On Thu, 13 Aug 2026 at 10:52, Bas Wijnen <[email protected]> wrote: > On 8/13/26 9:38 AM, Lucas Nussbaum wrote: > > Do we really know a lot about the environmental impact of AI? I have the > > impression that we have legitimate questions or concerns about it (that > > are acknowledged in several ballot options including the one I proposed), > > but not that we are experts on the topic. > > We are not experts on climate change, but really, saying at this point > that we need more research on that before we can agree that it is > happening, has extreme tobacco-industry-vibes. We know climate change is > real and that it is already destroying the planet. With the current > state of affairs, non-experts are allowed to say that. > [snip] > We can also see how much damage they do to the environment. Again, we > are not experts on that part, but we don't need to be; it's very > obvious. So while our impression of that damage may not be as accurate > as it could be, there is no reason to pretend that maybe the damage is > very minor. It obviously is not, and everyone can see it. >
Really? Is it? I do *not* see that at all. There is nothing inherently dirty in the AI technolgies or datacenters in general. They do not require using oil or coal or killing baby kittens to produce eitehr the models or the answers to questions. All they are using are electricity and hardware. There is absolutely no reason why AI can not be done in a sustainable way and .... *surprise* a LOT of the AI already out there already *is* being done in sustainable ways. Just because weak oversight in USA allows *a few* bad apples cut all the corners and power their AI datacenters with gas turbines or old coal power plants does not mean that *all* AI is dirty and damaging to the environment. It is the same fallacy that oil companies have been driving for decades against electric cars - "the grid is so dirty, the electric car can never be green!". Meanwhile the grid is becoming green and sustainable very rapidly and even an electric car powered by 100% coal-generated electricity produces less emissions than a gasoline car. We see the same fallacy in this discussion, in two ways: 1. Focus on worst case and throw out all (already successful) efforts of sustainable AI creation and usage. It's like refusing the whole concept of shoes only because some American money-barons figured out a cheap way to make shoes out of baby seals. All shoes do not become unsustainable just because *some* are. Same with AI - *all* does not become "climate destroying" just because some are. Not all power usage is equal. Using green power *funds* buildout of more green power. Building out battery backusp for your datacenter to better use cheap off-peak power *helps* the grid to become more stable and more green. And we see all of that happening in jurisdictions with proper regulation. 2. Ignoring efficiencies in technology and scale. Even if we for a moment assume that using more green, electrical power to achieve the same result is somehow bad. AI processing requests in a high efficiency data center can save me enough time and effort that my local computer will consume less power than the added power consumed by the request itself. There are *massive* AI efficiency gains in the pipeline, as scaling up the usage with the current efficiency is not feasible. See https://www.youtube.com/watch?v=3MKRjt59hh4 for example - AMD recently bought a company Taalas that designed an efficient way to "bake" a base LLM into the hardware of a chip. You can try it at https://taalas.com/products/ . The key is that it is producing approximately 50 *times* more tokens of output per second (and per watt) compared to generic AI inference chips. It is a 2.5kW server outputing 17k tokens per second. That is with 1 kWh of power it can output 24.5 *million* tokens in replies. For serious requests consuming, say, 20k tokens that would be 1000 replies per kWh or at EU power price of 0.35€/kWh about 0.00035€ worth of electricity used to respond to one request. I am pretty sure that some non-cached requests to Debian web services consume more power. Never mind any kind of CI test execution. And a few such requests can easily save me several hundreds of watts of running *my* computer in my appartment while I am debugging the same issues. So, excuse me if I do *not* see "how much damage AI does to the environment". *Some* AI companies (like xAI/Grok) *sure* are doing masive damage, but that is on those specific companies and the non-existent regulatory environment. It has basically nothing to do with AI itsef. If you care about the environment - advocate *for* green and sustainable AI. Not against all AI just because some of it is not green right now. -- Best regards, Aigars Mahinovs
