Oh I’m using more energy. I should really try to reduce it for the sake of the climate
I could just point out that the author of this ‘cheat sheet’ for why generative AI is not bad for the environment is Director of a Effective Altruism DC. I could leave it there. But I’ll engage with Andy Masley’s post, for a couple of reasons.
First, there are still plenty of people who don’t realise that reasonable-sounding ‘Effective Altruism movement’ is part of the TESCREAL tech bro cult. Second, Laura and I co-authored a paper for Friends of the Earth which is much more nuanced that this guy’s polemic.
So let’s get into it.
Throughout this post I’ll assume the average ChatGPT query uses 3 Watt-hours (Wh) of energy, which is 10x as much as a Google search. This statistic is likely wrong. ChatGPT’s energy use is probably lower according to EpochAI. Hugging Face released a similar much lower estimate. Google’s might be lower too, or maybe higher now that they’re incorporating AI into every search. We’re a little in the dark on this, but we can set a reasonable range. It’s hard for me to find a statistic that implies ChatGPT uses more than 10x as much energy as Google, so I’ll stick with this as an upper bound to be charitable to ChatGPT’s critics.
It seems like image generators also use 3 Wh per prompt (with large error bars), so everything I say here also applies to AI images.
Um, no. Creating an image using AI uses about as much energy as charging your phone. Before I worked on the Friends of the Earth report, I thought that perhaps developments in AI would spur development of renewable energy. And they have. It’s just that, as we mentioned in the report, for example “Between 2017 and 2023, all additional wind energy generation in Ireland was absorbed by data centres.”
ChatGPT uses 3 Wh…. You can look up how much 3 Wh costs in your area. In DC where I live it’s $0.00051. Think about how much your energy bill would have to increase before you noticed “Oh I’m using more energy. I should really try to reduce it for the sake of the climate.” What multiple of $0.00051 would that happen at? That can tell you roughly how many ChatGPT searches it’s okay for you to do.
According to the UN Information Centre, the average ChatGPT query costs approximately $0.0036 (0.36 cents). So seven times more than Masley quotes. But even then, you may think that’s not a lot of money.
Newer models, including the ones I use when doing research use a ‘chain of reasoning’ approach which are, in effect, running multiple queries. When everyone is doing this, the electricity usage grows exponentially. As we point out in the Friends of the Earth report, by 2027 the generative AI sector will have the same annual energy demand as the Netherlands. “Data centres worldwide are responsible for 1-3% of global energy-related GHG emissions (around 330 Mt CO2 annually), mainly due to the massive energy demands required to maintain server farms and cooling systems.”
Sigh. This chart 🙄
These things are not equal. Just like with a previous chart where Masley compares 50,000 ChatGPT searches with things like “living car-free” and “recycling” this misses the point. How many times do you “download a phone app” compared to the number of times you’re likely to prompt an AI if you’ve adopted it as your main search engine?
Masley also fails to realise, by shoving AI into everything, users are almost being forced into using the technology. This increases overall energy usage dramatically. In the Friends of the Earth report, we quote the UN Environment Programme as saying: “It is estimated that the global demand for water resulting from AI may reach 4.2–6.6 billion cubic metres in 2027. This would exceed half of the annual water use in the United Kingdom in 2023. Semiconductor production requires large amounts of pure water, while data centres use water indirectly or electricity generation and directly for cooling. The growing demand for data centres in warmer water scarce regions adds to water management challenges, leading to increased tension over water use between data centres and human need.”
So yes, Andy Masley, despite your protestations at the end of your “cheat sheet” the whole thing is whataboutism. There is no need to say that generative AI is somehow evil and must be banned to want governments to regulate Big Tech for the benefit of the environment. A more nuanced approach would be to say that there are systemic issues at play, and that blaming users isn’t perhaps the best strategy. Although I do think a bit more AI Literacy is needed, in general…
Source: The Weird Turn Pro
Image: Jean Woloszczyk