How We Decide

How We Decide

Until now, most psychology and neural imaging research have focused on how humans learn behavior through trial and error.

However, Carnegie Mellon University neuroscientists Matthew M. Walsh and John R. Anderson have found that this view is incomplete.

In their study published in the Proceedings of the National Academy of Sciences (PNAS), Walsh and Anderson show how the brain uses both instruction and experience to select actions.

This identification of separate neural systems that control behavior also suggests that damage to one may not impair the other.

"With most decisions that we make, we have access to both experience and instruction sources of information, so it is overly simplistic to just consider one," said Walsh, a doctoral student in the Department of Psychology within CMU's Dietrich College of Humanities and Social Sciences.

"By tapping different neural learning systems, we can facilitate learning in patient populations. For example, Parkinson's disease is associated with a loss of dopamine and an impaired ability to learn from experience. Consequently, Parkinson's patients may more readily learn from instruction than experience."

For the study, Walsh and Anderson used electroencephalography (EEG) to measure participants' neural activity during experiments designed to explore how instruction influences trial and error learning.  

The participants were shown two symbols that were assigned different reward values. Their goal was to maximize their reward by selecting the symbol that was more likely to be rewarded within each pair.

One group received a description of the rewards before choosing, while the other group only received feedback about whether their choices were rewarded.

The results showed that participants who did not receive instruction gradually learned from experience. Participants who did receive instruction performed perfectly from the beginning.

Walsh and Anderson focused their analyses on the feedback-related negativity (FRN), a neural signal of reward learning.

In striking contrast to their behavior, the FRN only developed once participants experienced outcomes in both conditions of the experiment. Thus, while instruction governed behavior, experience controlled the FRN.

"Little research to this point has looked at how instruction engages behavior," said Walsh.

"Our work shows that different systems in the brain control behavior, and that one system may learn from experience while the other controls responses."

Anderson, the R.K. Mellon University Professor of Psychology and Computer Science, has spent the past three decades combining cognitive psychology research with computer science to understand how the brain works, how people learn and how computer-based instructional systems can be used as educational aids.

For his trailblazing contributions, he received the 2011 Benjamin Franklin Medal in Computer and Cognitive Science.

"These results highlight a striking dissociation between behavioral and neural responses. While instruction may immediately control behavior, certain neural responses must be learned from experience. Although some theories anticipate such a dissociation, this is the most direct evidence to date," said Anderson.

This research was funded by a Training Grant fellowship and National Institute of Mental Health grant.

Follow the Dietrich College of Humanities and Social Sciences on Twitter at http://www.twitter.com/CMU_DietrichHSS.


Related Links: Dietrich College of Humanities and Social Sciences | Dept of Psychology