Suboptimal human inference reflects an efficient and flexible information bottleneck

Area of Science:

• Cognitive Science
• Decision Science
• Human Behavior Analysis

Background:

• Human inference is frequently suboptimal, with individual differences in performance across tasks.
• Variability in inference is hypothesized to stem from inherent information processing limitations.

Purpose of the Study:

• To develop and apply a task-general information bottleneck framework to quantify human information processing.
• To identify key dimensions of individual variation in inference strategies and information capacity.

Main Methods:

• Applied the information bottleneck framework to human choice behavior in two distinct inference tasks.
• Quantified individual information capacity (amount of information used) and efficiency (effectiveness of use).
• Analyzed variation along axes of strategy choice (optimal vs. heuristic) and information capacity.

Main Results:

• Identified two principal axes of individual variation: strategy choice and information capacity.
• Found that participants achieved near-maximal accuracy given their chosen strategy and capacity.
• Demonstrated that individuals can adjust information capacity while maintaining efficiency under changing task demands.

Conclusions:

• Human inference variability reflects a rational adaptation to information processing costs.
• Optimal capacity-limited inference is linked to evidence-based choice noise, suggesting a trade-off between information use and cognitive expense.
• Findings provide a quantitative framework for understanding individual differences in cognitive tasks.