|
A dopamine mechanism for reward maximisation| title | A dopamine mechanism for reward maximisation |
|---|
| start_date | 2024/11/18 |
|---|
| schedule | 15h15 |
|---|
| online | no |
|---|
| location_info | Room B10 |
|---|
| summary | Individual survival and evolutionary selection require biological agents to maximise reward. Conventional economic choice theories define the necessary and sufficient conditions for reward maximisation, and neuronal signals of decision variables provide mechanistic explanations for choosing rewards with the highest value. However, reinforcement learning (RL) by reward prediction errors (RPE) and its implementation by dopamine neurons provides an additional reward-maximising mechanism. The mechanism relies on the recursive nature of RL: RPEs affect predictions that in turn affect RPEs that affect predictions. Electrical and optogenetic self-stimulation experiments demonstrate that monkeys and rodents repeat behaviors that result in dopamine excitation that reflect positive RPEs. The dopamine RPE signal increases the reward prediction via RL, which requires agents to obtain a better reward for maintaining a sizeable RPE. This recursive reward-RPE-prediction iteration drives agents toward better and better rewards. The mechanism improves daily life and benefits evolutionary selection but may also induce restlessness and greed. Please see the PNAS 2024 review for details. |
|---|
| responsibles | Allen |
|---|
Workflow history| from state (1) | to state | comment | date |
|---|
| submitted | published | | 2024/11/13 13:36 UTC |
| |
|