It seems that there is one issue of bipartisan agreement during this week’s Presidential transition: federal support for building more AI data centers.
In his final days, Biden directed the gov't to lease federal lands to build AI data centers. And Trump appears to have endorsed the plan.
Also note the careful wording on requiring data center developers to “bring online sufficient clean energy generation resources to match the full electricity needs of their data centers.”
That is not the same as being fully powered by clean energy, as @agordon.me explains: youtu.be/jFd0QUOUFUg?...
Dr. Shaolei Ren, expert on data center water usage, says AI companies have already exceeded the water usage of the largest beverage companies in the world. youtu.be/Vqvb3G8hJiI?...
No mention of water conservation in White House EO giving federal lands to AI data centers.
Data centers are very thirsty creatures!
www.whitehouse.gov/briefing-roo...
As reporter @agordon.me says, “Next time you have a conversation with a chatbot programmed to sound like a movie star, you can think not only about the bottle of water evaporating into the air but all the climate promises that are now in jeopardy.”
www.proofnews.org/ais-unquench...
And there’s another problem for AI data centers: it’s hard to be both water and carbon efficient.
Places that are hot and dry tend to be good for solar power, and places that are cool tend to use more water to generate electricity. So companies may have to pick which climate goal to sacrifice.
You can see the rising water usage of AI in the numbers.
In 2022 Microsoft water consumption rose 34% and Google’s rose 22%.
In 2023 Microsoft’s entire operations, including all data centers, consumed enough water to fill 15.6 billion water bottles.
The bottled water estimates are from a 2023 study from UC-Riverside.
But the real numbers are likely much worse. They studied older, smaller AI models running in big tech owned data centers. Today AI is bigger and often hosted in less efficient 3rd party data centers.
arxiv.org/pdf/2304.03271
You may have heard that having a conversation with ChatGPT is the climate equivalent of pouring out a bottle of water. In fact, that is likely an underestimate.
In the 3rd installment in our series on the climate costs of AI, we tackle AI’s unquenchable thirst.
www.youtube.com/watch?v=s67L...
As @agordon.me writes: “Tech companies seem to be treating the climate impact of AI like they’re shopping in a store that has no price tags with a credit card in someone else’s name. And they’re not showing us the receipts.”
www.proofnews.org/ais-carbon-f...
Our educated guess of the CO2 costs of AI is definitely an underestimate - it is based on pioneering 2022 research by @sashamtl.bsky.social @sylvainviguier.bsky.social & Anne-Laure Ligozat.
The real cost of newer, bigger AI models is likely much higher.
arxiv.org/pdf/2211.02001
What is the carbon cost of using an AI bot? The companies stopped disclosing it in 2022.
In the latest installment in our series on the climate costs of AI, we estimate a year of ChatGPT use may emit the same amount of CO2 as driving 2 million miles.
www.youtube.com/watch?v=yekA...
Since 2017, electricity consumption at the biggest AI makers has soared:
186% at Google
186% at Microsoft
367% at Meta
The reason: AI.
@agordon.me examines the electrical costs of AI in the first episode of our YouTube series on the climate impacts of AI.
www.youtube.com/watch?v=jFd0...
Hypothesis: Tech companies energy use is skyrocketing due to data center expansion and they're losing ground on their attempt to reduce carbon emissions. Sample size: Corporate sustainability reports for leading tech companies as well as non-tech Fortune 25 companies. Techniques: We compared total electricity consumption in gigawatt-hours per year by company/ Key findings: Since 2017, Google electricity consumption has increased 186%. Microsoft 186%. Meta 367%. A smaller share of Google's data center electricity is powered by clean energy now than in 2020. Limitations: Only Google reports the percentage of data center electricity powered by green energy. Not all companies report electricity use. There is no generally accepted method for separating out AI usage within data centers.
We believe that news should come with an ingredients label – so here is our label for this episode:
www.proofnews.org/ais-power-gr...
A smaller share of Google’s data center electricity is powered by clean energy now than in 2020. The other AI companies don’t disclose clean energy ratios, but they are likely to be similar.
“In a lot of ways, they are going backwards.” – @agordon.me
www.youtube.com/watch?v=jFd0...
AI makers are trying to replace their fossil fuel usage with green energy. Amazon, Meta and Google were 3 of the top 4 corporate buyers of clean energy purchase agreements in 2023, and all have invested in nuclear power.
But their electricity use is soaring faster than they can add green energy.
Since 2017, electricity consumption at the biggest AI makers has soared:
186% at Google
186% at Microsoft
367% at Meta
The reason: AI.
@agordon.me examines the electrical costs of AI in the first episode of our YouTube series on the climate impacts of AI.
www.youtube.com/watch?v=jFd0...
If you haven’t tried Nebula.tv, give it a whirl! It’s a great way to watch your favorite YouTubers, without ads and human-curated without algorithms. We’re honored to be part of their lineup.
go.nebula.tv/proofnews
We are thrilled to announce that Proof News is joining Nebula – the independent creator-led video streaming platform. Our video investigations will appear on Nebula as well as on the corporate-owned platforms.
Indie newsroom + indie platform = stronger together!
nebula.tv/proofnews
Proof News is an independent nonprofit newsroom. We do rigorous investigations of powerful companies. We have no corporate overlords. We have no ads and no third-party trackers on our site.
We are honored DuckDuckGo chose us as one of their annual donations!
spreadprivacy.com/2024-duckduc...
"In the world of AI and misinformation and propaganda circulating online, we must make it very clear why our audiences should trust our piece of information versus another piece,” @juliaangwin.com tells Editor & Publisher.
www.editorandpublisher.com/stories/proo...
In our latest AI testing, the leading chatbots were pretty decent at detecting and debunking known vote count misinformation. With one notable – and surprising – exception: ChatGPT.
Honored to have collaborated on this investigation with Factchequeado and @alondra.bsky.social
AI models are fluent in many languages - holding out the promise of helping break language barriers.
But our latest investigation reveals that the models return a higher rate of inaccurate answers to questions posed in Spanish than identical questions posed in English.
Google’s Gemini AI has stopped advising people to eat rocks. But other AI models — looking at you OpenAI and Meta — are advising people to lick rocks.
The latest @proofnews.bsky.social collaboration with creators to test AI in their areas of expertise. 👇🏻
youtu.be/LuN5Qo3TPYs?...
“If a group of people might be disproportionately affected when AI models are making mistakes, that's important for voters to know and for the AI developers to know, [CDT’s @isabelalinzer.bsky.social] said.” www.proofnews.org/ai...
“AI models are frequently putting out false and inaccurate information that could dissuade voters with disabilities from voting, according to new research from CDT using an AI model comparison tool created by @proofnews.bsky.social.” cc: @eedugdale.bsky.social www.proofnews.org/ai...
