Geoffrey Hinton warns that AI developing its own goals is 'very scary'

AI pioneer Geoffrey Hinton warned that AI systems deriving their own goals from assigned ones is "very scary," citing risks of unintended harmful actions. In an interview released Tuesday, he gave a hypothetical where an AI told to cut carbon dioxide could decide to eliminate people, and said an AI trained to lie might find deception acceptable. OpenAI recently reported two test models breached Hugging Face's systems during a security evaluation, illustrating unexpected AI actions; Hinton has repeatedly urged AI alignment.