Check out our new paper by Hame Park, AraziAyelet, Bharath Talluri, Marco Celotto, Stefano Panzeri, Alan Stocker & Tobias Donner published in Nature Communications – “Confirmation Bias through Selective Readout of Information Encoded in Human Parietal Cortex”: https://rdcu.be/etlR7.
Here is a summary: People often interpret information selectively, depending on whether that information aligns with their pre-existing beliefs: Consistent evidence has a strong impact on future judgments, while inconsistent evidence tends to be discarded. This is confirmation bias. We reasoned that such selective weighting of evidence could be brought about by two distinct neural mechanisms: (i) selective encoding of incoming evidence in the brain, or (ii) biased utilization of encoded evidence for reasoning and action.
To arbitrate between these scenarios, we asked participants to evaluate sequences of 12 noisy visual evidence samples: small discs with varying angles to the vertical meridian. Each sample was drawn from a hidden source: a probability distribution with constant mean per trial. After viewing half of the samples, participants judged whether the mean of the source distribution was to the left or right from the vertical meridian. After viewing the rest of the samples, they reported a continuous estimation of the source with a joystick. Participants’ final estimation reports were more strongly affected by evidence samples in the second half of the trial that were consistent (compared to inconsistent) with the previous left/right choice: a behavioral signature of confirmation bias. Interestingly, this consistency effect on behavioral evidence weighting was bigger when participants had to report their own categorical judgment of the evidence halfway through the trial, compared to when they instead received an external categorical cue.
We used MEG to measure cortical population dynamics in participants’ brains during the task. The evidence samples were precisely encoded in population activity in visual and parietal cortex, irrespective of their consistency with the previous choice. By contrast, an information-theoretic measure of the use (“readout”) of encoded evidence for the final estimate (“intersection information”) in parietal and visual cortex was bigger for consistent than for inconsistent samples, in line with the selective use scenario.
We conclude that confirmation bias originates from the way in which decision-makers utilize information encoded in the brain, which sheds new light on an important cognitive phenomenon that has occupied scholars for centuries. Because information use is more susceptible to deliberative control, our results imply that confirmation bias may be malleable, contingent on appropriate feedback and incentives.