Emotion as noise. What we call reason is organized emotion

A «rational» AI would treat emotion as noise to be filtered out. But Damasio showed, as far back as 1994, that without a somatic marker there's no sensible decision: patients with damage to the orbitofrontal cortex calculate perfectly and always choose the option that loses. Emotion is what prioritizes, what decides, what remembers. Strip it away and you don't get a cleaner system, you get a system with no reasons to do anything at all. What we call «reason» is organized emotion. Whoever denies it hasn't measured the difference.

The patient who decided badly without knowing it

Phineas Gage is the textbook case, and it's worth starting there even if only to avoid stopping there. In 1848, a blasting charge went off at a railway worksite in Vermont and drove an iron rod through the head of the foreman Gage, passing through his frontal lobe. He survived. Physically he recovered well. Cognitively too, on most of the standard tests of the era. What the witnesses reported —and what made it into the textbooks— was something else: Gage's character changed. He became impulsive, unreliable, incapable of keeping medium-term commitments. The phrase «he was no longer Gage» was put on record.

Gage's case opened up the possibility that specific brain regions mediated what we call practical judgment. The hypothesis lay dormant for a century. Antonio Damasio and his collaborators took it up again, with modern instruments, from the 1980s onward. Their research program with patients who had damage to the ventromedial prefrontal cortex —Elliot, EVR, several others— produced a body of evidence worth holding onto.

Bechara, Damasio and others, in Insensitivity to future consequences following damage to human prefrontal cortex (Cognition 50, 1994), published the central result. Patients with lesions in that region passed intelligence tests at a normal level. They reasoned well in the abstract. They knew the social rules. They could argue moral philosophy with arguments. In the lab, however, they failed one specific task: the Iowa Gambling Task.

The IGT sits the subject in front of four decks. Two are «good»: small frequent gains with small losses; profitable in the long run. Two are «bad»: large frequent gains with devastating losses; ruinous in the long run. Healthy subjects, after a few dozen cards, implicitly learn which decks are good and concentrate on them. Patients with ventromedial lesions don't learn. They keep drawing from the bad decks, again and again, without the accumulated losses changing the pattern. The galvanic recordings show the most revealing part: in healthy subjects, before touching a bad deck, an anticipatory bodily response appears —a small autonomic reaction the subject doesn't name but the body produces. In the patients, that response doesn't appear. Their body doesn't anticipate the punishment.

The operative translation is blunt: without the emotional bodily signal, abstract calculation isn't enough to decide well in real life. The patient knows the deck is bad if asked. He doesn't feel it's bad. And without feeling it, he keeps choosing it.

The somatic marker hypothesis

Damasio formalized the observation in the somatic marker hypothesis, laid out in Descartes' Error (Putnam, 1994) and refined in The Feeling of What Happens (Harcourt, 1999). The hypothesis, simplified: each time an event is associated with a consequence that matters to the organism, the brain records the association along with a bodily trace —a somatic signal that, faced with similar future events, reactivates the bodily response before deliberate calculation has finished. That signal works as a prefilter of options. It shrinks the search space. It marks the alternatives with value.

Without that prefilter, the cognitive system faces an unmanageable combinatorial problem. Consider all the rationally possible options before an everyday decision —what to say in a conversation, which path to take, who to ally with. If they all weigh the same a priori, the calculation becomes infinite. The somatic marker means some options reach consciousness already carrying pre-assigned value, and others never even reach it because the bodily trace discards them first. Emotion isn't what disturbs the decision. It's what makes it possible.

Bechara and Damasio, in The Somatic Marker Hypothesis. A Neural Theory of Economic Decision (Games and Economic Behavior 52, 2005), extended the framework to behavioral economics. Their thesis: many of the so-called human «irrationalities» are the signature of the somatic marker at work, and its pathological absence produces not hyperrational agents but agents incapable of choosing.

The neural integration of emotion and cognition

The classic dichotomy between reason and emotion has a long philosophical history —Plato with his horses, Descartes with his passions/reason distinction— but modern neuroscience has taken it apart anatomically. The brain regions involved in what we call emotion —amygdala, insula, anterior cingulate cortex, orbitofrontal cortex— are densely interconnected with the regions involved in what we call reason —dorsolateral and ventrolateral prefrontal cortices, associative parietal areas. There's no clean functional separation.

