Dopamine is a famous chemical. It holds a prized position in the annals of popular science as the “reward” drug. Countless articles refer to the latest studies of foods, sex and exercise as boosting dopamine, and, by implication, pleasure. But is this characterization correct? What does dopamine actually do?
In the past two decades, research on the role of dopamine in the brain has proceeded apace. A key finding is that dopamine is not the “reward” chemical, at least not in the way that the media would have us believe.
In the 1990s, a now-classic experiment was carried out by Wolfram Schultz and colleagues. He recorded from brain cells that produce dopamine and deliver it to other brain areas. An increase in the firing of these neurons is a sign that more dopamine is being released in the brain. Schultz switched on a light, and then delivered a tasty drop of juice to the animal. Initially, the dopamine cells responded to the juice, consistent with the “reward” theory. But over time, as the animal began to understand that juice always followed lights, the dopamine response went away, even though the animals continued to gulp the juice down. Then, when the researchers unexpectedly stopped the flow of juice, dopamine decreased. This experiment provided an initial clue as to what dopamine was doing in the brain.
In parallel, computer scientists were building on early work on trial-and-error learning in psychology. A prominent theory, the Rescorla-Wagner rule, proposed that learning should occur when events are unexpected. This makes sense – if the food at your favourite restaurant is much as you expected, there is no point in updating your opinion of the chef. At the time Schultz was doing his experiments, computer scientists were using a variant of Rescola-Wagner learning to develop sophisticated computer programs that could learn, among other things, to play a good game of backgammon. These algorithms relied on computing the difference between what you expect and what you get – an error in prediction, or “prediction error”.
To illustrate the concept of prediction error, let’s imagine you have a decent bottle of wine that has been maturing for some years in the cellar. On New Year’s Eve, you decide the time is right to try the stuff. Initially your expectations (your predictions) are high, but when you open the cork, you notice it has crumbled. This piece of news leads to a negative prediction error, and you lower your expectations accordingly. Nevertheless, you plough on, pour it out and take a sip, and, despite the cork, it’s superb. Now a positive prediction error ensues – the wine is better than expected.
Revising our expectations using prediction errors is a very efficient way to learn, at least in the long run. Almost everything we do in life comes with both a prediction, and an associated prediction error signal. The jolt we feel getting on a stationary escalator is a manifestation of our prediction that the escalator is usually moving. The Schultz studies, and many subsequent experiments, have shown that the dopamine response in the brain closely tracks what one would expect from a prediction error signal. No dopamine is released later on in the experiment, because the animal has learnt to expect the juice.
So is dopamine a reward chemical? In a sense, yes – what we call rewards are often things that are unexpectedly good. Consider bumping into a good friend on the street or getting a pay rise. These unexpected events will lead to positive prediction errors, and increases in dopamine. But if we get what we expect, even if it’s a very pleasurable experience, dopamine probably plays less of a role. Even if that experience involves a special bottle of wine.