Relax: AI is not coming for your job, only for your energy and water
- Sergio Visinoni from Sudo Make Me a CTO <makemeacto@substack.com>
- Hidden Recipient <hidden@emailshot.io>
Hi, 👋 Sergio here! Welcome to another free post from the Sudo Make Me a CTO newsletter. If you prefer to read this post online, just click the article title. As this is a free newsletter, I do immensely appreciate likes, shares and comments. That's what helps other readers discover it! Relax: AI is not coming for your job, only for your energy and waterExploring the real implications of GenAI and LLMs in energy and water consumption, and the naive beliefs around nuclear fusion.
Today's article differs slightly from the usual reflection or recommendations for CTOs and other engineering leaders. It sits at the intersection of culture, technology, and environment. Everybody and their mother have been jumping on the GenAI psychedelic bus. As is often the case in online discourse, the discussion is acutely polarised. On the one side, people believe the immediate next step after GenAI is AGI, superintelligence, a modern take on the Philosopher's Stone¹. On the other side, we have people who believe AI will lead to the doom of humanity in the long term, starting by replacing the jobs of millions of people until it finally takes over control of the entire planet. While I'm wary of the totalitarian techno-optimist approach, for reasons that I'll detail later, I also don't believe AI, and specifically the current generation of GenAI solutions primarily based on LLMs, will end up having a massive impact suggested by the darkest predictions. Nobody is in a position to predict what the long-term implications of such technology will be for humanity. I'm particularly skeptical of individuals or organizations venturing into such predictions, as these have more to do with religion and faith than facts and observations. Humans are terrible at long-term predictions in two ways:
But while everybody is busy doing just that — trying to predict the long-term implications of this new technology — marking a new high in online tribalism, we're not paying enough attention to something more mundane that is already happening. GenAI is driving a surge in energy and water consumption right now when technology should be deployed to shift the trend in the opposite direction. If we don't reverse this trend, someone will have to revise the ridiculous prediction that 50 billion humans will live on this planet anytime soon.⁷ On a more serious note, let's look at some data and facts on our overall situation.
With all that in mind, let's look at the world of Big Tech and GenAI. Dr. Sasha Luccioni is the Climate Lead at Hugging Face, focusing on understanding LLMs' environmental implications. One interesting finding of her research is summarised in the following quote:
Let's put this into perspective. In 2022, Data Centers worldwide accounted for about 2% of global energy consumption. That number is expected to double by 2026. Roughly the amount of energy used by Japan:
I'm not sure how those estimates were made, but with more and more companies integrating their self-proclaimed “smart” tools into their existing products, this number could grow even faster. With Google rolling out its AI answers everywhere, we'll have enough data to update the forecast soon.¹² Some might argue that my concerns are unjustified. After all, Big Tech is pushing its renewable energy agenda. Everything they do has been green since its inception. Everything they do is carbon neutral, so why should I even bother? This reasoning is weak on two main points:
Let's take the example of Microsoft, the company that prides itself on being a first mover in the LLMs gold rush. In 2020, Microsoft made an ambitious commitment to make its operations carbon-neutral within the decade. That means by 2030. The reality today is quite different. Depending on the accounting approach, today's emissions at Microsoft are between 29% and 40% higher than in 2020. Definitely a disruptive way to reduce emissions. When asked about it, Melanie Nakagawa, Microsoft’s chief sustainability officer, has a beautiful answer to offer:
Nakagawa didn't say it explicitly, but she's referring to nuclear fusion, the final solution to climate change predicated by most Techno Optimists. There is a religious-like belief shared by most tech bro's that nuclear fusion will solve all our problems with renewable energy. Maybe. Google's SREs made famous their motto: “Hope is not a strategy”. Yet, that's precisely the strategy deployed in this case. While nuclear fusion could play a significant role in the long term, its viability within the short time horizon we need to operate in to avoid hard-to-predict catastrophic events is unlikely to be substantial. In the words of another Microsoft employee, Kate Crowford:
For those unaware of it, Crowford is talking about Helion's Energy. Sam Altman is the company's board chair and has invested $375 million in it. OpenAI is apparently in talks with Helion to buy massive amounts of fusion power… whenever that will be produced. Another rabbit hole I don't have the intention to dig into right now.¹⁵ Then, there is the problem of water consumption. A paper from October 2023¹⁶ shed some light on the less talked about externality of GenAI. According to the research, the impact is far from trivial:
