Temporal Prediction Errors in
a Passive Learning Task Activate Human Striatum
Samuel McClure, Gregory S Berns,
P Read Montague
Introduction
Behavioral studies
have shown that animals track error in predicting reward obtainment. Subsequent
neural studies seem to show that dopaminergic cells track prediction error.
Thus, it would be interesting to use fMRI to see what areas respond to prediction
error.
Task
1. The first two
runs only had control trials in which a light was followed by juice after
six seconds.
2. The third run
consisted mostly of control trials but some experimental trials occured where
their was a dealy of 4 seconds in delivering the juice.
3. To control for
sequnce another session was run identical to this task execept juice was exchanged
for a red light.
Prediction
1. A negative prediction
error should occur when anticipated juice fails to be delivered.
2. A positive prediction
error should occur when unanticipated juice is delivered (after the delay
period).
Results
1. In the control
session (lights) no areas were significantly active.
Positive Prediction Error: Unexpected
versus Expected Juice Delivery
1. Unpredicted delivery of Juice
shows increased right putamen activity.
Negative Prediction Error: Absence
of Expected Juice Delivery versus Periods of No Juice Delivery during Normal
Events.
1. Absence of predicted juice
shows decreased left putamen activity.
2. Region of interest analysis
on the left putamen for the negative prediction error shows difference between
the control and experimental condition.
Region of Interest
Analysis: Nucleus Accumbens
1. Unexpected delivery of juice
caused greater bold response at times points 4,6,and 8 but was not significant.
Unexpected absence of juice only showed a weak increase in response at the
time point 4.
Discussion
1. Using classic conditioning
to design an fMRI task this study was able to show that in the putamen, increases
in the bold signal were linked to a positive prediction error while decreases
were linked to a negative prediction error of a behavioral task.
2. There are two reasons for
why only one area showed changes to prediction error.
a. The task was simple allowing only one way to perform the task.
b. In the whole brain analysis a specific time course was assumed.
3. This task shows changes in the dorsal striatum while other studies with
similar tasks show changes in the ventral striatum. This difference is explained
as a difference in task. This study used a classic conditioning paradigm while
the other studies used an instrumental paradigm.
4. Recent work on the reward system
shows that many different kinds of rewards can affect the system. Thus, it
might be reasonable to argue that the reward system processes rewards free
of the modality of the reward.
5. Further experiments might be
able to draw closer connections between human neural mechanism and the prediction
error hypothesis found in animal studies. Others studies have shown this link.
6. Finally, studies have shown
that decision making tasks can activate the reward system without their being
an apparent reward.
