I think the environmental aspect is interesting and worth discussing. Around the offices the common joke is that people will "just burn down a piece of the rainforest" when they fire up their AI to solve some complex problem. Which certainly isn't what the world needs right now, and you can't have the "tool" without also the massive water consumption in a world where not everyone has access to clean water. Though as the fatalism in the burning down the rainforest implies, people around here have sort of accepted that the world is going to get hot.

On the other hand. If we apply the same sort of fatalism to AI, then we can expect AI to lead to civil uprising and a world which will probably be a lot more sustainable once most of us are dead.

I don't think the automation is any different from what we've seen the past 150 years. Except that perhaps this time AI is the tool which is actually going to do to the office what the assembly line did to the factory.

I think there's a larger problem of externalities (i.e. environmental costs) not being fairly represented in transactions. AI is just the latest and biggest symptom of this. If someone has a business process which involves generating a waste product (or utilising some resource) then that's only a problem if the cost isn't being paid by that business (and ultimately reflected in the price charged for the product/service).

Obviously that's not the whole problem - the other issue is the financial resources these companies have access to. Presumably if they wanted to, instead of pushing up the worlds prices for DRAM or electricity they could have distorted pretty much any market they wanted to given the money they have. If instead of datacentres they decided to buy the world's supply of coffee then presumably people would be paying $100 for a tin of coffee beans. You could introduce all sorts of restrictions and market controls but for me a better reform would be ensuring that they didn't have hundreds of billions of dollars to spend in the first place.

>Except that perhaps this time AI is the tool which is actually going to do to the office what the assembly line did to the factory.

I think this is probably right, but for the people who worked on assembly lines they (or more realistically their kids) could go find an office job to do instead. It's not clear what the kids of todays office workers are going to do for employment (if anything).

Neither the power, nor cooling, requirements are necessarily environmentally harmful. As always ymmv.

First water; there are plenty of places on the planet where water is bountiful. If you site a data center well (say on a big lake in a rainfall area) then water is not an issue. An "using" water in this way doesn't affect water scarce areas.

Equally lots of electricity can be generated in environmentally harmless ways. Solar, wind, hydro are all clean. (Hydro depending on how and where.)

Yes, in the long run, it may make sense to put data centers outside the US. Norway for example has no problem cooling things down. And hydro is abundant.

Plus, water "consumption" is also variable. Water is used for cooling, but is not necessarily "lost". Its "used" in the sense of "made use of" but may also then be "used for something else".

I wonder if this fatalism comes from a feeling that the enormous power of AI must somehow come with some sort of proportional cost. There isn't really any clearly known cost so maybe that's why people talk up the water and global warming and social impact fears and all that.

In comparison, does anybody say they're burning down a bit of rainforest when they go to the toilet (and flush it)? Or when they cook dinner at home with more waste heat than factory cooked food? Or throw away a pen that's run out of ink? The environmental impact of all the mundane things in life is far greater than using AI but the impact of each mundane thing is perhaps also far less impressive.