Abstract representations of events arise from mental errors in learning and memory

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Information Theory

Background:

• Humans excel at identifying abstract associations, but the cognitive mechanisms are unclear.
• Existing theories suggest complex mental processes underlie higher-order learning.
• An alternative perspective posits natural errors in learning and memory are key.

Purpose of the Study:

• To propose and validate a novel model where higher-order associations emerge from errors in learning and memory.
• To explain human expectations and predict behavior in probabilistic tasks using this error-driven model.

Main Methods:

• Derivation of a maximum entropy model based on the free energy principle.
• The model represents internal representations of stimulus transitions.
• Utilizing information theory and Bayesian inference to bridge concepts.

Main Results:

• The model provides a concise analytic form.
• Qualitatively explains how transition network structure impacts human expectations.
• Quantitatively predicts human reaction times in sequential motor tasks.

Conclusions:

• Natural mental errors significantly and predictably influence abstract world representations.
• Findings suggest implications for designing more learnable information sources.
• Highlights the role of imperfections in cognitive processes.