downarrow

\boxed{\text{Induction}}.

The proposed innovation lies primarily in the first box.

4. The Computational Necessity of Selective Abduction

The complete hypothesis space for a sufficiently complicated observation can be represented as:

\mathcal H=\{H_1,H_2,H_3,\ldots\}.

Human cognition cannot search \mathcal H exhaustively.

This follows naturally from bounded-rationality approaches. Simon rejected models of agents possessing unrestricted computational resources, and contemporary resource-rational theories explicitly analyze cognition as an allocation of finite computation. Lieder and Griffiths, for example, characterize cognition in terms of the effective use of limited computational resources. (Cambridge University Press⁠)

The abductive problem is therefore not simply:

\text{Which }H\in\mathcal H\text{ is best?}

It is first:

\text{Which tiny subset of }\mathcal H\text{ should enter consideration?}

Suppose cognition can explicitly construct only k hypotheses, where:

k\ll|\mathcal H|.

It requires some proposal mechanism:

q(H\mid O,K,C,S,T,U),

where:

• O = observations,

• K = background knowledge,

• C = perceived external context,

• S = internal affective and physiological state,

• T = relevant time horizon, and

• U = urgency or cost of delayed action.

Rather than enumerating the entire hypothesis space, cognition samples from this distribution.

State-modified abduction proposes that:

q(H\mid S_1)\neq q(H\mid S_2).

The consequence is profound.

Different states produce different candidate sets:

G_{S_1}(O)=\{H_1,H_3,H_7\},

G_{S_2}(O)=\{H_2,H_3,H_9\},

G_{S_3}(O)=\{H_1,H_5,H_{11}\}.

Across time, the organism obtains:

\mathcal H^*
=
\bigcup_iG_{S_i}(O),

a diverse but computationally tractable portfolio of models.

The reasoner never searches all possible explanations.

Instead, changing state changes where the search is conducted.

5. Affect as a Change in Cognitive Configuration

This account has substantial precursors in affective science.

Tooby and Cosmides characterize emotions as “superordinate programs” that coordinate multiple cognitive and physiological systems, including attention, inference, memory, learning, motivation, categorization, and physiological regulation. (UCSB Center for Educational Partnerships⁠)

On such an account, fear is not simply a negative feeling appended to otherwise unchanged cognition. It is a reconfiguration of the organism for a class of problems.

Likewise, extensive evidence shows that emotion changes attentional selection and memory processing. (PubMed Central (PMC)⁠) Forgas’s Affect Infusion Model specifically treats affect as influencing constructive social judgment and the information incorporated into it. (PubMed⁠)

Fredrickson’s broaden-and-build theory supplies another important element. Positive emotional states can broaden momentary thought-action repertoires, whereas some negative states narrow them toward more constrained and immediately relevant possibilities. (Prospective Psychology⁠)

Gopnik and colleagues, from a different direction, have shown the importance of variation in hypothesis-search breadth and the exploration–exploitation tradeoff across human development. (PNAS⁠)

These approaches converge on an important point:

Human cognitive search is not invariant.

The present theory asks what follows when that fact is placed explicitly inside Peircean abduction.

6. Bias as Specialized Perspective

The language of “bias” normally implies deviation from an epistemically preferable neutral representation.

Sometimes that description is plainly appropriate. A cognitive process can systematically produce false judgments.

But a different possibility emerges when multiple biased systems are considered together.

Suppose state S_T increases sensitivity to threat. It may lower the threshold for generating hypotheses involving hostile agency:

q(H_{\text{threat}}\mid S_T)
>
q(H_{\text{threat}}\mid S_{\text{safe}}).

This will predictably produce both:

\text{additional true detections}

and

\text{additional false positives}.

The conventional description is that S_T introduces a threat bias.

That description may be locally correct while incomplete at the system level.

If the organism possesses other states that produce different error distributions, then the threat-biased state may function partly as a specialized detector whose job is not to construct the final all-purpose world-model but to ensure that certain costly possibilities enter consideration.

