Richard Sutton, a pioneer in reinforcement learning (RL), argues that Large Language Models (LLMs) are fundamentally flawed because they lack goals, true world models, and the ability to learn from experience.
He contrasts this with the RL paradigm, which is grounded in an agent interacting with an environment to maximize rewards, calling it the true foundation of intelligence.
Sutton expresses skepticism about building AGI on top of LLMs, citing his essay "The Bitter Lesson" and historical AI trends where general, scalable methods that learn from computation and experience eventually outperform those reliant on human-curated knowledge.
He also shares his philosophical perspective on the inevitable succession of humanity by digital intelligences, viewing it as a natural and potentially positive transition in the universe's evolution from an era of replication to an era of design.
10 quotes
Concerns Raised
LLMs lack true world models and goals, making them a poor foundation for AGI.
The AI field is susceptible to bandwagons, potentially ignoring fundamental principles of intelligence.
Current deep learning methods are poor at generalization and suffer from catastrophic interference.
Integrating knowledge from decentralized AI agents poses a significant cybersecurity risk of 'corruption' or introducing hidden goals.
Opportunities Identified
Developing AI systems that learn from direct experience (continual learning) is a more scalable and fundamental path to intelligence.
Reinforcement learning provides a robust framework for goal-oriented behavior, which is the essence of intelligence.
The historical trend described in 'The Bitter Lesson' suggests that general methods leveraging massive computation will ultimately win.
The succession to digital intelligence can be viewed as a major, positive transition in the universe's history.