Neural correlate of decision confidence and uncertainty

Adam Kepecs1, Naoshige Uchida1,2, Hatim A. Zariwala1,3 & Zachary F. Mainen1,4

Humans and other animals must often make decisions on the basis of imperfect evidence1, 2. Statisticians use measures such as P values to assign degrees of confidence to propositions, but little is known about how the brain computes confidence estimates about decisions. We explored this issue using behavioural analysis and neural recordings in rats in combination with computational modelling. Subjects were trained to perform an odour categorization task that allowed decision confidence to be manipulated by varying the distance of the test stimulus to the category boundary. To understand how confidence could be computed along with the choice itself, using standard models of decision-making3, 4, 5, 6, we defined a simple measure that quantified the quality of the evidence contributing to a particular decision. Here we show that the firing rates of many single neurons in the orbitofrontal cortex match closely to the predictions of confidence models and cannot be readily explained by alternative mechanisms, such as learning stimulus–outcome associations7, 8, 9, 10. Moreover, when tested using a delayed reward version of the task, we found that rats' willingness to wait for rewards increased with confidence, as predicted by the theoretical model. These results indicate that confidence estimates, previously suggested to require 'metacognition'11, 12 and conscious awareness13, 14, are available even in the rodent brain, can be computed with relatively simple operations, and can drive adaptive behaviour. We suggest that confidence estimation may be a fundamental and ubiquitous component of decision-making.

This is a really nice paper by Adam Kepecs et al from CSHL. The gist of the paper is in the following figure. The model (a-d) illustrates the perceptual decision process of determining which category the odorant stimulus (s) belongs to, by comparing the stimulus (si) to the memory of the boundary (bi), each drawn from its respective distribution. This leads to the determination of the perceptual category (si>bi or si less than bi) and the decision confidence (absolute difference between si and bi).  The behavior of the decision confidence construct (d) is matched perfectly with neuronal responses in the OFC (e, g). 

One important concern of the authors' interpretation is whether the OFC neuronal activity signals uncertainty or an error signal. This is nicely addressed in the method section:

The observed selectivity of neural activity for the upcoming outcome might arise if, after executing a choice, extra sensory or memory information enters decision-making circuits and causes the realization that an error occurred even before obtaining feedback. According to this interpretation the negative outcome selective population of OFC neurons would signal error44 instead of uncertainty. In contrast, the highest observed firing rates were associated with near chance level performance and not errors (Fig. 4g, f). To test this more rigorously, we asked whether an ideal observer could obtain better performance than the experimental subject if it could switch choices based on the firing rate after the choice and before feedback is provided. In all but one negative outcome selective neuron (1/133), the highest firing rates (top 5% of trials) were associated with chance level performance (within the 95% confidence interval). Therefore negative outcome selectivity does not imply that OFC neurons are actually able to predict error trials but rather that high firing rates predict near chance level performance consistent with an uncertainty signal.

The discovery of this neural correlate of decision confidence opens many new research questions. How is uncertainty/confidence calculated in the neural circuits? What are the influences of decision confidence on behavioral responses, learning and memory, and attention? No doubt these questions will be further explored by combining elegant experimental designs with a sound theoretical framework. Excellent work!


kaleidoscopik.com said...

I think your html went a little screwy when you tried to use a less than symbol, so you lost some text.

Also, the jpg probably moved because the paper is no longer an Advanced Online Publication.

I'm about to add my two cents over at my blog even though I know nothing about the OFC.

SCLin said...

Thank you for the pointers. I will edit them now.

Anonymous said...

I am a student in a Master's of Education Program and am reading an article about how neuroscience relates to Education. So this post has interesting implications as far as it applies to learning, don't you think?

Moose withthoughtslikemine said...

I am a PhD student examining uncertainty and decision-making in humans. This study is great... I can't think of another study that correlated a statistical measure with behavioural uncertainty. Most just relate the degree if uncertainty (in the presented stimuli) with behaviour...

Alicia said...
This comment has been removed by a blog administrator.