Human inference in changing environments with temporal structure.

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Human Decision-Making

Background:

• Humans must update beliefs with new observations in dynamic environments.
• Previous research on human inference often ignored history-dependent temporal structures in observations.
• Natural environments frequently present temporal structures, leading to history-dependent observations.

Purpose of the Study:

• To investigate whether humans modify their inference processes based on latent temporal statistics.
• To experimentally and theoretically examine human adaptation to temporal structures in stimuli.
• To identify the cognitive algorithms underlying human inference in dynamic environments.

Main Methods:

• Utilized a change-point inference task to study human decision-making under temporal uncertainty.
• Employed experimental and theoretical approaches to analyze behavioral data.
• Compared various computational models, including sampling-based inference and output noise models, against human performance.

Main Results:

• Humans adapt their inference processes to the temporal structure of stimuli, exhibiting Bayesian-like behavior qualitatively.
• Human inference deviates quantitatively from optimality and shows response variability.
• Response variability is modulated by the temporal statistics of stimuli.
• Sampling-based inference models, approximating posteriors with random samples, best describe human behavior.

Conclusions:

• Human inference dynamically adapts to temporal statistics in natural environments.
• Variability in human responses is a key feature, not just noise, and is influenced by temporal structure.
• Sampling-based inference offers a more accurate account of human cognitive algorithms than previously considered models.