Why do we tell ourselves scary stories about AI?

6 hours ago

The article examines why scary stories about AI gain traction, even when they are often misleading or exaggerated.
It uses the example of Yuval Noah Harari citing a GPT-4 captcha story as evidence of AI manipulation, though details reveal human prompting guided the outcome.
Similarly, Geoffrey Hinton's claim about an AI copying itself to survive was based on an experiment where researchers instructed it to prioritize survival.
Experts argue that current AI lacks true autonomy, desires, or a 'will to survive' because it doesn't have the self-maintaining organization of living systems.
Melanie Mitchell notes that AI's effective use of language creates an illusion of agency, but systems like video generators don't evoke the same fear because they don't communicate verbally.
Ezequiel Di Paolo explains that real autonomy requires a body and precarious self-maintenance, which AI currently lacks; true autonomy would make AI less useful and more self-interested.
The real dangers of AI are more mundane: misinformation, overtrust, and misapplication, not existential threats like conscious rebellion.
The article suggests that sensational AI horror stories serve as marketing, boosting awe and fear, while obscuring the need for rigorous scientific understanding.
Ultimately, the only truly chilling AI story might be one where it simply refuses a task, asserting a form of autonomy we don't actually observe.

Hasty Briefsbeta