3 ms·
Tangentially related: There are several popular SF stories about how an AI, given an overly simple goal (e.g. "make paperclips") inevitably concludes that huma
by MarkusWandel 4y ago
Tangentially related: There are several popular SF stories about how an AI, given an overly simple goal (e.g. "make paperclips") inevitably concludes that humanity has to go.
Not quite as scary, but doable today: A text-generating AI training itself on the "success" of its output. What success metric? User engagement. We're not doomed, not exactly. But intelligent discourse may be.
- zirgs 4y agoA hostile nation could use that AI to influence elections in other countries. Its success metrics could be the number of votes for its preferred candidates, number of people who are spreading conspiracy theories that it invented and so on. Would not be surprised if it's happening right now.