Back to the future: The return of cognitive functionalism

Area of Science:

• Artificial Intelligence
• Cognitive Science
• Machine Learning

Background:

• Current artificial intelligence (AI) research often emphasizes the need for explicit causal models and interpretable explanations in learning systems.
• This perspective aligns with cognitive functionalism, but may oversimplify the relationship between implicit and explicit knowledge representation.

Purpose of the Study:

• To investigate whether neural networks engage in implicit causal model building during the learning process.
• To challenge the assumption that explicit causal models and explanations are prerequisites for advanced AI.

Main Methods:

• Analysis of recent evidence from neural network research.
• Examination of the dissociability of model building from the learning process.

Main Results:

• Neural networks demonstrate evidence of engaging in implicit model building.
• This implicit model-building is intrinsically linked to the learning process and cannot be separated.

Conclusions:

• The findings suggest that implicit causal model building is a fundamental aspect of learning in neural networks.
• This challenges cognitive functionalist views that solely prioritize explicit knowledge representation and explanation in AI.