Gary Marcus: Toward a Hybrid of Deep Learning and Symbolic AI | Lex Fridman Podcast #43

Cognitive Models

Gary Marcus emphasizes the importance of cognitive models in addressing AI's common sense challenges. He argues that current AI systems rely too heavily on statistical approximations, lacking true understanding of everyday phenomena, such as the relationship between a bottle and its cap 1. Marcus believes that developing cognitive models that incorporate both physical and psychological knowledge is crucial for AI to gain common sense 2. He suggests that a committee of wise individuals, including ethicists, should be involved in translating these cognitive models into actionable AI systems 3.

A good solution to AI starts by having real, what I would call cognitive models of what's going on.

--- Gary Marcus

Marcus's insights highlight the need for AI systems to move beyond mere data correlations to achieve a deeper understanding of the world.

   

Challenges & Solutions

The challenges AI faces without common sense are significant, and Gary Marcus proposes solutions to address these issues. He argues that AI must transition from deep learning to deep understanding, incorporating abstract concepts like harm and physical object recognition 4. Marcus acknowledges the complexity of common sense, noting that even traditional AI tools like symbols and logic fall short in fully capturing it 5. He advocates for a hybrid approach, combining deep learning with human-guided active learning to enhance AI's ability to understand and interact with the world 6.

The first thing we have to do is to replace deep learning with deep understanding.

--- Gary Marcus

This approach aims to create AI systems that are not only more intelligent but also more aligned with human values and needs.