206 – A myriad of maps – The Conscious Look

How we hold many contradictory models at once

In 1956, the social psychologist Leon Festinger published a book that began with an unusual research method. Festinger and two colleagues had infiltrated a small doomsday cult in the American Midwest whose members believed that a great flood would destroy the world on December 21, 1954, and that flying saucers would arrive to rescue the true believers before the catastrophe. The group had sold their possessions, quit their jobs, and assembled to await the end. When December 21 came and went without flood or spacecraft, Festinger documented what happened next. The group did not disband. They did not revise their beliefs. They announced that their faith had been so powerful that God had decided to spare the world, and their proselytizing became more fervent than before. The disconfirmation of the prophecy did not weaken the belief. It strengthened it.¹

Festinger called the discomfort of holding simultaneously contradictory beliefs cognitive dissonance, the tension that arises when two elements of a person’s belief system cannot both be true. What his research revealed was that this discomfort is real and motivating, but that the most common response to it is not the revision of the belief that is contradicted by evidence. It is the revision of the relationship between the belief and the evidence: the generation of an explanation for why the disconfirmation does not actually disconfirm. The group did not conclude that they had been wrong about the flood. They concluded that their faith had averted it. The map had been preserved by changing the story about the territory.

This article is about a property of human cognition that Festinger’s research illuminates but does not fully describe. We do not carry a single, coherent, internally consistent model of the world. We carry dozens, perhaps hundreds, of domain-specific maps built at different times, from different sources, in different emotional contexts, for different purposes, and we rarely check whether they are consistent with each other. The result is not pathology. It is the normal condition of a mind that has been built for navigation rather than for logical coherence, and that maintains multiple maps precisely because the world it navigates is too large and too varied for any single map to cover. The question the article examines is not how to achieve perfect consistency (that ambition misunderstands the architecture) but what the characteristic failure modes of inconsistency look like, and what can be done to manage them honestly.

The pluralist mind

The unified self is a useful fiction. The experience of being a single, coherent person with a stable set of beliefs, values, and dispositions is one of the most powerful and most persistent constructs the human mind generates, and it is, in significant part, a construction. The neuroscientist Michael Gazzaniga’s split-brain research demonstrated this with unusual precision: patients whose corpus callosum had been severed (disconnecting the two hemispheres of the brain) could be shown to hold simultaneously contradictory beliefs, one in each hemisphere, with neither hemisphere aware of the contradiction.² The left hemisphere, which governs language and narrative, would generate confident explanations for behaviors that had been initiated by the right hemisphere for entirely different reasons, explanations that the left hemisphere had invented on the spot to make the behavior seem coherent, because the left hemisphere had no access to the right hemisphere’s actual motivation. Gazzaniga called this narrative function the interpreter: the module whose job is to generate a coherent story about what is happening, even when no coherent story is available.

The intact brain does not present the same clean experimental separation, but the architecture it reveals is recognizably the same. The psychologist Jonathan Haidt, drawing on both the evolutionary history of the brain and the experimental literature on moral judgment, describes the relationship between the two systems through an image that is simpler and more honest than most available alternatives: the elephant and the rider.³ The elephant is the large, fast, emotionally powerful system that generates responses: the system that produces immediate attraction or aversion, trust or suspicion, moral approval or disgust, before any deliberate reasoning has occurred. The rider is the small, slow, linguistically competent system that sits on top and constructs explanations: coherent, articulable, apparently reasoned accounts of why the elephant went where it went. The rider did not choose the direction. The rider arrived after the elephant had already moved and composed a narrative about the journey.

Haidt’s evolutionary argument is important and is often missed in popular presentations of the metaphor. The rider did not evolve to govern the elephant. It evolved primarily to justify the elephant’s behavior to others, to serve as the social interface through which an inherently social animal negotiates its standing in a group. This means that the rider’s explanations are not straightforwardly honest accounts of causes. They are constructions optimized for social persuasiveness, which is a different optimization target from truth. When the rider says “I believe this policy is wrong because of these principled reasons,” the more accurate description is often: the elephant had a strong negative response to this policy, and the rider has generated the most persuasive available account of why that response is justified. The principled reasons are real enough in themselves (they are not made up) but they are selected from the available space of reasons by the prior response of the elephant, rather than being the causes of the response. This is what Gazzaniga’s interpreter does in the split-brain patient: it selects from available explanations the one that best accounts for the behavior already performed. The difference in the intact brain is that the behavior and the explanation happen closer together in time, and the gap between them is correspondingly harder to notice.

Haidt has documented a related phenomenon he calls moral dumbfounding: the experience of having a strong moral intuition (a firm conviction that something is wrong) without being able to articulate any coherent reason why. When subjects are presented with scenarios that trigger moral discomfort but that involve no identifiable victim and no apparent harm, they typically generate reasons, find those reasons refuted, generate more reasons, find those refuted too, and eventually arrive at the position Haidt calls “I know it’s wrong, I just can’t explain why.” What they are experiencing is the collision between a moral map built at one level (the level of intuition and emotion, in the elephant) and a reasoning process built at another level, in the rider, and the failure of the two to generate a consistent output.⁴ The maps were not designed to be consistent with each other. They were built for different purposes and they answer different questions.

