Resource-rational account of sequential effects in human prediction

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Decision Making

Background:

• Sequential effects in human predictions deviate from optimal Bayesian inference.
• Existing explanations attribute these deviations to assumptions of environmental non-stationarity.
• This study proposes an alternative explanation rooted in cognitive constraints.

Purpose of the Study:

• To propose and test a framework of costly inference to explain sub-optimal human predictions.
• To differentiate between two classes of cognitive constraints: precision cost and predictive power.
• To compare model predictions against human behavior in a sequential prediction task.

Main Methods:

• Developed two classes of costly inference models: precision-cost and predictive-power.
• Conducted a prediction task with static binary random stimuli (probabilities 0.05-0.95).
• Analyzed sequential effects, including attractive and repulsive effects, in human predictions.

Main Results:

• Human predictions fluctuated and showed biases toward recent stimuli (attractive effect), consistent with both models.
• Only the precision-cost model accurately reproduced a repulsive effect, where distant past stimuli biased predictions away.
• Experimental results showed systematic modulations in sequential effects.

Conclusions:

• Sub-optimal and fluctuating human predictions arise from rational decision-making under cognitive constraints.
• The precision-cost model provides a better account of observed sequential effects, including repulsive effects.
• Costly inference offers a viable alternative to environmental non-stationarity assumptions for explaining prediction deviations.