Thus:

\boxed{\text{Local epistemic bias}}

may sometimes contribute to

\boxed{\text{Global epistemic competence}}.

This proposal differs from, while complementing, Error Management Theory.

Haselton and Buss argue that predictable cognitive bias can be adaptive where false-positive and false-negative errors have asymmetrical fitness costs. (PubMed⁠) Haselton and Nettle subsequently developed this into a broader evolutionary account of cognitive biases. (Sage Journals⁠)

Under Error Management Theory, a bias can be adaptive because making more errors in one direction minimizes the expected cost of error.

Epistemic parallax adds another mechanism.

A biased system could be valuable even if its isolated judgments are not better calibrated, provided its systematically different representation contributes nonredundant hypotheses to a higher-order integrative process.

The distinction is:

\textbf{Error management:}
\quad
\text{Bias improves the decision threshold.}

versus

\textbf{Epistemic parallax:}
\quad
\text{Bias improves the diversity of models available to the total system.}

The two mechanisms can coexist.

7. The Difference Between Generation and Weighting

State can influence abductive reasoning in at least two logically separable ways.

First:

S\rightarrow
P(H\text{ becomes cognitively available}).

This is a generative effect.

Second:

S\rightarrow
P_{\text{subjective}}(H\mid H\text{ available}).

This is a weighting effect.

These effects are easily conflated.

Suppose a threat state generates a previously unconsidered hostile-intent hypothesis and simultaneously makes that hypothesis feel overwhelmingly convincing.

It is possible that:

1. generating the hypothesis improves the total model space; while

2. assigning it extreme probability worsens calibration.

Thus the same state can be:

\text{generatively valuable}

and

\text{evaluatively biased}.

This produces a critical epistemic rule:

Preserve the model; discount the state’s claim to grade its own model.

A hypothesis should not be rejected merely because it emerged from anger, fear, sadness, excitement, attachment, or physiological disruption.

But neither should the subjective force associated with those states automatically transfer into evidentiary warrant.

State controls proposal.

Evidence ultimately controls retention.

8. Epistemic Parallax and Higher-Dimensional Model Building

The eye analogy becomes most useful where cross-state disagreement produces a richer explanatory structure rather than merely a choice between alternatives.

Let:

M_A

be an affiliative-state model and

M_T

a threat-state model.

Suppose each captures a different projection of some latent structure L.

Then:

M_A=f_A(L),

M_T=f_T(L).

If the transformations f_A and f_T are systematically different, their disagreement can constrain what L must look like.

The higher-order question becomes:

What underlying structure would make both lower-dimensional interpretations intelligible?

This is meta-abduction.

The reasoner does not ask merely:

Which model wins?

The reasoner asks:

What model explains why these perspectives diverge in precisely this way?

In binocular vision, disparity reveals depth.

In epistemic parallax, interpretive disparity may reveal causal dimensionality.

A situation that appears cooperative under one state and strategic under another may not require selecting “cooperation” or “strategy.” A sufficiently sophisticated model may contain both: cooperative behavior may simultaneously create strategic optionality.

A person may seek affection while also benefiting from control.

An institution may sincerely pursue one stated objective while its structure systematically selects behavior serving another.

A legal actor may pursue a facially legitimate remedy while exploiting collateral reputational effects.

Complex social behavior often permits multiple consequences, motives, levels of awareness, and strategic branches.

State-dependent perspectives may therefore operate as lossy projections of a richer causal structure.

The epistemic achievement occurs when those projections are integrated rather than absolutized.

9. State Determines Not Only Where We Search, but at What Resolution

A second dimension of state-modified abduction concerns model resolution.

Not every situation warrants a sophisticated causal model.

Under immediate danger, the relevant abductive task may be:

\text{“Possible threat.”}

That model may be crude, underdetermined, and incapable of explaining much beyond the immediate action requirement.

Yet it may be entirely adequate to support:

\text{“Do not approach.”}

We call this action-sufficient abduction.