Domain-specificity and its costs

The architecture of the pluralist mind is not a design flaw. It is the solution to a genuine engineering problem. A single, fully coherent, universally applicable model of the world would need to be so complex that it could not be computed in real time, and so abstract that it would lose the domain-specific precision that makes models useful. The brain’s solution (many domain-specific models, each built and maintained separately, each optimized for its particular context) is computationally efficient precisely because it does not try to be globally coherent. The chess grandmaster’s model of a chess position does not need to be consistent with their model of a business negotiation. The parent’s model of their child does not need to be consistent with their model of children in general. Each model does its job in its own domain.

The cost of this architecture is the characteristic inconsistency that Festinger, Haidt, and Gazzaniga’s work illuminates from different angles. But there are two additional costs that are worth naming specifically, because they are the ones most likely to produce consequential errors in the domains this series is most concerned with.

The first is what might be called the selective skepticism problem: the tendency to apply very different evidential standards to claims in different domains, depending on which model is activated. The person who demands rigorous double-blind trials as evidence for pharmaceutical efficacy may accept testimonials as sufficient evidence for alternative therapies, not because they have consciously decided that different standards apply in different domains but because the two beliefs are stored in different models that have never been directly compared. Their epistemic standards are not globally consistent; they are locally applied to whichever domain is currently active, and the inconsistency is invisible until the two domains are placed explicitly side by side. Article 207 of this series examines this pattern in the specific context of the homeopathy debate, where exactly this double standard is most clearly visible and most practically consequential.

The second cost is model lag: the tendency for one domain-specific model to update more slowly than another, producing an inconsistency that the person is not aware of because the two models are rarely activated simultaneously. The scientist who holds a rigorously empirical model of causation in their professional domain and a loosely evidential model of causation in their personal life (attributing their professional successes to their own ability while attributing their personal difficulties to bad luck) is experiencing model lag. The models were built at different times, from different evidence, and have accumulated different degrees of evidential updating. They are inconsistent not because the person is irrational but because the updating process is domain-specific rather than global. New evidence in one domain does not automatically propagate to related beliefs in another.

The special case of identity-adjacent beliefs

Not all beliefs are equally resistant to revision. The beliefs that are most difficult to bring into consistency with other beliefs are the ones that are most closely associated with the person’s sense of who they are: their political identity, their religious commitments, their professional self-concept, their tribal affiliations. These beliefs have a property that purely factual beliefs do not: they are not held primarily as models of how the world is but as markers of who the believer is and which communities they belong to. In the terms of article 211’s discussion of crony beliefs, they are held for social and identity reasons rather than purely epistemic ones, and their revision therefore carries a social cost that factual belief revision does not.

The consequence is that identity-adjacent beliefs form the most stable nodes in the network of maps: the points around which other beliefs are arranged and to which they are adjusted, rather than the other way around. When a new piece of evidence arrives that conflicts with a factual belief, the factual belief is revised. When a new piece of evidence arrives that conflicts with an identity-adjacent belief, it is more likely that the evidence will be reinterpreted, questioned, or assimilated into a narrative that preserves the identity-adjacent belief intact. The map is preserved by changing the story about the territory, exactly as Festinger’s doomsday cult members did in 1956.

This is not a phenomenon that afflicts only the credulous or the poorly educated. Research by Dan Kahan and colleagues at Yale Law School has consistently found that higher levels of scientific literacy and quantitative reasoning skill are associated not with more accurate assessments of politically contested empirical questions but with more sophisticated generation of reasons to support whichever position is consistent with the respondent’s cultural identity.⁵ Greater cognitive ability, in the presence of identity-adjacent beliefs, produces better motivated reasoning rather than better reasoning. The map has more resources available for its own defense, and it uses them.

Holding multiple maps well

None of this is an argument for abandoning the domain-specific architecture of the pluralist mind. The architecture is not optional (it is the way the mind works), and the appropriate response to understanding it is not the impossible demand for global consistency but the more achievable practice of periodic, deliberate comparison.

The comparison that is most valuable is not between randomly selected beliefs but between beliefs in adjacent domains that ought, on reflection, to be governed by similar evidential standards. Someone who believes that clinical trials are necessary to establish the efficacy of pharmaceutical interventions can ask whether the same standard applies to non-pharmaceutical health interventions. Someone who believes that poverty is primarily explained by structural factors when discussing public policy can ask whether the same framework governs how they explain the economic outcomes of people they know personally. Someone who believes that the historical track record of an institution is irrelevant when arguing for reform can ask whether the same standard holds when the institution is one they are attached to.

