Learning a theory of causality.

Area of Science:

• Cognitive Science
• Developmental Psychology
• Computational Neuroscience

Background:

• Early abstract knowledge acquisition is often attributed to innate predispositions.
• Understanding the learning mechanisms behind abstract concepts like causality is crucial for cognitive development theories.

Purpose of the Study:

• To investigate the learning speeds of abstract versus specific knowledge within a Bayesian framework.
• To explore the role of innate structure versus learning mechanisms in acquiring causal knowledge.
• To determine if causal knowledge can be learned solely from event co-occurrence.

Main Methods:

• Utilized a Bayesian framework to model the learning of abstract and specific knowledge.
• Employed a logical language for relational theories to represent causal Bayes nets and alternatives.
• Simulated simultaneous inductive learning of an abstract theory of causality and specific causal models.

Main Results:

• The abstract theory of causality was learned relatively quickly, often preceding the learning of specific causal theories—termed the 'blessing of abstraction'.
• Auxiliary evidence, particularly from perceptual input analyzers, significantly aided the bootstrapping of abstract knowledge.
• Computational models suggest efficient causal knowledge acquisition relies on learning mechanisms and perceptual support rather than solely innate abstract causality.

Conclusions:

• The 'blessing of abstraction' facilitates rapid learning of domain-general causal theories.
• Perceptual supports and robust inductive learning mechanisms are key to efficient causal knowledge acquisition.
• Findings suggest a shift from emphasizing innate abstract knowledge to powerful learning architectures in cognitive development.