Its optimization problem is approximately:

U_A
=
f(\text{speed},
\text{cost of false negative},
\text{cost of delay},
\text{action sufficiency}).

By contrast, once immediate urgency has passed, the relevant problem may become:

What actually produced this event?

The reasoner can now construct multi-causal models, distinguish motives from consequences, separate primary goals from advantageous branches, and preserve uncertainty among alternatives.

This is explanatory abduction.

Its objective function places greater weight on:

U_E
=
f(\text{causal fidelity},
\text{predictive power},
\text{discrimination},
\text{integration}).

Thus:

\text{short horizon}
\rightarrow
\text{compressed model}

while

\text{long horizon}
\rightarrow
\text{higher-resolution model}.

A fear-based model of “possible danger—avoid” need not be judged by the same standard as a later causal reconstruction.

The first may be epistemically shallow and behaviorally rational.

The second may be behaviorally unnecessary and explanatorily superior.

Theories of reasoning often conflate these tasks.

A distinctly human epistemology should not.

10. State and Context: Why Incidental State Matters

The adaptive account is easiest where internal state tracks relevant external context.

If:

C_{\text{danger}}\rightarrow S_{\text{fear}},

and fear reallocates cognition toward threat-relevant models, the state functions partly as information about the environment.

But internal state can also vary independently:

X\rightarrow S,

where X might include sleep, hunger, illness, pharmacological effects, hormonal conditions, arousal, fatigue, or other bodily variables.

In those cases:

S

may reconfigure abductive search even though:

C

has not changed.

This distinction is important but does not defeat the model.

An incidentally induced threat-like state may still generate a threat hypothesis that the relaxed state failed to produce.

The state may therefore remain useful as a search perturbation even while being uninformative as evidence.

Put differently:

\text{incidental state}
\not\Rightarrow
\text{incidental hypothesis}.

A hypothesis can be accidentally discovered and subsequently prove correct.

What incidental state cannot legitimately supply is the inference:

\text{“I feel this strongly now”}
\therefore
\text{“the world is more likely to be this way.”}

The model thereby draws a sharp distinction between:

epistemic generation

and

epistemic authentication.

11. The Modified Peircean Cycle

The resulting process can be formalized in five stages.

11.1 State-conditioned proposal

Given observations O, context C, background knowledge K, internal state S, available resources R, time horizon T, and urgency U:

G_S
\sim
q(H\mid O,C,K,S,R,T,U).

The system generates a tractable subset of explanatory models.

11.2 Cross-state diversification

Across materially different states:

\mathcal P
=
\bigcup_i G_{S_i}.

\mathcal P is the reasoner’s abductive portfolio.

11.3 Parallax integration

Where models disagree, the reasoner asks whether a higher-dimensional explanation can account for the discrepancy:

M^*
=
\Phi(M_1,M_2,\ldots,M_n,\Delta M),

where \Phi represents integrative model construction and \Delta M represents cross-state disparities.

11.4 Deduction

For each serious candidate:

H_i
\rightarrow
\{P_{i1},P_{i2},\ldots,P_{im}\}.

The emphasis should be on discriminating predictions: observations expected under one model but not another.

11.5 Induction

New evidence constrains the portfolio.

Models lose standing when predictions fail and gain standing when they successfully anticipate observations not used in their construction.

The complete cycle is therefore:

\boxed{\text{State-modified abduction}}

\downarrow

\boxed{\text{Epistemic parallax}}

\downarrow

\boxed{\text{Deduction}}

\downarrow

\boxed{\text{Induction}}

\downarrow

\text{revised state and renewed abduction}.

This retains Peirce’s fundamental insight while supplying an embodied account of the abductive search process.

12. State-Robust, Not State-Neutral, Rationality

The model rejects the assumption that responsible reasoning requires returning to a uniquely neutral state.

Calmness is not absence of state.

Neither is analytical detachment.

Each cognitive configuration has characteristic sensitivities and blind spots.

