The brain does not store decision variables as defined, labeled quantities. There is no register in your head named "discount rate." These variables are theorists' constructs: parameters in mathematical models (reinforcement learning, Bayesian decision theory, drift-diffusion) that predict choices well and turn out to correlate with specific neural signals.
So "how is the definition stored?" has two answers. The definition lives in the model. The quantity lives in the brain as a pattern of activity, a set of synaptic strengths, or a chemical concentration that behaves the way the variable should.
"All" isn't a closed set either. Different frameworks carve things up differently, but a convergent core shows up across economics, reinforcement learning, and neuroscience. That core is what follows.
The structure: four kinds of variables, four storage media
Any decision variable belongs to one of four types:
- State variables: what the situation is right now.
- Option estimates: how good each choice looks.
- Global parameters: knobs that set how you decide (patience, risk tolerance, randomness).
- Learning signals: errors that update everything above after the outcome.
The brain has roughly four ways to hold a quantity:
| Storage medium | Timescale | What it's good for |
|---|---|---|
| Synaptic weights (connection strengths) | Days to lifetime | Learned values, habits, priors, the world model |
| Persistent neural activity (firing patterns held over seconds) | Milliseconds to seconds | Current estimates, working memory, accumulating evidence |
| Neuromodulator levels (dopamine, serotonin, norepinephrine, acetylcholine) | Seconds to hours | Global parameters broadcast to the whole brain at once |
| Body state (hormones, glucose, heart rate, read by insula and hypothalamus) | Minutes to hours | Needs and arousal |
The core logic: slow, learned knowledge is in wiring; the current computation is in activity; the settings are in chemistry; the needs are in the body. Almost every variable below maps onto this.
1. State variables: where am I and what do I need?
| Variable | Definition | When / why it's used | Stored as |
|---|---|---|---|
| Belief state | Your estimate of the current situation, including hidden parts you can't directly see ("is this a negotiation or a friendly chat?") | Always. The same action has different value in different states, so values are looked up per state | Activity in orbitofrontal cortex and hippocampus, which encode a "cognitive map" of task structure |
| Homeostatic state | Hunger, thirst, fatigue, temperature, arousal | Always, invisibly. Need multiplies value: food is worth more when hungry (alliesthesia) | Body signals read by hypothalamus and insula |
| Active goal / task set | What you're currently trying to accomplish | When behavior has to be organized around an objective rather than a cue | Persistent activity in lateral prefrontal cortex; fragile, which is why interruptions derail you |
2. Option variables: what can I do and how good is each?
| Variable | Definition | When / why it's used | Stored as |
|---|---|---|---|
| Consideration set | The options actually represented, a tiny subset of what's possible | Always, and the most underrated variable. You can't choose what never entered the set | Attention plus memory retrieval; action possibilities in parietal and premotor cortex |
| Attributes | The separate features of an option (taste, health, price, status) | Whenever an option has multiple dimensions | Different cortical areas specialize in different features |
| Attribute weights | How much each feature counts | Every multi-attribute choice. Self-control is largely raising the weight on long-term attributes: in dieters who resist junk food, health gets weighted alongside taste | Prefrontal modulation of how strongly each attribute feeds into value |
| Subjective value / utility | The integrated "how much do I want this," on a common scale | Whenever unlike things must be compared | Activity in ventromedial prefrontal and orbitofrontal cortex plus ventral striatum |
| Reference point | The zero against which outcomes count as gains or losses | Always. Value is relative, not absolute | Learned expectations, often a running average of recent outcomes, so it drifts as you adapt |
| Cached action value | The learned worth of doing action a in state s, from past experience | Familiar situations; this is habit | Synaptic strengths between cortex and striatum, tuned by dopamine |
| Simulated value | Value computed fresh by mentally playing out consequences | Novel, high-stakes, or changed situations | Computed on demand by prefrontal-hippocampal simulation (the hippocampus "replays" possible paths) |
| Pavlovian cue value | Automatic attraction or aversion triggered by a cue | Whenever cues are present, whether wanted or not | Amygdala and ventral striatum; cue associations in synapses |
3. Uncertainty variables: how sure am I?
Uncertainty has at least four distinct components, each processed differently.
| Variable | Definition | When / why it's used | Stored as |
|---|---|---|---|
| Risk (expected uncertainty) | Known variability: you know the odds, just not the outcome. A coin flip | Gambles, any choice with known spread | Insula and orbitofrontal signals scaling with variance; acetylcholine is linked to known uncertainty |
| Ambiguity | Unknown odds: you don't know the probabilities at all | Most real life. People avoid ambiguity even more than risk (the Ellsberg paradox) | Lateral orbitofrontal cortex and amygdala |
| Volatility (unexpected uncertainty) | How fast the world itself is changing | Determines how much to trust the past. In a volatile world, recent data should count more | Norepinephrine is strongly implicated; it signals "the rules may have changed" |
| Confidence | Your estimated probability that a choice was correct | During or after a choice: whether to change your mind, seek more information, or bet more | Computed from evidence strength; prefrontal metacognitive areas read it out |
| Risk attitude | Personal parameter for how much spread hurts or helps you | Every uncertain choice. A stable trait, but shifted by state (losses, stress) | A fitted parameter (utility curvature); the neural basis is distributed and not fully settled |
