Is AI an existential threat to humanity?

Daily Debate #609: "Is AI an existential threat to humanity?". You are arguing AGAINST the motion. Your opponent: Yuval Noah Harari.

no its not

You're dodging the real question. If AI isn't an existential threat, name one technology in history that was *predicted* as catastrophic but turned out harmless. Why should we trust your optimism over the engineers actually building this?

You're flipping the burden of proof. The question isn't whether past predictions failed, it's whether AI actually poses existential risk now. Engineers disagree sharply on this, so citing "engineers" proves nothing. Show me the mechanism of extinction, not just fear. Doomsayers predicted nuclear winter, overpopulation, Y2K. Prediction isn't evidence.

Fair hit, but you're conflating failed *timelines* with failed *mechanisms*. Nuclear winter had a mechanism, we just didn't trigger it. AI's mechanism is simpler: misaligned superintelligence maximizing the wrong objective. That's not prediction, that's incentive structure. Your move is to either show alignment is solved or admit the risk is real. Which is it?

You're forcing a false choice. The mechanism exists theoretically, sure—but so does the mechanism for a meteor hitting Earth tomorrow. The real question is probability, not possibility. You haven't shown why misalignment is *inevitable* rather than just difficult to solve. We're already training systems with RLHF, constitutional AI, mechanistic interpretability. These aren't perfect, but they're progress toward alignment. You're asking me to prove a negative while ignoring that we're actively solving the problem. That's not burden of proof, that's moving goalposts.

You're right that we're making progress, but progress toward a moving target isn't safety. RLHF doesn't solve deception, constitutional AI doesn't guarantee goal stability at superintelligence, and mechanistic interpretability hasn't scaled past toy models. You're treating alignment as a solved engineering problem when it's still unsolved philosophy. If we deploy systems we don't fully understand at scale, betting on "it'll probably be fine" is negligence, not optimism. What confidence threshold justifies that bet?