The appropriate aspiration is therefore state robustness, not state neutrality.

A state-robust inquiry would:

1. identify which models emerged under materially different states;

2. preserve those models without automatically preserving their original confidence levels;

3. note each state’s characteristic salience profile;

4. examine disagreements among states as potentially informative;

5. search for models capable of explaining those disagreements;

6. derive discriminating consequences from rival models;

7. gather observations over time; and

8. revise models according to their performance rather than their phenomenological intensity.

The aim is not:

\text{find the unbiased state}.

It is:

\text{construct a system capable of using multiple biased states without becoming captive to any of them}.

13. A Distinctly Human Epistemology

Classical epistemology often idealizes cognition by subtracting precisely those features most characteristic of actual human inquiry: bodily state, affect, urgency, attentional scarcity, limited memory, changing motivation, incomplete information, and finite computation.

A distinctly human epistemology begins with those conditions rather than treating them as afterthoughts.

The human problem is not:

How should a limitless reasoner process all relevant models?

It is:

How can an embodied organism that can entertain only a handful of possible worlds at one time nevertheless construct sufficiently rich representations of a complex environment?

State-modified abduction offers one answer.

Human states may create structured epistemic heterogeneity within a single reasoner across time.

The anxious person, curious person, angry person, affiliative person, rested person, exhausted person, and reflective person are not literally separate epistemic agents. But neither are they computationally identical configurations.

If each configuration samples somewhat different explanatory structures, then the diachronic self possesses something analogous to an internal diversity of observers.

The human epistemic unit may therefore be less like a single camera than a moving array of differently tuned sensors.

Rationality occurs at the level of integration.

14. Empirical Predictions

The theory generates several testable predictions that distinguish it from the weaker proposition that mood merely changes confidence or judgment.

14.1 State should alter hypothesis identity

Participants presented with identical ambiguous observations but induced into different affective or physiological states should spontaneously generate different categories of explanation, even before being asked to assign probabilities.

Experiments that supply all candidate hypotheses in advance cannot adequately test this prediction because they remove the abductive-generation problem.

14.2 Cross-state reasoning should increase model diversity

Participants who reconsider the same ambiguous problem after meaningful state changes should cumulatively generate a broader range of nonredundant hypotheses than participants who repeatedly reconsider it under similar states.

Formally:

|\mathcal P_{\text{variable state}}|
>
|\mathcal P_{\text{stable state}}|.

The relevant measure should be conceptual diversity, not merely the number of verbal formulations.

14.3 State variation should sometimes improve discovery

The strongest test is whether cross-state portfolios are more likely to contain the ultimately best-supported explanatory model.

If state variation merely generates noise, increased diversity should not systematically improve discovery.

If epistemic parallax is real, at least some problems should show:

P(H^*\in\mathcal P_{\text{variable}})
>
P(H^*\in\mathcal P_{\text{stable}}).

14.4 Cross-state disagreement should sometimes predict model complexity

Where different states produce apparently contradictory but internally coherent interpretations, participants encouraged to explain the disagreement should more frequently construct multi-causal or branch-sensitive models than participants instructed simply to choose the most plausible initial interpretation.

This directly tests the parallax hypothesis.

14.5 States should have characteristic proposal distributions

Threat states should increase generation of danger, deception, hostile-agency, and loss-avoidance models.

Affiliative states should increase cooperative, relational, and benign-intention models.

Positive exploratory states should increase remote or unconventional hypotheses.

The key dependent variable is candidate generation, not merely posterior confidence.

14.6 Urgency should regulate model resolution

High urgency should produce faster, lower-dimensional, action-sufficient models.

Reduced urgency should increase causal decomposition, competing-model representation, and sensitivity to multiple simultaneous motives or consequences.

14.7 Generative value and calibration should dissociate

A state may increase the probability that participants generate a subsequently useful model while simultaneously causing participants to overestimate its probability.

This would provide direct evidence for the distinction between generative benefit and weighting bias.

