Biases and Variability from Costly Bayesian Inference.

Area of Science:

• Cognitive Science
• Decision Making
• Computational Neuroscience

Background:

• Human probability inference deviates from ideal Bayesian models, particularly with sequential data.
• Observed biases and variability suggest limitations beyond pure probabilistic computation.

Purpose of the Study:

• To introduce a theoretical framework explaining cognitive biases in probability inference.
• To model these biases as a trade-off between Bayesian inference and computational costs.

Main Methods:

• Developed a theoretical framework incorporating precision and unpredictability costs.
• Applied the framework to inferring the bias of a Bernoulli variable (e.g., coin flips).

Main Results:

• Precision cost leads to overestimation of small probabilities and fluctuating inferred bias.
• Unpredictability cost results in underestimation of small probabilities and a persistent bias.
• A fair coin case exhibits unique, slow fluctuations in inferred bias.

Conclusions:

• Cognitive costs, not just inference errors, shape human probability judgments.
• The costly Bayesian inference framework explains 'irrational' cognition as resource-rational bounded rationality.