Error Correction and AI Alignment

There are calls to pause or shut down AI development. I want to stay out of the widespread, tribalist, parochial political bickering because I think those activities are unproductive. But I can use these issues for examples to apply Critical Fallibilism (CF) philosophy to. How could people approach the issues more productively? What can CF tell us about the situation?

This is a companion discussion topic for the original entry at
1 Like

If I genuinely thought AIs had a high risk of killing all humans, and especially if I had a lot of public attention, I would not write about off-topic political controversies like capitalism. Or spend 3 paragraphs of writing effort advocating consumption-based taxation:

Especially if I had no website to direct people to, nor even an FAQ, nor any list of debates I already won about the issue. Surely Yudkowsky has plenty of urgent things to do, in addition to having a lot of reason to avoid getting distracted and avoid alienating people with off-topic claims.

Yudkowsky is not acting similarly to how I would act if I had reached his conclusion. His actions seem more like someone who uses an issue for fundraising but still wants to continue his normal activities, not someone acting rationally regarding a serious concern about the extinction of humanity.

Yann LeCun is the “Chief AI Scientist at Meta” and he disagrees with the concerns about AI alignment/safety.

If some ill-intentioned person can produce an evil AGI, then large groups of well-intentioned, well-funded, and well-organized people can produce AI systems that are specialized in taking down evil AGIs.
Call it the AGI police.

No need to bomb data centers.

I take bombing data centers as a direct reference to (and disagreement with) Yudkowsky’s article in Time.

The Indians didn’t get to design and build the British to be subservient to them.

LeCun and I both think Yudkowsky is wrong. But I think LeCun’s arguments are noticeably worse than Yudkowsky’s. I suspect opposition like this gives Yudkowsky confidence that he’s right (which is a mistake – you can’t infer from a bunch of your opponents being awful and/or wrong that you’re right).