Win-Stay, Lose-Sample: a simple sequential algorithm for approximating Bayesian inference.

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Developmental Psychology

Background:

• Human cognition often aligns with Bayesian models, yet exact Bayesian inference is computationally intensive.
• The cognitive mechanisms enabling approximate Bayesian inference in humans remain an active area of research.
• Understanding how individuals update beliefs sequentially based on new evidence is crucial for cognitive modeling.

Purpose of the Study:

• To investigate potential algorithms underlying approximate Bayesian inference in human cognition.
• To test if a sequential "Win-Stay, Lose-Sample" (WSLS) algorithm can model human belief updating.
• To examine the applicability of this algorithm across different age groups and learning scenarios.

Main Methods:

• Developed a Bayesian approximation using the "Win-Stay, Lose-Sample" (WSLS) principle.
• Employed a "mini-microgenetic method" to track belief updates in real-time during causal learning tasks.
• Conducted experiments with adults and preschoolers involving deterministic and stochastic causal inference scenarios.

Main Results:

• Behavioral data from adults and preschoolers on causal learning tasks were consistent with the proposed Bayesian WSLS algorithm.
• The WSLS algorithm effectively approximated Bayesian inference in both deterministic and stochastic environments.
• Evidence suggests that humans may utilize such sequential sampling strategies for efficient belief updating.

Conclusions:

• The "Win-Stay, Lose-Sample" algorithm provides a plausible computational mechanism for approximate Bayesian inference in humans.
• This finding offers a novel perspective on human judgment and decision-making, particularly in causal reasoning.
• The study highlights the potential of simple sequential algorithms to explain complex cognitive behaviors across development.