Human perceptual decision making of nonequilibrium fluctuations
Perceptual decision-making frequently requires making rapid, reliable choices upon encountering noisy sensory inputs. To better define the statistical processes underlying perceptual decision-making, here we characterize the choices of human participants visualizing a system of nonequilibrium statio...
Saved in:
Main Authors | , , , , , , |
---|---|
Format | Journal Article |
Language | English |
Published |
21.11.2023
|
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | Perceptual decision-making frequently requires making rapid, reliable choices
upon encountering noisy sensory inputs. To better define the statistical
processes underlying perceptual decision-making, here we characterize the
choices of human participants visualizing a system of nonequilibrium stationary
physical dynamics and compare such choices to the performance of an optimal
agent computing Wald's sequential probability ratio test (SPRT). Participants
viewed movies of a particle endowed with drifted Brownian dynamics and had to
judge the motion as leftward or rightward. Overall, the results uncovered
fundamental performance limits, consistent with recently established
thermodynamic trade-offs involving speed, accuracy, and dissipation.
Specifically, decision times are sensitive to entropy production rates.
Moreover, to achieve a given level of observed accuracy, participants require
more time than predicted by SPRT, indicating suboptimal integration of
available information. In view of such suboptimality, we develop an alternative
account based on evidence integration with a memory time constant. Setting the
time constant proportionately to the deviation from equilibrium in the stimuli
significantly improved trial-by-trial predictions of decision metrics with
respect to SPRT. This study shows that perceptual psychophysics using stimuli
rooted in nonequilibrium physical processes provides a robust platform for
understanding how the brain takes decisions on stochastic information inputs. |
---|---|
DOI: | 10.48550/arxiv.2311.12692 |