nam_

This is one of the more lucid plans on ai safety and alignment, although in fairness it would have to involve massive amounts of civilisational cooperation- I would not necessarily put it past humanity's abilities (see transnational agreements on nuclear warheads and their proliferation). I think currently there is only possible path for the complete execution of plan R:

• There are some warning shots as to the potential danger of AI (current trajectory would suggest that some sort of cybersecurity danger is imminent, maybe the appearance of large scale super

The labs are racing not because they desperately want to become rich, but because they are afraid of losing to another lab.

The prospect of 'losing to another lab' is only threatening because it implies massive deductions in revenue and users. Without AI labs interacting with each other and competing in a market-esque system, the loss associated with not having the latest model is comparatively lower.

For example: nuclear projects around the world were not 'racing' with each other to build warheads out of pure scientific reasons (I think scientists would ha... (read more)

I agree with the thrust of your article. I do not think that this is a case where the moral nature of the beings in question (whether it be fly connectomes or AI models) is relevant though. As you put it yourself:

You don't have to know that AI has moral worth to want to stop people from torturing it, just in case, and if it's easy.

Reflecting on why this is the case might be fruitful. I think it is because of the fact that the willingness to torture reveals something about the actor's ethical disposition; who knows what they will do, given that they are wil... (read more)