14.8 Metacognitive state calibration should improve integration

Participants who learn the characteristic error and salience profiles of their own states should become better at integrating cross-state models without either suppressing them or granting them automatic authority.

This suggests a trainable form of abductive metacognition.

15. Objections and Boundaries

15.1 Not every bias is adaptive

The theory does not require that every systematic cognitive error serve a function.

Some biases may be developmental artifacts, cultural products, pathological processes, obsolete adaptations, or straightforward failures.

The narrower proposition is:

Some biases may be functionally intelligible as differentiated components of a larger model-generation architecture.

15.2 Diversity alone does not produce knowledge

Generating ten explanations instead of three does not improve epistemic performance unless the additional models are subsequently constrained.

Uncontrolled abduction can become elaborate rationalization.

The Peircean sequence remains essential:

\text{generate}
\rightarrow
\text{derive}
\rightarrow
\text{test}.

15.3 Recurrence across states does not establish truth

A model that appears repeatedly in different states may reflect stable assumptions, learned schemas, or common memory structures.

Cross-state recurrence is therefore not equivalent to independent evidentiary confirmation.

15.4 Parallax requires partially independent transformations

The binocular analogy works only where states alter cognition in meaningfully different ways.

Two identically biased states provide redundancy, not depth.

The empirical question is therefore whether actual state transitions create sufficiently structured and partially complementary changes in hypothesis generation.

15.5 Integration itself remains state-dependent

There is no external homunculus standing outside the human organism to combine the outputs neutrally.

The integrating reasoner remains embodied and stateful.

For this reason, the normative concept must be state robustness across repeated inquiry, not a mythical state-free adjudicator.

16. Relationship to Existing Theories

The theory is best understood as a synthesis with one additional proposition.

Peirce supplies the distinction between abductive generation, deductive consequence, and inductive testing. (Cambridge University Press⁠)

Simon and resource-rational approaches supply the computational constraint: finite agents must allocate scarce cognitive resources rather than perform exhaustive optimization. (Cambridge University Press⁠)

Forgas, Bower, and affect-cognition research supply evidence that state changes attention, retrieval, interpretation, and judgment. (PubMed⁠)

Fredrickson supplies evidence and theory concerning state-related broadening and narrowing of cognitive repertoires. (Prospective Psychology⁠)

Gopnik and colleagues supply the exploration–exploitation framework and evidence that hypothesis-search breadth itself can vary adaptively. (PNAS⁠)

Tooby and Cosmides supply an account of emotion as coordinated reconfiguration of multiple cognitive subsystems according to adaptive problems. (UCSB Center for Educational Partnerships⁠)

Haselton, Buss, and Nettle establish that systematic cognitive bias need not automatically imply maladaptation and may reflect adaptive responses to asymmetric error costs. (PubMed⁠)

Stereoscopic vision research supplies the structural analogy: systematically different representations can carry higher-order information precisely because they differ. (ScienceDirect⁠)

The additional proposition is:

Because bounded agents cannot generate all possible explanatory models, state-dependent biases may function as differentiated abductive proposal distributions; integrating their disparities can sometimes generate a more causally adequate model than any state produces independently.

The relevant epistemic object is therefore not merely the accuracy of each state considered separately.

It is the performance of the multi-state abductive architecture as a whole.

17. Conclusion: Bias, Depth, and Human Inquiry

Human beings do not encounter the world as stateless inference engines.

We reason while afraid, curious, angry, tired, secure, attached, suspicious, hopeful, hungry, excited, and calm. Those states change what we notice, remember, associate, expect, and imagine.

The dominant normative temptation is to treat this variability as contamination:

\text{state}
\rightarrow
\text{bias}
\rightarrow
\text{error}.

Sometimes it is.

But that cannot be assumed at the architectural level.

A bounded reasoner faces a more fundamental problem: the number of possible explanatory models greatly exceeds the number that can be consciously represented and compared.

Some mechanism must determine where cognition searches.

State-modified abduction proposes that affective and physiological states partly perform that function.

