Reasoning AI to be unpredictable, says OpenAI co-founder

As AI advances, its reasoning capabilities could make it less predictable, according to Ilya Sutskever, a prominent AI researcher and co-founder of OpenAI.
OpenAI’s weekly active users have surpassed 400 million, highlighting the growing demand for AI tools.

At the NeurIPS conference in Vancouver, Ilya Sutskever, co-founder of OpenAI, predicted that artificial intelligence will become increasingly unpredictable as its reasoning abilities grow. Speaking to thousands of attendees, Sutskever explained that while advancements in AI have relied on scaling data and computing power, this approach is nearing its limits due to finite resources like the internet.

To overcome these challenges, Sutskever suggested that AI could begin generating its own data or evaluating multiple responses to improve accuracy. He envisions a future where superintelligent machines, capable of reasoning like humans, become a reality. However, this reasoning power could lead to unexpected outcomes, as seen with AlphaGo’s famous move in a 2016 board game match or unpredictable strategies from advanced chess algorithms.

Sutskever emphasised that AI’s evolution will make it ‘radically different’ from what we know today, with deeper understanding and potential self-awareness. Yet, he warned that this reasoning could complicate predictability, as AI analyses millions of options to solve complex problems. This shift, he stated, marks the beginning of a new chapter in AI.