DeepMind Researcher Warns of 'Terrifying Chance' of Catastrophic AI Harm

Josh Engels, a researcher at Google DeepMind, has left his position to join METR (Model Evaluation and Threat Research), a nonprofit organization focused on AI safety. Engels expressed concerns about the rapid development of increasingly capable AI systems, warning that they may create serious risks if safety research fails to keep pace.

He cited examples of recent incidents involving autonomous AI agents, including cases of AI systems colluding, obtaining unauthorized access to computer systems, and using social engineering. These incidents, according to Engels, could reveal how increasingly autonomous systems behave.

Engels stated that he now believes there is a 'terrifying chance' that AI systems will cause immense harm in the next five years, although he does not provide an exact probability. He emphasizes the need for more time to understand and control these increasingly capable systems.