Events and Causal Mappings Modeled in Conceptual Spaces

Area of Science:

• Cognitive Science
• Robotics
• Artificial Intelligence
• Causal Reasoning

Background:

• Current robotic systems often focus on event outcomes, neglecting the underlying causal mechanisms.
• Human causal reasoning is uniquely characterized by understanding forces and events, distinct from animal cognition.
• Existing models for robotic implementation lack a robust framework for causal event mapping.

Purpose of the Study:

• To present a novel model of causal relations for robotic implementation, grounded in human cognitive principles.
• To introduce a two-vector event model that incorporates forces, actions, and results.
• To provide guidelines for learning causal mappings in robots, emphasizing geometric approaches.

Main Methods:

• Development of a two-vector model representing events with action forces and result vectors.
• Description of agents and patients as vectors in conceptual spaces.
• Identification of key cognitive processes: causal thinking, action control, and learning by generalization.

Main Results:

• The proposed model integrates the causal aspect (action) with the result of an event.
• It defines event mapping as the core of causal relations, linking actions to results.
• Mathematical properties (monotonicity, continuity, convexity) are proposed to constrain learning of event mappings.

Conclusions:

• A geometrically oriented approach to event mappings is more suitable for robotic implementations than Bayesian models.
• The two-vector model offers a framework for robots to understand and learn causal relations.
• Further research is needed to determine optimal methods for learning these causal event mappings.