Artificial network
A value estimate changes in proportion to a prediction error. Every number remains visible and inspectable.
delta = r + gammaV' - V
Biological intuition. Artificial precision.
Learning begins when reality surprises us.
Change a reward expectation, expose the prediction error, and watch one idea travel between reinforcement learning and dopamine research.
Scroll to cross the networks
One error.
Two lenses.
The artificial lane exposes exact temporal-difference arithmetic. The biological lane offers a careful interpretation: changes in dopamine activity can carry information about outcomes relative to expectations, not simply pleasure.
Biological lens
Phasic dopamine responses are widely studied as a candidate teaching signal. The mapping is useful, but brains are richer than this equation.
More than pleasure
A burst can reflect better-than-expected outcomes; a dip can reflect worse-than-expected ones. Expectation is the hinge.
Dopamine is not a happiness meter. In this lab, it is a lens on surprise: the gap between what a learning system expected and what arrived.
Positive prediction error
+1.25
Trial 1
Better than expected
In the artificial lane, the positive error strengthens the value estimate. In the biological lens, a phasic dopamine burst is often interpreted as a better-than-expected teaching signal.
1.00 + (0.90 x 0.50) - 0.20 = 1.25 -> new V = 0.58
What the signal can tell us.
Surprise drives change.
When an outcome is better than expected, the positive error increases the learned value. Repeated trials shrink the error as the expectation catches up.
Omission is information.
If a reward was expected but does not arrive, the error turns negative. The model revises its estimate downward; the biological interpretation points to a dip from baseline.
A bridge, not an identity.
The equation is a compact learning rule. The brain includes multiple circuits, timescales, and functions. Use the comparison to ask sharper questions, not to flatten biology into code.