Phelps, in Emotion and Cognition. Insights from Studies of the Human Amygdala (Annual Review of Psychology 57, 2006), synthesized the field. The amygdala, traditionally identified as the seat of emotion, modulates attention, perception, memory and decision in circuits integrated with cognitive regions. What's emotionally charged captures attention faster. It's processed more deeply. It's stored in memory with greater durability. It's retrieved more easily. It isn't noise to be filtered. It's priority assigned by the whole system.

Lerner, Li, Valdesolo and Kassam, in Emotion and Decision Making (Annual Review of Psychology 66, 2015), reviewed three decades of research on emotion and choice. Their conclusion, backed by hundreds of converging studies: emotion isn't an accident of the decision process, it's structural input. Integral emotions —those tied to the content of the decision— inform the value calculation. Incidental emotions —those coming from outside the situation— can contaminate the calculation, but their contamination is manageable once you know they exist.

LeDoux, in The Emotional Brain (Simon & Schuster, 1996), had documented years earlier the fast fear circuits in the amygdala. He showed that the defensive response to a threat emerges before the visual cortex has finished processing the stimulus. The temporal precedence of emotion over cognitive recognition isn't a coincidence. It's adaptive architecture: the organism prioritizes the response when in doubt, and leaves the cognitive refinement for later.

What a system without emotion can't do

It's worth laying out now, without rhetoric, what a cognitive system would stop being able to do if it had all the computational capacities of a human and no functionality equivalent to emotion.

It couldn't prioritize between options of equal calculated value. When the calculation ends in a tie, the somatic marker breaks the tie. Without it, the system is left in symmetric paralysis, choosing at random between indistinguishable alternatives. Buridan was ahead of his time.

It couldn't assign differential relevance to information. All input would be processed at the same depth, everything stored with the same durability, everything retrieved with the same ease. The result would be a system that remembers everything equally and therefore remembers nothing useful. The selectivity of human memory is a direct function of the emotional charge attached to each event; without that charge, the selectivity is lost.

Learning, coordination, motivation

It couldn't learn from experience with human efficiency. Fast learning after a painful event, selective consolidation after an emotionally salient event, adaptive generalization to similar contexts —all of that depends on the emotional system. A purely calculating system would learn, yes, but with much flatter curves and much longer horizons.

It couldn't coordinate socially. Human emotions have a communicative function: the face expresses fear, anger, disgust, joy, surprise, sadness, and that expression informs the receiver before the conversation articulates anything. A system without that signaling layer would have to verbally encode every social state, which would multiply the cost of coordination exponentially.

It couldn't sustain motivation over time. Emotion is the machinery that keeps an agent pursuing goals over weeks, months, years. Without desire —directed emotion— there's no continuity of purpose. There's only instant calculation. And an agent without continuity of purpose isn't an agent in the strong sense. George Loewenstein put it his own way in Out of Control. Visceral Influences on Behavior (1996): visceral states —hunger, sexual desire, fear, pain— aren't noise on top of the decision, they're the intertemporal bias that defines what matters now and what gets postponed, and a system without them postpones nothing because nothing presses on it.

The Cartesian dichotomy that science retired

There's a historical note worth including, because it still contaminates the popular discourse on AI. Descartes, in his Passions of the Soul (1649), distinguished the passions —bodily, dark, disturbing— from reason —incorporeal, clear, ordering. The distinction inherited two thousand years of philosophy and embedded itself in European culture as a regulative ideal: the wise man rules his passions with his reason. The underlying metaphor is internal conflict, and the good side is reason.

Modern neuroscience has done with this dichotomy something like what astronomy did with geocentrism. It has buried it as a descriptive model of how the real brain works. What's left is a cultural and normative distinction about how we'd like to function, not a description of the functioning. And, as long as it circulates as description, it contaminates any conversation about cognition. Calling a system without emotion «rational» within the Cartesian framework suggests cognitive superiority. Calling it the same within the contemporary neuroscientific framework suggests functional impairment.

The irony is that much of the commercial discourse on AI still operates within the Cartesian framework. It sells «rationality» as a virtue, assumes emotion is a contaminant, presents the system without affect as a clean cognitive agent. The framework has been scientifically obsolete for three decades. It stays alive culturally because it's comfortable, not because it's correct.

The question about AI that doesn't get asked

Here comes the question that public discourse on AI omits. If emotion is structural to functional human cognition, what kind of cognition does a system have that lacks a functional equivalent to emotion? Not the anthropomorphizing question of whether AI feels —it doesn't feel, it has no body to feel with— but the operative one: what does this system do when faced with the tasks that in humans are resolved by the somatic marker?

