I think the AI safety community is largely failing this test (with most people failing one but not both bullets). Many people fail the first by arguing for some weaker point, like “AI is becoming increasingly powerful, and powerful things can be dangerous”. And maybe… - ThreadSky

About ThreadSky

daniel-eth.bsky.social • 8 days ago

I think the AI safety community is largely failing this test (with most people failing one but not both bullets). Many people fail the first by arguing for some weaker point, like “AI is becoming increasingly powerful, and powerful things can be dangerous”. And maybe…

Comments

daniel-eth.bsky.social•8 days ago

they convince their audience of this weaker point. But what good does that do? Perhaps a little, but it’s so weak and removed from the real concern that it probably doesn’t do much. Not clear it even helps much with then convincing them of the main point. Meanwhile…

daniel-eth.bsky.social•8 days ago

Lots of people *do* talk about the main thing, but often with front-loading parts that are hard to swallow, without first giving background context to close the inferential gap. Sometimes they used language that’s likely to cause a knee-jerk reaction against whatever’s said…

daniel-eth.bsky.social•8 days ago

I get why it’s hard to do both. Avoiding the first pitfall requires not flinching away from awkwardness of arguing something that’s “crazy”. Avoiding the second requires having intellectual empathy to understand where the audience is now & translating the case to their language…

daniel-eth.bsky.social•8 days ago

But ultimately, I do think you need to do both to convince people. (At least for now - once it’s a high salience topic that, say, 15% of people can talk about ~reasonably, then I’d expect problems on both ends to go down a lot. But we need to get to that point first)

Posting Rules

Be respectful to others
No spam or self-promotion
Stay on topic
Follow Bluesky's terms of service

Comments

Posting Rules

Reply