Glossary
Reward prediction error
What is reward prediction error?
Reward prediction error is the difference between the reward you expect and the reward you get. It helps explain how experience changes expectations.
How it works
In classic primate research, Schultz and colleagues linked dopamine activity to unexpected rewards and changes in reward expectations. These findings helped connect brain activity with reinforcement-learning models, including temporal-difference learning. Some dopamine responses resemble these models' prediction-error signals; dopamine neurons do not all send the same signal.
The classic dopamine response pattern
In the primate recordings described by Schultz and colleagues, an unexpected reward was accompanied by a brief increase in firing in the studied dopamine neurons. After an animal learned that a cue predicted the reward, the response shifted toward the cue: the expected reward itself produced little or no additional response. When an expected reward failed to arrive, firing fell below baseline around the time it should have appeared.
These are the classic positive, approximately zero and negative reward-prediction-error patterns. The comparison is with what was expected at that moment, so a response to a predictive cue is compatible with little response to the reward that follows. This account describes the reported neural pattern, not a claim that every dopamine neuron behaves identically.
Example
You expect one candy bar from a vending machine, but two fall out. That is a positive reward prediction error. If the machine keeps dispensing two, you may come to expect the bonus. Getting only one would then be a negative prediction error.
If the reward matches the expectation, there is no difference between expected and received reward in this simple comparison.
Why it matters
The same reward can be better or worse than expected. Prediction-error models make that difference explicit and give researchers a way to test how expectations change through experience.