These comparisons are uncomfortable precisely because they surface inconsistencies that the domain-specific architecture was designed to prevent from being noticed. This is the discomfort of The Conscious Look applied to the plurality of one’s own maps: the experience of placing two models that have been stored separately side by side and asking whether, if they were both true, they could be true simultaneously. The question is diagnostic rather than devastating: the goal is not to achieve a perfect logical consistency that the architecture does not support, but to identify the inconsistencies that are consequential enough to be worth addressing and to address them honestly rather than by generating ever more sophisticated explanations for why they are not actually inconsistencies.

The deepest form of this practice is the one that Festinger’s doomsday cult members could not perform: the willingness to let disconfirmation do its work. When evidence conflicts with a belief, the question worth asking is not how the evidence can be accommodated within the existing map but whether the existing map needs revision. The two responses (accommodation and revision) are both available in most situations. The Conscious Look is the practice of noticing when accommodation is being chosen for reasons of identity preservation rather than genuine epistemic assessment, and of asking what the alternative would look like if it were tried.

Notes

¹ Festinger, L., Riecken, H. W., and Schachter, S. (1956). When Prophecy Fails: A Social and Psychological Study of a Modern Group That Predicted the Destruction of the World. University of Minnesota Press. The group Festinger studied was led by a woman he called Marion Keech, a pseudonym for Dorothy Martin, a housewife from suburban Chicago who claimed to receive messages from alien beings she called the Guardians. The specific prediction (December 21, 1954, not 1956 as sometimes misreported) failed as described. The group’s post-disconfirmation response, and the increase in proselytizing activity that followed, became the primary evidence for Festinger’s cognitive dissonance theory and for the specific prediction that the theory makes about the circumstances under which belief is strengthened rather than weakened by disconfirmation.

² Gazzaniga, M. S. (1967). The split brain in man. Scientific American, 217(2), 24-29. Gazzaniga’s research with Roger Sperry, for which Sperry received the Nobel Prize in Physiology or Medicine in 1981, established the functional lateralization of the brain’s two hemispheres and the role of the corpus callosum in integrating their outputs. The interpreter function (the left hemisphere’s tendency to generate post-hoc explanations for behaviors initiated by processes it has no access to) is described in detail in Gazzaniga’s later popular work, particularly The Social Brain (1985) and Who’s in Charge? Free Will and the Science of the Brain (2011).

³ Haidt, J. (2006). The Happiness Hypothesis: Finding Modern Truth in Ancient Wisdom. Basic Books. The elephant and rider metaphor is introduced here and developed in the context of the relationship between conscious deliberation and automatic emotional response. The metaphor is extended and given a more explicitly evolutionary and social framing in Haidt’s later work, The Righteous Mind: Why Good People Are Divided by Politics and Religion (2012), where the rider’s primary function is identified as social justification rather than rational governance. The evolutionary argument (that language and reasoning evolved primarily as social tools rather than as cognitive tools) is developed in parallel in the work of the evolutionary psychologist Robin Dunbar, whose social brain hypothesis proposes that the large human neocortex evolved primarily to manage the complexity of social relationships rather than to solve abstract problems. Dunbar, R. I. M. (1998). The social brain hypothesis. Evolutionary Anthropology, 6(5), 178-190. The two arguments converge on the same conclusion: the rational, language-using system is, in evolutionary terms, younger, smaller, and less powerful than the emotional system it appears to govern, and its governance is substantially more limited and more post-hoc than the phenomenology of deliberate reasoning suggests.

⁴ Haidt, J. (2001). The emotional dog and its rational tail: A social intuitionist model of moral judgment. Psychological Review, 108(4), 814-834. The moral dumbfounding experiments, in which subjects maintained strong moral convictions they could not rationally justify, were reported in Haidt, J., Koller, S. H., and Dias, M. G. (1993). Affect, culture, and morality, or is it wrong to eat your dog? Journal of Personality and Social Psychology, 65(4), 613-628. The scenarios used included harmless taboo violations (eating a chicken carcass purchased from a supermarket for sexual purposes, cleaning a toilet with the national flag) that subjects reliably judged to be morally wrong despite being unable to identify any victim or any harm.

⁵ Kahan, D. M., Peters, E., Dawson, E., and Slovic, P. (2013). Motivated numeracy and enlightened self-government. Behavioural Public Policy, 1(1), 54-86. The key finding was that subjects with higher numeracy skills were better able to solve a neutral statistical reasoning problem correctly, but that when the same statistical problem was framed in terms of a politically contested question (gun control), subjects with higher numeracy were more likely to reach the conclusion consistent with their political identity rather than the mathematically correct one. The effect was symmetric: it held for both liberal and conservative identifiers, and it was larger for higher-numeracy subjects. The paper’s title, “Motivated Numeracy and Enlightened Self-Government,” captures the political implication precisely.

206 – A myriad of maps