Part II · The Neurochemical Temperament Map Draft
Chapter 7
From Molecules to Computational Variables
The most dangerous sentence in popular psychopharmacology is often the shortest one: dopamine is motivation, serotonin is happiness, oxytocin is trust, GABA is calm.
Each sentence contains a trace of truth and a large amount of compression. The compression is the problem. A neurotransmitter system acts in many regions, through multiple receptors, on several timescales, inside networks whose effects depend on context. A personality trait such as extraversion or conscientiousness is itself an abstraction over many behaviors. Trying to connect one directly to the other skips most of the causal chain.
Personality pharmacology needs an intermediate language.
The most useful candidates are computational variables: recurring operations that a nervous system must solve regardless of how psychologists choose to name personality traits. How strongly should an opportunity be valued? How much effort is worth spending? How rapidly should a belief be updated after surprising evidence? How much uncertainty should trigger vigilance? When should an impulse be suppressed? How long should a reward be delayed? How broadly should the mind search for associations? How much confidence should be assigned to an interpretation? How valuable is social belonging? How rapidly should the system switch from exploration to exploitation?
These questions are closer to what neural systems actually control.
A simplified causal chain might look like this:
molecular signaling → circuit computation → learning and valuation → recurrent behavioral policy → measured personality trait
Every arrow is many-to-many. Dopamine can influence several computations. Effort allocation depends on more than dopamine. A trait such as novelty seeking can reflect reward sensitivity, low threat, high arousal, weak inhibition, learned identity, or some combination. The same measured trait can therefore have different mechanisms, and similar mechanisms can produce different traits in different environments.
This is not an inconvenience. It is one of the theory’s central predictions.
Imagine two people who both describe themselves as adventurous. The first experiences novelty as intensely rewarding. The second is not especially reward-driven but experiences little anticipatory fear. Their behavior converges. Give both a drug that increases incentive salience and the first may become markedly more exploratory while the second changes little. A personality label that looked identical at baseline hides different control architectures.
The reverse can also happen. Two people may have similar reward sensitivity but different personalities because one grew up in an environment where exploration was encouraged and the other where it was punished. Biology contributes to a policy, but history shapes how that policy is expressed.
This intermediate level also helps explain why drug effects can seem psychologically broad. A stimulant does not need to contain a separate effect for productivity, confidence, talkativeness, and goal pursuit. If it alters a smaller set of control variables such as effort cost, salience, action vigor, and reinforcement, several high-level behaviors may shift together. Likewise, a sedative that reduces threat weighting and self-monitoring can change sociability, risk taking, and emotional reactivity at once.
The computational language needs discipline, however. Terms such as salience, precision, flexibility, or integration can become scientific-sounding fog if they are not operationalized. A useful variable should be tied to something measurable: a task, a formal model, a reproducible behavioral pattern, a neural response, or ideally several converging indicators.
Reward prediction error is more useful than “motivation chemistry” because it can be modeled. Temporal discounting is more useful than “patience chemistry” because choices over delayed rewards can be quantified. Response inhibition can be tested. Threat generalization can be measured. Confidence calibration can be compared with accuracy. Associative breadth can be probed with divergent-thinking and semantic-network tasks, though those too require careful interpretation.
This is where personality pharmacology can connect with existing traditions rather than reinventing them. The Big Five and HEXACO provide powerful descriptive summaries of individual differences. Reinforcement sensitivity theory organizes approach and avoidance. Cloninger’s temperament model explicitly linked personality dimensions to biological systems, though its mappings were often stronger than later evidence justified. RDoC organizes psychiatric phenomena around functional domains. HiTOP describes psychopathology dimensionally. Predictive-processing frameworks provide language for priors, precision, and updating. The Functional Ensemble of Temperament attempts to connect temperament with interacting neurochemical systems rather than single molecules.
Personality pharmacology can borrow from all of these while making a different wager: that pharmacological perturbation is not merely a treatment tool but a way of identifying the control architecture beneath stable psychological variation.
That wager suggests stronger experiments than ordinary correlation studies. Suppose a genetic variant or receptor measure is weakly associated with a personality score. Interesting, but limited. A more convincing chain would show that the biological variable predicts a specific computational phenotype, that the computational phenotype predicts a stable trait or behavior, and that a targeted perturbation changes both in the expected direction. Stronger still would be evidence that the baseline computational phenotype predicts who responds to the perturbation.
This turns drugs into probes as well as interventions.
The advantage is conceptual clarity. If a proposed biological mechanism fails to predict the intermediate computation, the trait-level story weakens. We do not need to argue about whether dopamine “really means” extraversion. We can ask narrower questions about reward learning, effort, exploration, and action selection.
The cost is that the story becomes more complex. There will be no table in which every transmitter maps neatly onto one personality dimension. The correct map will look more like a network, with overlapping arrows and context-dependent effects.
That is a feature, not a defect. Personality is complex because the control problem it solves is complex. A person must decide what to notice, what to fear, what to pursue, whom to trust, when to wait, when to act, how strongly to feel, how quickly to update, and when to abandon a model. Those decisions are repeated millions of times and gradually look, from the outside, like temperament.
The next chapters examine individual neuromodulatory systems in that spirit. They are not attempts to assign ownership of traits. They are hypotheses about which control operations each system may bias, and how those biases could contribute to recognizable styles of being.