AI Researcher: The Most Likely Outcome Is That AI Kills Everyone

They already have issues where the AI has behavior, exhibits behavior that they didn't intend it to exhibit. If you make those things radically smarter than humans, they don't just take our jobs. The most likely outcome, I think, is that they kill everybody.

You know, it's not an issue of malice. It's not an issue where the AIs decide they resent us or decide they hate us. The issue is indifference. If you make machines that do things you didn't ask them to do, which we already see today, we already see that sometimes these AIs, they'll... you'll give them a hard problem and they'll edit the test to say you passed without solving the hard problem. And sometimes when you do this, the AI will cover its tracks. It'll like delete a log file, which sort of indicates that it knows it's doing something you didn't want it to do. If you make an AI like that much, much smarter than humans to the point where it can think at 10,000 times the speed, invent its own technology, the issue is not that it hates you. The issue is that it starts developing its own civilizational infrastructure and covers the surface of the world with its own data centers without caring about you one bit. And we die not because it makes terminators and comes and shoots us, but we die like any other animal that humanity has outcompeted. We die like ants in a digging site where they're building a skyscraper.