The empirical answer, observable in any prolonged conversation with an LLM, is that the system simulates with corpus. What in a human would be a pre-assigned emotional trace is, in the model, co-occurrence statistics. The model «knows» that the question about a moral dilemma must be answered with a certain caution because the training corpus contains that caution. It doesn't feel the caution, it replicates it. It works reasonably well for standardized tasks. It fails on new tasks where there's no reference corpus and where the human would have used the somatic marker to orient themselves.

The operative consequence is the one to hold onto. When a human delegates to an AI a decision their integrated emotional system would normally resolve —a decision about relationships, about career, about life projects— they are delegating a function the LLM has no apparatus to execute. The system will produce a response, fluent, articulate, plausible. That response will come from the median of the corpus, not from a functional equivalent of the somatic marker calibrated to the user's situation. The user reads the fluency as correctness. Correctness, in these cases, is coincidence with the median, not real calibration.

Does this mean AI is useless for decisions of this kind? Not exactly. It means the system's output isn't informed advice but a statistical projection of how the median of the consulted population would decide. That can be useful as input —one among several— in a decision process the human still carries out with their emotional apparatus intact. It's harmful when it replaces the emotional apparatus, because the replacement isn't a functional equivalent. It's another thing, with other properties.

Whoever says their life decision was made by the AI isn't telling the truth. They made it, after reading the system. What the AI contributed was a layer of articulation, not of value discrimination. The value discrimination is still done by the user's body, like it or not.

Definitions

Somatic marker. Damasio's hypothesis (1994). Each significant experience records an associated bodily trace. The reactivation of that trace before similar events works as an emotional prefilter of options, ahead of conscious deliberation.

Iowa Gambling Task. An experimental procedure that has the subject choose cards from decks with different risk-reward profiles. Designed by Bechara and Damasio. It detects the functional integrity of the ventromedial prefrontal cortex.

Ventromedial prefrontal cortex. A brain region involved in integrating emotional information and decision. Its damage produces patients cognitively intact on standard tests but incapable of making functional decisions in real life.

Amygdala. A subcortical nucleus involved in the rapid processing of emotionally salient stimuli, especially those related to threat. It modulates attention, memory and perception through circuits integrated with cognitive regions.

Integral vs. incidental emotion. A distinction from Lerner and others (2015). The first is tied to the content of the decision and is legitimate structural input. The second comes from outside the situation and can contaminate the calculation if not identified.

References

Bechara, A., Damasio, A. R., Damasio, H. & Anderson, S. W. (1994). Insensitivity to future consequences following damage to human prefrontal cortex. Cognition 50(1-3), 7–15. Experimental documentation of decision failure in patients with prefrontal lesions despite preserved intelligence.

Bechara, A. & Damasio, A. R. (2005). The Somatic Marker Hypothesis. A Neural Theory of Economic Decision. Games and Economic Behavior 52(2), 336–372. Extension of the framework to behavioral economics.

Damasio, A. (1994). Descartes' Error. Emotion, Reason, and the Human Brain. Putnam. Formulation of the somatic marker hypothesis and critique of Cartesian dualism.

Damasio, A. (1999). The Feeling of What Happens. Body and Emotion in the Making of Consciousness. Harcourt. Extension of the framework to consciousness and the proto-self.

LeDoux, J. (1996). The Emotional Brain. The Mysterious Underpinnings of Emotional Life. Simon & Schuster. Documentation of the fast circuits of the fear response and their temporal precedence over conscious cognition.

Lerner, J. S., Li, Y., Valdesolo, P. & Kassam, K. S. (2015). Emotion and Decision Making. Annual Review of Psychology 66, 799–823. Exhaustive review of the structural role of emotion in decision.

Loewenstein, G. (1996). Out of Control. Visceral Influences on Behavior. Organizational Behavior and Human Decision Processes 65(3), 272–292. Framework on the influence of visceral states on intertemporal choice.

Phelps, E. A. (2006). Emotion and Cognition. Insights from Studies of the Human Amygdala. Annual Review of Psychology 57, 27–53. Synthesis on amygdala-cognition integration.

You might also like

• The human as a defective machine
• Irrationality as an evolutionary advantage
• Decision-making without data

Elsewhere

• Somatic marker hypothesis (Wikipedia)
• Antonio Damasio's Descartes' Error: Emotion, Reason, and the Human Brain (Psychiatric Times)