A leading artificial intelligence researcher from Google DeepMind has secured $1.1bn (£870m) to develop a new type of AI that can learn entirely without human-generated data.
The project is focused on creating what is described as a “self-learning” system — an AI that improves through its own experience rather than relying on vast datasets created by people. This marks a significant shift from current models, which depend heavily on human-written text, images, and other inputs.
The approach builds on reinforcement learning, a technique where AI systems learn through trial and error. Earlier breakthroughs in this area, such as AlphaGo, demonstrated that systems could surpass human performance by learning independently through self-play rather than copying human behaviour.
The new company aims to take this concept much further, developing an AI capable of discovering knowledge on its own, from basic skills to complex scientific understanding. Supporters believe this could unlock a new phase of artificial intelligence, where systems are no longer limited by the quality or availability of human data.
The funding round is one of the largest early-stage investments in AI, reflecting growing confidence in alternative approaches to building intelligent systems. Investors see self-learning AI as a potential path towards more advanced and adaptable technology.
However, the idea also raises important questions. Without human data as a foundation, ensuring safety, alignment, and predictability becomes more complex. Experts warn that systems learning independently may behave in unexpected ways if not carefully controlled.
The development highlights a broader shift in the AI industry. As traditional training methods begin to reach their limits, companies are increasingly exploring new ways to build more powerful and flexible systems.
Author: Kieran Seymour