Causal generative models are just a start.

Area of Science:

• Cognitive Science
• Artificial Intelligence
• Computer Vision

Background:

• Current models of human reasoning, particularly in vision, overemphasize image synthesis theories.
• World knowledge in vision likely involves a complex integration of geometric, physical, and other information, not solely causal image production models.

Purpose of the Study:

• To challenge the prevailing focus on image synthesis in understanding human reasoning.
• To highlight the broader scope of knowledge and inference mechanisms involved in human cognition.

Main Methods:

• Conceptual analysis of existing theories in cognitive science and artificial intelligence.
• Critique of causal generative models in the context of physical and intuitive psychological reasoning.

Main Results:

• Human reasoning is more multifaceted than current synthesis-centric models suggest.
• Physical reasoning can operate via constraint satisfaction rather than complex physics engines.
• Intuitive psychology often bypasses detailed causal generative models for inference.

Conclusions:

• Theories of human reasoning should encompass a wider range of knowledge types beyond causal synthesis.
• The mechanisms enabling reliable inference under incomplete information remain an open and significant research question.