Reasoning With Causal Cycles.

Area of Science:

• Cognitive Science
• Computational Linguistics
• Artificial Intelligence

Background:

• Causal graphical models (CGMs) successfully explain category-based judgments but cannot represent causal cycles.
• Reasoning with cyclical causal relationships is a key aspect of human cognition and complex system understanding.

Purpose of the Study:

• To evaluate formalisms for representing causal cycles in human reasoning.
• To compare the performance of Dynamic Bayesian Networks (DBNs), alpha centrality, and chain graphs in modeling cyclic causal reasoning.

Main Methods:

• Introduced and evaluated Dynamic Bayesian Networks (DBNs), chain graphs, and unfolded chain graphs.
• Assessed an existing model of causal cycles (alpha centrality).
• Conducted four experiments where participants reasoned about categories with cyclically related features.

Main Results:

• Experimental data provided evidence against Dynamic Bayesian Networks (DBNs) and alpha centrality.
• The results supported the efficacy of the two types of chain graphs in modeling causal cyclic reasoning.
• Chain graphs represent the equilibrium distribution of dynamic systems, aligning with observed human reasoning.

Conclusions:

• Chain graphs are promising models for human causal reasoning in complex systems with feedback loops.
• This research extends the application of causal graphical models to cyclic causal structures.
• Findings have implications for understanding category-based judgments and other cognitive phenomena involving cyclical causality.