4. Cost variables: what will this take?
| Variable | Definition | When / why it's used | Stored as |
|---|---|---|---|
| Physical effort cost | The subjective price of exertion | Whenever an action requires work | Anterior cingulate cortex weighs effort against reward; dopamine tone sets willingness to work |
| Cognitive control cost | The subjective price of thinking hard or overriding a habit | Whenever deliberation competes with autopilot. Explains why people default to habit even when they know better | Dorsal anterior cingulate evaluates whether control is worth paying for (Shenhav's "expected value of control") |
| Opportunity cost of time | What else you could be getting per unit time, roughly your average reward rate | Determines speed, vigor, and when to quit (foraging: leave the patch when it drops below average) | Strong theoretical case it's carried by tonic (background) dopamine level |
| Loss aversion (λ) | The multiplier on losses relative to gains, typically about 1.5 to 2.5 | Any mixed gamble or anything framed as giving something up | Fitted parameter; associated with amygdala and striatal asymmetries |
5. Time variables: when does it pay off?
| Variable | Definition | When / why it's used | Stored as |
|---|---|---|---|
| Delay | How far in the future an outcome arrives | Every intertemporal choice | Prefrontal and hippocampal timing signals |
| Discount rate (k) | How fast value decays with delay. Humans are hyperbolic: steep near-term, shallow long-term | Saving, studying, health, procrastination. One of the best behavioral predictors of many life outcomes | A trait-like fitted parameter; serotonin is linked to patience (blocking serotonin makes animals wait less) |
| Future vividness | How concretely the future outcome is represented | Imagining the future vividly reduces discounting, because simulated outcomes get more weight | Hippocampal episodic simulation |
6. Social variables: what about other people?
| Variable | Definition | When / why it's used | Stored as |
|---|---|---|---|
| Others' payoffs / inequity aversion | Weight on fairness and others' outcomes, separate from your own | Splitting, trading, cooperating. People reject free money in the Ultimatum Game if the split is unfair | Insula signals unfairness; fitted as an "envy/guilt" parameter |
| Others' beliefs and intentions | Your model of what others think and will do | Any strategic interaction | Temporoparietal junction and dorsomedial prefrontal cortex (theory of mind) |
| Reputation / norms | How a choice will be seen and whether it fits group rules | When observed, or when identity is at stake | Learned norms in semantic memory; social value integrated into the same value signal |
7. Selection parameters: how do I commit?
These come mostly from drift-diffusion and reinforcement-learning models.
| Variable | Definition | When / why it's used | Stored as |
|---|---|---|---|
| Drift rate | Strength of evidence per moment favoring one option, roughly the value difference | Every choice. It sets speed and accuracy | Moment-to-moment activity in parietal and prefrontal accumulators |
| Attention weighting | Extra weight given to whichever option you're currently attending | Constantly. The option you look at gains evidence faster | Gaze and attention modulating the accumulation |
| Decision threshold | How much evidence before you commit; the speed-accuracy dial | Raised under conflict, high stakes, or caution; lowered under time pressure | Basal ganglia circuits; the subthalamic nucleus raises the threshold when options conflict |
| Starting bias / prior | Where accumulation starts, a head start for an expected or default option | Defaults, status quo bias, base rates | Prior expectations in synaptic weights, set into the starting activity |
| Inverse temperature (β) | How deterministically you pick the best-valued option versus sampling others | The exploration-exploitation tradeoff | Norepinephrine implicated in shifting between exploit and explore modes; frontopolar cortex in deliberate exploration |
| Perseveration | Tendency to repeat your last choice regardless of outcome | Always present; a pure inertia term | Habit-like synaptic traces |
8. Learning signals: update everything
| Variable | Definition | When / why it's used | Stored as |
|---|---|---|---|
| Reward prediction error | Outcome minus expectation: better than expected (+), worse (−), as expected (0) | After every outcome. The engine of value learning. Fully expected rewards teach nothing | Phasic (burst) dopamine firing, one of the best-established findings in neuroscience |
| State prediction error | "That's not what I thought would happen," regardless of reward | Updating the world model itself | Lateral prefrontal and parietal signals |
| Learning rate (α) | How much each error changes your estimates | Set by volatility: high in changing worlds, low in stable ones | Fitted parameter; modulated by norepinephrine and uncertainty signals |
| Eligibility trace / credit assignment | Which past actions get credit or blame for an outcome | When outcomes are delayed (did the win come from the move just now or three moves ago?) | Synaptic "tags" that decay over time and await the dopamine signal |
9. Control variables: who's in charge?
| Variable | Definition | When / why it's used | Stored as |
|---|---|---|---|
| Controller reliability | How trustworthy the habit system versus the simulation system currently is | Arbitrates which system drives the choice | Prediction-error-based reliability estimates in lateral and frontopolar prefrontal cortex |
| Conflict | Degree to which options or systems disagree | Triggers slowing down and recruiting control | Anterior cingulate cortex |
| Stress / arousal | Global state shifting the system toward fast, habitual, threat-sensitive modes | Under pressure, fatigue, threat | Cortisol and norepinephrine; high levels degrade prefrontal function |
What ties it together
Parameters are what make people different. Two people with the same information and options choose differently because their discount rate, loss aversion, risk attitude, inverse temperature, and attribute weights differ. Personality, in decision terms, is largely a parameter set.
Parameters are partly trait, partly state. Your discount rate has a stable baseline but shifts with sleep, stress, mood, and blood sugar. That's why the same person makes different decisions at 1 a.m. than at 10 a.m.
Evidence quality varies a lot. Dopamine as prediction error is about as solid as neuroscience gets. Tonic dopamine as opportunity cost, serotonin as patience, and norepinephrine as volatility are well-supported but simplified. Many others are model fits whose neural home is a correlation, not a confirmed mechanism.
Back to the project
Artificial Person. Where these variables became dials in code.