Should We Ban Imitative AI to Save the Environment?

Imitative AI is an environmental disaster.

It takes an enormous amount of water and electricity to cool and power the systems that produce imitative AI results. Imitative AI is using so much power that it is delaying the retirement of coal burning plants. It is very difficult to find out exactly how much water AI data centers use, likely because they don’t want us to know, but all research and investigations suggest it is a lot. This is not part of a broader tech industry issue, either. Imitative AI just requires a lot more compute and storage than other kinds of systems, meaning they use a lot more of both for the same results. An imitative AI search query, for example, takes up to ten times as much energy as a normal search request. And to get what?

Not a lot. Even putting aside the issue of taking people’s material to train these systems without permission, they really haven’t lived up to the hype. They aren’t good at pretty much anything not related to misinformation, especially when you consider their environmental cost. So the question then arises: should we ban these systems in order to protect the environment?

I don’t like the idea of banning research. Bluntly, we don’t do enough research in this country, and it is not outside the realm of possibility that imitative AI could produce some breakthrough that makes it or its successor truly useful. But right now, that seems a distant prospect and it has to be weighed against the very real damage that AI is doing to the environment. It’s not just the power and water, by the way. Building more data centers often requires extensive carbon outlays — extensive enough that Microsoft’s planned building spree, largely driven by AI needs, is going to likely mean they cannot be net-carbon neutral as a company as they had originally planned.

The climate crisis is very real and while it is not a sure thing, recent projections offer some hope that we are making progress in bringing down the carbon emissions that drive it. But AI leaders are now talking about how they cannot get to carbon neutrality. Sam Altman has said that “there is no way to get there without a breakthrough”, for example. But that breakthroughs aren’t coming — Bill Gates next generation nuclear technology is suffering setback after setback. And we have been five years away from commercial fusions plants for almost as many decades as we have been five years away from artificial general intelligence.

Given these facts, it is hard not to think that the time has come to get rid of imitative AI. And please do not tell me that imitative AI — a system that produces a best guess about what should come next — is going to help solve the climate crisis. Its own math means that it drives toward the median of its training data. You will get no solutions from it, and any help it provides can certainly be provided by more efficient machine learning and classification algorithms. At some point, the externalities must be taken into consideration for every technology, policy, and plan. And right now, it looks very much like imitative AI is going to significantly damage the planet in order to produce deepfakes and email summaries that you probably need to verify before you trust them anyway.

That really doesn’t seem like a good trade off to me.