Artificial intelligence is gradually starting to reach the limits of human data, and is expected to unleash more potential in a new "era of experience", with continual learning as one of the cores, an expert has said at the ongoing 2025 Inclusion Conference on the Bund in Shanghai.

The 2024 Turing Award winner Richard Sutton pointed out at the event's opening ceremony that, in the current era of human data, AI is trained to predict words and labels constructed by humans, and is fine-tuned by human experts. The purpose of most machine learning today is to transfer existing knowledge from humans to static, non-learning AI.

"We are gradually starting to reach the limits of human data and realize that we can't generate generally new knowledge with this approach," said Sutton.

"This method is not suitable for continual learning, which is critical to the intelligence's utility."