Calendrier du 25 juin 2020
PSE Internal Seminar
Du 25/06/2020 de 12:00 à 13:30
ZOOM
MILCENT Carine(PSE)
COIMBRA Nuno(PSE)
*
Behavior seminar
Du 25/06/2020 de 11:00 à 12:00
Online
WYART Valentin (ENS)
ANNULE - Variable by choice or by chance? The surprisingly large contribution of computation noise to human decision-making under uncertainty
Efficient learning and decision-making in uncertain environments constitutes an important challenge for human and machine intelligence. Even for simplest, binary decisions, it requires inferring properties of one’s environment on the basis of imperfect evidence. This cognitive inference process has been studied across a wide range of vastly different paradigms, from categorizing ambiguous stimuli to choosing among risky reward sources. A common, defining feature of human decisions made under uncertainty is their large trial-to-trial variability. Prominent theories have assigned the source of this decision variability to the peripheries of cognitive inference: sensory errors during perceptual decisions, and stochastic choices during economic decisions. During this talk, I will show how the ubiquitous variability of human decisions made under uncertainty is driven by neither of these two sources, but by computation noise during cognitive inference itself. During probabilistic reasoning, the majority of suboptimal errors arise from random noise in otherwise near-optimal computations. During reward-guided learning, the same type of computation noise drives the majority of non-greedy decisions - which existing theories usually describe as the result of a trade-off between choosing a known option and exploring uncertain alternatives. In the brain, computation noise correlates with fluctuations of activity in prefrontal regions implicated in decision-making and cognitive control, and increases during pharmacological manipulation of the locus coeruleus-norepinephrine (LC-NE) system - a neuromodulatory pathway involved in arousal and exploratory behavior. Together, these findings indicate most behavioral variability, rather than reflecting stochastic choices or active exploration, is due to the limited precision of mental computations.