I’m very much a luddite when it comes to AI, I’m unequivocally against its usage and development. The social and environmental costs AI incurs and proposes to incur are overwhelming. The celebration of AI with Chinese characteristics is often justified by the idea that AI can be useful, and the reason why it’s OK to be vehemently against it otherwise but support it when it’s being implemented by China is because AI can only be a force for evil within the capitalist framework. I agree with the latter point obviously but find it hard to accept that China driving AI advancement is a good thing for the world.

I don’t want to have a debate about whether China is a revisionist state, but I think we can all agree that China has refused to take the role of the Soviet Union in exporting socialism to the world. What happens when it exports AI to countries world over that very much operate within a capitalist framework while China does nothing to advance world wide socialism?

you are viewing a single comment's thread
view the rest of the comments
[–] 4 points 11 hours ago* (last edited 8 hours ago)

Good question, but I wouldn’t be addressing this if it wasn’t being implemented at scale in a desperate manner.

I guess my point is that capitalism is the driver for the implementation at scale in a desperate manner.

Is there any way to operate AI with significantly lower resource demand to the point that is a non-factor?

I can't back this up with facts and figures, but my I understanding is yes, based on these reasons:

  • at least some portion of the built out is speculative. I'm not sure how the popularly reported resource consumption data is generated, but I imagine it includes planned and or approved data centres that may or may not ever get turned on, or even built. Maybe the dotcom bubble is a good analogy here, people wee just buying up all sorts of domains and putting together basically slop websites with the hope they could cash in. It doesn't mean all those websites materialized.

  • if the computing infrastructure is planned and incorporates good practices, there is no reason renewables can't be used for the energy (including cooling energy). I think this addresses the electricity and cooling resources.

  • As for the hardware and mineral resources needed for those, we know a lot of that is speculative production. I hope models will become more efficient, but efficient models don't sell ram 😉

And we still haven’t addressed the social consequences of it.

I tried to touch on that with the infrastructure but I did skip the less tangible aspects. Again, I think the rush to be the first and create a demand for AI has been the cause of many of the social harms. It's just thrust upon us as this great thing...without any real instructions or notes about limitations. If it was developed to solve specific problems and not just bump up usage numbers for quarterly reports, perhaps a lot of that could have been avoided.

I have found it useful for some things, but I'm also old. I think that (for better or worse) this lets me see this as a new thing I can use, instead of "the way the world is" as someone who uses it in school might see it. Without remembering how we used to do things, and what the pros and cons of using AI for these tasks are, I think we are in for a heap of trouble.

Eta: I believe the vast majority of data centres approved in Ontario use air cooling, so yes there are places where water cooling isn't depended on.

  • source
  • parent