technologist (noun): a specialist in technology.
Merriam-Webster Medical Dictionary
technologist (noun): a person employed to look after technical equipment.
Google, via Oxford Languages
I recently had the pleasure to read Drew deVault's provocatively-titled "Please stop externalising your costs directly into my face" - and its equally provocative assertion:
If blasting CO2 into the air and ruining all of our freshwater and traumatizing cheap laborers and making every sysadmin you know miserable and ripping off code and books and art at scale and ruining our fucking democracy isn’t enough for you to leave this shit alone, what is?
This isn't deVault's first bout with what is currently being marketed as "AI": in an earlier post he asserts that while there's a bubble, the generative technologies being peddled aren't going to go anywhere, and capitalism is going to push them to be used in ways that make our world worse.
While there is much discussion of the human cost of using these technologies, including:
- Liabilities arising from the collection of training data
- The concern that Generative AI technologies will cheapen the work of artists
- Job security threatened by the adoption of automation
- The ramifications of AI-generated works and copyright law
the body of evidence supporting the environmental impact of Generative AI technologies has somewhat lagged the development of those technologies. Historically, analysis of the long-term impacts of the development of a new technology has been reserved for after the technology has been assessed for its short-term viability. With "prompt engineering" at one point being considered a legitimate career pathway (and, I will note, a hilarious one - I was taught how to formulate queries for early search engines when I was ten - and early search engines were a lot less clever at processing natural language than modern parsers), it would seem that the pathway to relevance as a worker remains the same as it always has when automation has arrived: learn a new skill bro. just one more skill and you'll be economically secure bro i promise.
Sadly, automation is here, and no matter the skills you cultivate, the job market, by all accounts, sucks (however, the hard data would seem to suggest that it's getting better) - especially given the seeming lack of consideration given to applicants by many firms, and what commentators have argued to be a shift in the social contract underpinning the employer-employee relationship.
I'm not an economist, and I don't completely understand the complex factors leading to our modern-day hellscape save for the unrestrained hyperaccelerated selfishness of the lucky few trickling down pain and suffering on us unwashed masses. Instead, as a technologist (defined earlier), I have to look at technosocial developments through the lens of my responsibilities - and the idea that if you keep chipping away at a finite resource - say, the biosphere of an entire planet - it will eventually be depleted, or perhaps worse, transformed beyond recognition.
My jobs require that I tend both people and machines - in the open source spaces I frequent, fostering collaboration is not simply done with a git forge and a tutorial - it is done by building a community people want to be part of. In my position assisting children to develop social skills, I don't just have to do tech support to make sure the games work, I have to be able to talk to a person, and to be able to encourage them to talk back. In providing end-user support, I not only have to be able to diagnose a power supply failure, but also manage expectations on repair, and explain to the user clearly how a failure has occurred, and how we're going to fix it. If you think technical jobs don't require people skills, you might be better suited to being a deep space investigator somewhere (spoiler: you end up needing people skills for that, too).
I'm not going to refuse to work with people who've used or built Generative AI tools. I'm seldom going to turn away someone coming to my projects or jobs in good faith. But I do think it's time to admit that the fantasy of getting a computer to do all our work for us is just that - a fantasy.
How much water would be consumed if every person on the planet (some sources assert ~8.1bil, but the true number may be higher; I'll compute for eight billion people as a nice round number) took a shower? A perfectly average shower? (We assume the average shower lasts eight minutes and consumes 15L.min-1 based on a spread of shower water use for a total of 120L water consumed.) Eight billion people, all using 120L - that's 960×109L - nearly one trillion litres. Based on OECD projections, I'm going to estimate the current yearly use of water by the training and querying of Generative AI tools at 3.5 billion cubic meters, or 3.5 trillion litres - 3500×109L (don't @ me, mathematicians). While this refers to withdrawn water, not consumed water, it is important to understand that this water consumption itself requires pumping, which requires energy, which uses water. It is also necessary to note that we can't just dump saltwater into cooling systems; water for cooling (the primary use associated with the water consumption of datacenters) must be pure. To use saltwater would require desalination and purification; to my knowledge, datacentres generally are not installed near the sea, nor are they fitted with desalination systems.
In effect, in a year, all the AI models used worldwide withdraw the same amount of water as three and a half days in which every single person alive worldwide takes a nice, satisfying eight-minute shower.
I would focus on electricity, but quite frankly, it's the easiest thing in the world to get. Nuclear fission has been capable of providing clean and safe energy since the 1970s (mismanagement aside), and solar power and wind power both are free and abundant if you don't mind barely-school-age kids strip-mining the shit out of the African continent for tantalum, lithium, and lanthanum. I will not be citing these assertions, as they are well-established in the literature already.
I didn't actually have all the numbers before writing this, and I already understood that the never-ending hunger for power shared by both successful capitalists and Generative AI systems was catastrophic to the environment. I have the numbers now, and stand firm in my belief that protecting technical assets (the environment, the welfare of people, the welfare of computer systems) is incompatible with the further adoption of Generative AI tools. The only way that you can ethically work on these tools at this point is to be actively working, on the smallest scale feasible, at the task of making these "AIs" less resource-intensive to run.
Automation has, since the earliest days of the industrial revolution, tantalised us with the promise of workers needing to do less, to be able to spend time with their families, to have recreation and rest. And since those heady days, the rich have used it to get richer and richer without materially improving the conditions of the working class.
There's never been a better time to be proud of doing work yourself - creating art, fixing problems, making things, serving people - and to demand a fair wage. Because they can't replace us all with AI, and if they somehow try, the skies will burn from what we do to the Earth.
And if you want to work on making datacentres more sustainable, and you're willing to invest to lead the way - talk to me. I have ideas, and they can turn a profit if you're willing to play the long game.
Further reading:
- The Price of Prompting: Profiling Energy Use in Large Language Models Inference
- The growing energy footprint of Artificial Intelligence
- LLMs and the effect on the environment
- A Survey of Sustainability in Large Language Models: Applications, Economics, and Challenges
- LLMCARBON: MODELING THE END-TO-END CARBON FOOTPRINT OF LARGE LANGUAGE MODELS
- From Words to Watts: Benchmarking the Energy Costs of Large Language Model Inference
- Carbon Emissions and Large Neural Network Training
- anything else you want, the more information you absorb, the more right to an opinion you have honestly