Google DeepMind executive says trillion-dollar AI investment bets on "recursive self-improvement"

Odaily News: Jasjeet Sekhon, Chief Strategy Officer at Google DeepMind, stated that one of the key investment theses behind the unprecedented capital influx into the AI industry is betting on future AI systems achieving "Recursive Self-Improvement" (RSI).

Sekhon recently stated at the Agentic AI Summit held at the University of California, Berkeley, that the massive capital currently being poured into the AI industry ultimately needs to be validated by a more advanced capability—namely, AI's ability to autonomously improve itself and create more capable next-generation systems.

He noted that RSI has become a critical component of AI investment logic. Unlike traditional models that rely on manual research, development, and training iterations, recursive self-improvement envisions AI systems that can autonomously optimize algorithms, refine model architectures, and continuously enhance their own capabilities.

Sekhon believes that RSI is replacing AGI (Artificial General Intelligence) as the new core narrative in the AI field. As companies like OpenAI, Google, and Anthropic, along with cloud computing giants, continue to invest enormous sums in computing infrastructure, the market is betting that AI may transition from a "scaling" phase to a "self-enhancement" phase in the future.

However, whether RSI can actually be achieved remains highly contentious. While current AI models already possess capabilities such as code generation, tool invocation, and automated optimization, moving toward true autonomous recursive improvement still faces significant technical, safety, and controllability challenges. (The Information)