Abstracts

Probabilistic Reasoning and Its Neural Mechanism in Macaque Monkeys

TIANMING YANG Institute of Neuroscience, Shanghai Institutes for Biological Sciences
Chinese Academy of Sciences

Many things in life are not certain. It is important for the survival of an organism to be able to base its behavior on uncertain or probabilistic information. Probability theory and statistics have provided mathematical solutions, but humans and presumably animals too have been using their “good sense” to carry out probabilistic reasoning way before such knowledge was available. It is a curious question whether their behavior is consistent with theoretical solutions and achieves certain optimality.

We trained macaque monkeys to perform a probabilistic categorization task based on multiple pieces of probabilistic information. Two monkeys were trained to make decisions after looking at 4 shapes randomly chosen with replacement from a set of 10. The shapes were sequentially presented on a computer display along with two eye movement targets in the peripheral visual field. Each shape was assigned with a probability that one of the two eye movement targets was associated with a juice reward. The monkeys indicated their choice by looking at the target of its choice after all shapes were presented and received juice rewards if the choice was correct. After extensive training, the monkeys were able to choose the more likely rewarded target most of the time. Their behavior was consistent with the log likelihood ratio (logLR) test. Furthermore, we found that a group of neurons in a cortical area called lateral intraparietal (LIP) cortex represented logLR during monkeys’ decision process.

Next, we extended our study and allowed the monkeys to look at an endless stream of shapes and decide freely when to make their responses. The shapes and the probabilistic assignments of rewards were similar to the previous task. Allowing monkeys to make their decisions whenever they wanted gave them freedom to see more shapes when the evidence was weak and to commit to a decision sooner when the evidence was strong. We found that their choice accuracy and speed could be explained by a model using sequential probability ratio test, in which one accumulates logLR of individual pieces of evidence toward a preset decision boundary. Again, neuronal activities in LIP reflected the decision-making process.

In conclusion, our studies showed that macaque monkeys have capabilities of complex probabilistic reasoning that is consistent with probability theory. We identified LIP as a potential neural substrate underlying such decision process.