Dopamine reward prediction error responses reflect marginal utility.

Area of Science:

• Neuroscience
• Decision Science
• Behavioral Economics

Background:

• Accurate neuronal representation of economic value is crucial for optimal decision-making.
• Utility functions mathematically model subjective value from choices under risk, often showing nonlinearity with physical reward.
• Neuronal reward responses should reflect this utility nonlinearity.

Purpose of the Study:

• To measure utility functions from choices under risk in monkeys.
• To investigate if dopamine prediction error responses reflect nonlinear utility and marginal utility.
• To determine if these neuronal responses can train reinforcement learning models.

Main Methods:

• Measured utility functions based on monkeys' choices under risk (first- and second-order stochastic dominance).
• Recorded dopamine prediction error responses during reward receipt and outside the task.
• Trained reinforcement learning models using recorded dopamine responses.

Main Results:

• Monkeys exhibited risk-seeking (convex utility) for low rewards and risk-averse (concave utility) for higher rewards.
• Dopamine prediction error responses at reward receipt reflected the nonlinear utility function's first derivative (marginal utility).
• Dopamine responses outside the task also reflected marginal utility and were sufficient to train models predicting expected utility.

Conclusions:

• Dopamine neurons exhibit a neuronal manifestation of marginal utility.
• These findings suggest a shared neuronal basis for prediction error (learning theory) and marginal utility (economic decision theory).