Though there are expectations that energy efficiency and new GPUs will improve energy efficiency, this inefficiency is real today. Widespread usage and glorification of LLM-based tools are making the problem more acute. Where is all this taking us? Should we stop using AI and become Luddites? Not at all. But we should be mindful of the implications of our decisions and actions. We should ask ourselves whether the benefits of using specific tools are worth the environmental cost or externalities, as economists call them. This is something we can do at the individual level. When it comes to the overall tech industry, this is ultimately an issue of prioritization. Money is a finite resource, as is time. As any good investor does, opportunity cost should be evaluated when deploying capital for some purpose. When all the VC money in the world is invested in a technology that so far is only making matters worse for the climate, we have the moral obligation to consider the opportunity cost of such investments: what would be the impact on life on this planet if part or all of that capital was deployed to reduce carbon emissions as quickly as possible? And don't tell me that this is utopian as the ROI would not be worth the deployment of capital, markets self-regulate, and that GDP growth is the only thing that matters. Call me a socialist if that makes you feel better; I don't care. But while you do so, please share your arguments to explain the ROI calculation of making life on this planet miserable for billions of people.¹⁸ I haven't found a manifesto covering that yet. If you found this valuableIf you found this valuable, here are other ways I can help you and your company:
1 Quoting verbatim from the totalitarian cult-like “Techno Optimist Manifesto” by people that seem to have lost contact with reality over at a16z: https://a16z.com/the-techno-optimist-manifesto/ 2 More details can be found here if you want to nerd out about Bakelite and its inventor: https://en.wikipedia.org/wiki/Bakelite 3 Though the study could not determine with certainty any health impact of such a finding, in a study conducted on samples of human and dog testicles, they found that every single one of them contained microplastics. https://www.scientificamerican.com/article/are-the-microplastics-found-in-testicles-a-health-danger/ 4 Another fascinating example is Thomas J. Midgley, who invented both leaded gasoline and CFC. In his case, though I don't feel confident enough to buy the thesis, he was fully unaware of the consequences, as he experienced the side effects on his own health. https://www.history.com/news/cfcs-leaded-gasoline-inventions-thomas-midgley 5 If you're not familiar with the word, here is a summary of how precise it is at conveying a temporal reference: https://www.spanish.academy/blog/why-ahorita-in-spanish-almost-never-means-now/ 6 This phenomenon seems to be more acute in the US than anywhere else. See this fascinating article for a more in-depth look at the issue. 7 This is another reference to the “Techno Optimist Manifesto” from a16z. 8 See https://www.reuters.com/business/environment/january-was-worlds-warmest-record-eu-scientists-say-2024-02-08/ for more details. 9 https://www.reuters.com/business/environment/global-warming-will-reach-15c-threshold-this-decade-report-2023-11-02/ 10 See https://www.reuters.com/business/environment/fossil-fuel-use-emissions-hit-records-2023-report-says-2024-06-19/ for a recent study published just a few days ago. 11 For the article where the quote is from, see here: https://www.vox.com/climate/2024/3/28/24111721/ai-uses-a-lot-of-energy-experts-expect-it-to-double-in-just-a-few-years 12 Current studies estimate that an answer provided by Google AI uses about 10x more energy than a regular Google search: https://www.scientificamerican.com/article/what-do-googles-ai-answers-cost-the-environment/ 13 For the full article see here: https://www.geekwire.com/2024/microsofts-carbon-footprint-keeps-growing-as-ai-drives-data-center-expansions/ 14 Here you can find the full article https://www.nature.com/articles/d41586-024-00478-x 15 If you want to go down this rabbit hole: https://www.geekwire.com/2024/openai-reportedly-in-talks-with-helion-energy-to-buy-vast-quantities-of-fusion-power/ 16 Here is the full paper https://arxiv.org/pdf/2304.03271 17 This is the article where the footnote is taken: https://deepgram.com/learn/how-ai-consumes-water. For more on this subject, see also https://www.newsweek.com/why-ai-so-thirsty-data-centers-use-massive-amounts-water-1882374 18 According to specific models, by 2050, about 5 billion people will be exposed to at least a month of health-threatening extreme heat when outdoors in the sun https://www.washingtonpost.com/climate-environment/interactive/2023/extreme-heat-wet-bulb-globe-temperature/ Sudo Make Me a CTO is a free newsletter edited by Sergio Visinoni. If you found this post insightful, please share it with your network using the link below. If you or your company need help with one of the topics I talk about in my newsletter, feel free to visit my website where you can schedule a free 30 minutes discovery call. I'd be delighted to investigate opportunities for collaboration! |
Similar newsletters
There are other similar shared emails that you might be interested in: