Deciphering the subtle, complex electrical signals emanating from the brain has yielded important insights into the still-mysterious neural mechanisms that underlie behavior. By recording a subject's faint brain signals via arrays of scalp electrodes during cognitive tasks, researchers can extract "event-related potentials" (ERPs) that enable them to pinpoint with great accuracy when the brain reacts to components of the tasks.
One such intriguing ERP signal, called "error-related negativity" (ERN), is associated with activity in a brain area called the anterior cingulate cortex (ACC). The ACC is known to be activated during demanding cognitive tasks, and ERNs are typically more negative after participants make incorrect responses compared to correct choices. Researchers believe that the ERN reflects an abrupt dip in signaling between neurons that depend on the neurotransmitter dopamine. A key question, though, has been exactly what kind of processing ERNs reflect in the ACC. While one theory holds that ERNs reflect a mechanism by which the brain detects errors in decisions, another holds that it reflects the processing of conflicts, of which errors are just one case.
In an article in the August 18, 2005, issue of Neuron, researchers led by Michael J. Frank and colleagues at University of Colorado at Boulder offer new insight into the neural computational function represented by ERNs. They explored the processing underlying ERNs by wiring volunteers with scalp electrodes and measuring ERNs while the subjects were asked to pick between pairs of Japanese characters on a video screen. They used Japanese characters as symbols that would be meaningless to the volunteers and that could be randomly imbued by the researchers as being either a "correct" or "incorrect" choice. When asked to choose one symbol or the other, the correct choice was indicated by a smiley face and an incorrect choice by a red crossout symbol.
Thus, the researchers could precisely explore how ERNs reflected subjects' reactions to error under different experimental circumstances. In their studies, Frank and his colleagues found that "the relative size of the ERN predicts the degree to which participants learn more about the negative, as compared to positive, consequences of their decisions."
However, intriguingly, their detailed studies also found a difference in ERNs in "positive" and "negative" learners. The former are people who perform better at choosing the correct response than avoiding the wrong one, and the latter are those who learn better to avoid incorrect responses. The negative learners, they found, showed larger ERNs, suggesting that "these individuals are more affected by, and therefore learn more from, their errors. This notion makes the strong prediction that the feedback negativity should also be relatively larger in these participants to negative compared with positive feedback, which could potentially reflect the neural mechanism causing them to be more sensitive to their mistakes." The researchers also tested whether ERN might also reflect processing of conflicts and did find some effect.
They concluded that, while they found no overall effect on ERNs of conflict, "an effect of conflict was revealed that depended on the participants learning bias. Positive learners had larger ERNs when faced with high-conflict win/win decisions among two good options than during lose/lose decisions among two bad options, whereas negative learners showed the opposite pattern.
They explained that "In other words, positive reinforcement learners appear to have experienced greater conflict when choosing between two stimuli that were each previously associated with positive (compared with negative) feedback, whereas negative reinforcement learners may have experienced greater conflict when choosing among negative stimuli."
"Thus, we show that the ERN is not only an error detection mechanism, but that its relative magnitude actually predicts the degree to which participants learn from errors and that its indexing of conflict depends on the type of decision faced by the particular learner," concluded Franks and his colleagues.
"These results demonstrate that the ERN predicts the degree to which participants are biased to learn more from their mistakes than their correct choices and clarify the extent to which it indexes decision conflict," they wrote.
"Taken together, our findings provide valuable constraints toward advancing theoretical perspectives on the role of the ERN in reinforcement learning and decision making," concluded Frank and his colleagues. "First, our results are consistent with a large literature pointing to a role for the ERN in error detection but go beyond these studies to demonstrate that the ERN also reflects error correction."