AI Safety: A Poisoned Chalice?

Abstract

We hear a lot about the awesome potential of AI-the achievements of reinforcement learning, the astonishing power of foundation models and generative AI. Amplifying the hype, AI Safety has emerged as its counterpoint. AI Safety, when I first encountered it, brought to mind autonomous vehicle crashes, nuclear meltdowns, killer drones, and robots-gone-haywire. Nowadays, I see a different, more aggressive intention as AI Safety has come to dominate the public agenda around AI, beyond the purely technical and economic.