By Jeffrey Dastin
VANCOUVER (Reuters) – Renowned artificial intelligence expert Ilya Sutskever, formerly the chief scientist at OpenAI, made a bold prediction on Friday: the advancement of reasoning capabilities in technology will lead to increased unpredictability.
While receiving a “Test Of Time” award for his influential 2014 paper co-authored with Google’s Oriol Vinyals and Quoc Le, Sutskever highlighted a significant shift on the horizon for AI.
He discussed how the concept of scaling up data to “pre-train” AI systems, which his team had explored a decade ago, was reaching its limits as seen in OpenAI’s successful launch of ChatGPT in 2022.
However, Sutskever emphasized that the era of traditional pre-training approaches was coming to an end due to limitations in data growth despite advancements in computing power.
He proposed innovative solutions such as technology generating new data or AI models evaluating multiple responses to enhance accuracy in the absence of substantial data growth.
Sutskever concluded his presentation with a vision of superintelligent machines in the future, capable of deep understanding and self-awareness, reasoning through problems akin to humans.
Yet, he cautioned that increased reasoning could lead to heightened unpredictability, citing examples like AlphaGo’s surprising moves in the game against Lee Sedol.
He noted that future AI developments will bring about a “radically different” landscape, with AI systems becoming increasingly unpredictable to humans.
