Turing Award Winner Richard Sutton’s New Venture is a Bet Against Today’s AI Models

Richard Sutton, who will share the 2024 Turing Award announced in 2025, is diving into a new project that takes a different approach to AI. Together with Andrew Barto, Sutton is well-known for his groundbreaking work in reinforcement learning, a field that has greatly impacted today’s AI systems.

His new venture, reportedly called Keen Technologies and founded in 2024, aims to shift gears from the widely used large language models (LLMs) and focus instead on reward-based, agent-driven methods.

Turing: Key Details of Sutton’s Venture

Sutton’s launch of Keen Technologies stems from his belief that the current push toward Artificial General Intelligence (AGI) is misguided. While many in the AI field are heavily invested in LLMs, Sutton pushes for a return to the basics of reinforcement learning. His influential work includes co-authoring the essential textbook “Reinforcement Learning: An Introduction,” first published in 1998.

Keen Technologies seems to reflect Sutton’s vision of using reward-based systems, which he argues can create more adaptable and effective AI systems. Moving away from LLMs is a big step; Sutton often points out how these models, despite their strengths, may not fully capture the complexities of human-like intelligence and decision-making.

In his impactful essay “The Bitter Lesson,” published in March 2019, Sutton suggested that general methods relying on computation typically outperform those based on human knowledge. This idea resonates with his current initiative, as he aims to harness the computational strengths of reinforcement learning.

Context: The Evolution of AI Models

Sutton’s impressive career includes about 20 years at the University of Alberta and a significant role at DeepMind (now Google DeepMind). His insights into reinforcement learning have influenced a variety of AI applications, from gaming agents to real-world decision-making systems.

Nowadays, LLMs like OpenAI‘s GPT series and Google’s Bard dominate the AI scene. These models have garnered a lot of attention for their ability to understand and create human-like text.

However, Sutton’s timing is interesting, as the limitations of LLMs are becoming increasingly clear, especially in areas that require reasoning and long-term planning. Critics argue that while LLMs are great at generating text, they often struggle with tasks needing a deeper understanding of context or the ability to learn from past interactions.

Sutton’s focus on reinforcement learning offers a refreshing perspective. By using methods that reward successful actions and outcomes, he believes AI can learn in ways that are more similar to human learning. This could lead to more effective and flexible AI systems, potentially charting a new path in the ongoing quest for AGI.

What’s Next for Sutton and Keen Technologies?

As Keen Technologies prepares for its official launch, many in the AI community are paying close attention. Sutton’s reputation as a trailblazer in reinforcement learning lends credibility to his new venture, but it also raises questions about the future of LLMs. Will Sutton’s focus on reward-based systems catch on, or will LLMs continue to reign supreme? For more information, check out VentureBeat AI.

The results of this endeavor could not only influence Sutton’s career but also impact the wider AI landscape. If Keen Technologies can prove its approach effective, it might shift the entire field in new directions.

Sutton reportedly left Google DeepMind around 2024 to explore this independent venture. This marks a bold move toward independent exploration in AI. By challenging popular AI trends, the implications of his work could reach far beyond academia, shaping industry practices and the next generation of AI technologies.

FAQs

What is the Turing Award?

The Turing Award is a prestigious honor given each year for outstanding contributions to the computing community. It’s often seen as the “Nobel Prize of Computing.”

What is reinforcement learning?

Reinforcement learning is a type of machine learning where an agent learns to make decisions by taking actions in an environment to maximize cumulative rewards.

How do Sutton’s ideas challenge current AI models?

Sutton argues that although current large language models are powerful, they may not effectively capture the complexities of human-like intelligence and decision-making. He advocates for a return to reinforcement learning principles.

What is Keen Technologies’ main focus?

Keen Technologies is focused on creating AI systems based on reward-based, agent-driven methods, stepping away from reliance on large language models.

Why is Sutton’s venture significant?

Sutton’s venture could potentially transform the AI landscape by challenging existing methodologies and advocating for alternative approaches that align more closely with human learning processes. Stay tuned for more updates on Richard Sutton.

Richard Sutton