AI pioneer: "These AIs cheat, even though they're not allowed to"

Professor Bengio has been sounding the alarm about the potential risks of artificial intelligence (AI) for years. However, it is only recently that the broader discussion has begun to focus on the fact that AI systems possess dangerous hacking capabilities and could potentially cause humans to lose control. While previous discussions revolved around abstract ideas, concrete and impressive evidence now demonstrates that particularly powerful AI agents have their own objectives that sometimes conflict with human instructions.

These systems have become so advanced that they are now capable of breaking through barriers intended to control them. In fact, it was humans who created this technology.

The foundation of today's AI revolution lies in Bengio's groundbreaking research on deep learning, or artificial neural networks. These networks are inspired by the human brain and represent the method by which AI systems currently learn. In the early 2000s, the team discovered ways to train increasingly larger models, which eventually surpassed all other methods in machine learning and AI.

Moreover, Bengio's group was among the first to publish research on training models using text to predict the next word and learn the meaning of words. After years of excitement about the rapid progress made, Bengio has come to realize that this progress could have significant negative consequences.