VANCOUVER — Ilya Sutskever, a leading figure in artificial intelligence and former chief scientist at OpenAI, has made a bold prediction about the future of AI. Speaking at the NeurIPS conference, he foresees that as AI develops advanced reasoning capabilities, it will become increasingly unpredictable.
Accepting the prestigious “Test of Time” award for his groundbreaking 2014 paper with colleagues Oriol Vinyals and Quoc Le, Sutskever reflected on how far AI has come. He noted that the strategy of scaling up data to pre-train AI models, which led to innovations like ChatGPT, is nearing its limits. “While computing power continues to grow, the data isn’t expanding because we have only one internet,” he remarked.
To overcome this challenge, Sutskever suggested that AI could start generating new data or evaluate multiple responses to improve accuracy. However, his most striking prediction was about the emergence of superintelligent, self-aware AI agents capable of human-like reasoning. “The more it reasons, the more unpredictable it becomes,” he said, emphasizing that AI’s future behavior might be beyond our expectations.
He drew parallels with past AI achievements, such as AlphaGo’s surprising moves in defeating Go master Lee Sedol and advanced chess programs outmaneuvering top human players. These examples highlight how AI can operate in ways that even experts cannot foresee.
Sutskever’s insights hint at a transformative era in technology. As AI evolves, it may not only solve complex problems but also challenge our understanding of predictability itself. “AI as we know it will be radically different,” he concluded.
Reference(s):
AI with reasoning power to be less predictable, Ilya Sutskever says
cgtn.com