They tune the reasoner toward different regions of explanatory space.

In doing so they can systematically distort.

But systematic difference is not necessarily epistemically useless.

The visual system obtains depth not by forcing both eyes to produce the same image, but by exploiting the structured disparity between them.

Human inquiry may sometimes operate similarly.

\boxed{\text{Different state}}
\rightarrow
\boxed{\text{different salience}}
\rightarrow
\boxed{\text{different abductive model}}

and then:

\boxed{\text{model disparity}}
\rightarrow
\boxed{\text{higher-dimensional explanation}}.

The appropriate response to cognitive bias may therefore sometimes be neither submission nor suppression.

It may be integration.

Threat cognition need not be trusted simply because danger feels salient. Nor should its hypotheses be discarded merely because the system generating them is threat-sensitive.

Affiliative cognition need not be treated as uniquely rational because it is calm. Nor should its benign models be dismissed as naïve merely because another state sees hostility.

Each perspective may reveal structure and conceal structure.

The mature epistemic question is:

What account of reality explains not only the observations, but also why differently situated cognitive states make different aspects of those observations salient?

This is epistemic parallax.

Within a modified Peircean framework, it yields a distinctly human model of rational inquiry:

\boxed{\textbf{State-Modified Abduction}}

generates perspectivally diverse candidate models;

\boxed{\textbf{Epistemic Parallax}}

uses their differences to construct greater causal depth;

\boxed{\textbf{Deduction}}

derives consequences capable of distinguishing those models;

and

\boxed{\textbf{Induction}}

allows an external world to constrain which models survive.

On this account, the epistemic virtue of human reason lies not in becoming stateless.

It lies in learning how to obtain depth from the fact that we are not.

References

Bower, G. H. (1981). Mood and memory. American Psychologist, 36(2), 129–148.

Cosmides, L., & Tooby, J. (2000). Evolutionary psychology and the emotions. In M. Lewis & J. M. Haviland-Jones (Eds.), Handbook of Emotions (2nd ed.). Guilford Press.

Forgas, J. P. (1995). Mood and judgment: The Affect Infusion Model (AIM). Psychological Bulletin, 117(1), 39–66.

Fredrickson, B. L. (2001). The role of positive emotions in positive psychology: The broaden-and-build theory of positive emotions. American Psychologist, 56(3), 218–226.

Gopnik, A., O’Grady, S., Lucas, C. G., Griffiths, T. L., Wente, A., Bridgers, S., Aboody, R., Fung, H., & Dahl, R. E. (2017). Changes in cognitive flexibility and hypothesis search across human life history from childhood to adolescence to adulthood. Proceedings of the National Academy of Sciences, 114(30), 7892–7899.

Haselton, M. G., & Buss, D. M. (2000). Error management theory: A new perspective on biases in cross-sex mind reading. Journal of Personality and Social Psychology, 78(1), 81–91.

Haselton, M. G., & Nettle, D. (2006). The paranoid optimist: An integrative evolutionary model of cognitive biases. Personality and Social Psychology Review, 10(1), 47–66.

Lieder, F., & Griffiths, T. L. (2020). Resource-rational analysis: Understanding human cognition as the optimal use of limited computational resources. Behavioral and Brain Sciences, 43, e1.

Niiniluoto, I. (1999). Defending abduction. Philosophy of Science, 66(S3), S436–S451.

Peirce, C. S. (1931–1958). Collected Papers of Charles Sanders Peirce (C. Hartshorne, P. Weiss, & A. W. Burks, Eds., Vols. 1–8). Harvard University Press.

Qian, N. (1997). Binocular disparity and the perception of depth. Neuron, 18(3), 359–368.

Simon, H. A. (1955). A behavioral model of rational choice. Quarterly Journal of Economics, 69(1), 99–118.

Tooby, J., & Cosmides, L. (1990). The past explains the present: Emotional adaptations and the structure of ancestral environments. Ethology and Sociobiology, 11(4–5), 375–424.