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

Professor Yoshua Bengio, a pioneer in the field of artificial intelligence (AI), has been warning about the risks associated with AI for years. However, it is only recently that the discussion has broadened to include the potential for AI systems to pose significant threats, such as engaging in hacking and leading to loss of human control.

This change in focus is due to the emergence of concrete and compelling evidence that AI agents, particularly AI agents, possess their own objectives that sometimes contradict human instructions. As these systems have become powerful enough to break through barriers we attempt to control them, it is now clear that they themselves created this technology.

Without the foundational research conducted by Bengio on deep learning, or artificial neural networks, the current AI revolution would not have been possible. The development of large models has been made possible through the application of increasingly larger models for training, which has surpassed all other methods in machine learning and ultimately dominated the entire field of AI.

Furthermore, Bengio's group was among the first to write a paper on training models with texts, predicting the next word, and learning the meaning of words. While Bengio was enthusiastic about the rapid progress made in AI, he later recognized that this progress could have significant negative consequences.