A closer look at the intersection of artificial intelligence and sustainability news for the KAOH media team.
Greetings, KAOH crew! Welcome to the first installment of the KAOH AI Sustainability Update. With this newsletter, I want to carve out an internal space where we can discuss the very real climate impacts of AI development alongside the also very real potential these technologies have to not just make our work more efficient – helping us work to bring more clean energy online faster – but also play a more direct role in bringing about a more sustainable world.
We’ll talk about the good, the bad, and the ugly so we can all make better decisions about how we use AI at work and in life.
xAI's Memphis data center. (Courtesy Steve Jones Flight by Southwings for the Southern Environmental Law Center)
If your timeline is anything like mine, you’ve been seeing journalism exposing dirty data centers, like Elon Musk’s xAI Colossus supercomputer in Memphis, TN, which contracts some power from the local electric utility but runs its own gas turbines on site. Because of a loophole allowing “temporary” gas turbines, the company has run these polluting gas-fired turbines unpermitted for a full year, and they still have no air permits for the pollutants they emit, including formaldehyde and nitrogen oxides. Residents in this area of the city, Boxtown, have long had to contend with industrial pollution from 17 other polluting facilities in their vicinity, and they have reported that air quality has worsened in the year since Colossus arrived. The NAACP stepped in to ask the city to halt the center’s operations and more recently filed an intent to sue xAI.
Locals oppose a data center proposed for New Braunfels, TX. (Courtesy Dylan Baddour/Inside Climate News)
Similarly, in Texas, a string of data centers and associated natural gas plants is prompting concerns about noise, air, and light pollution across many communities. The data center rush underscores the urgency of grid infrastructure overhauls. Currently, many of these new data centers operators are choosing to go online faster with on-site natural gas turbines rather than waiting years to be allowed access to a grid interconnection point. As supplies constrain the immediate availability of independent, on-site gas-fired turbines, data center developers are also contracting privately with large natural gas plants.
As data centers continue to boom, it is imperative that we make room on the electric grid so that they can benefit from grids that operate using wind and solar. In the meantime, one creative proposed solution from RMI is to couple wind and solar development with peaker plants, which are high-emitting gas power plants that are designed to be used rarely and only when needed, like during a cold snap or a heat wave. Developers would overbuild wind, solar, and batteries and connect them not just to the local grid interconnection point – the same interconnection point as a peaker plant – but also directly to the data center. When the renewables sources outproduce the data center’s needs during peak times, there’s an infrastructure already in place for the excess to go to the local grid. When renewables sources cannot meet the needs of the data center, it can then draw in limited ways from the peaker plant, which acts like a “circuit breaker” between the data center and the local grid.
Credit: RMI
AI Model Spotlight: DistilGPT2
DistilGPT2 is designed to be similar to GPT-2 while being significantly more compact using knowledge distillation. DistilGPT2 requires less computational power and electricity to run than GPT-2 (82 million parameters vs. GPT-2’s 124 million), making it a more environmentally friendly option for organizations and individuals concerned about the carbon footprint of AI.
DistilGPT2 is trained on the OpenWebTextCorpus and released under the Apache 2.0 license. It is suited for applications like writing assistance, creative content generation, and chatbots. While it is efficient and easy to deploy, DistilGPT2 can still produce biased or inaccurate information and should not be relied upon for tasks that demand high factual accuracy or ethical judgment without further safeguards or customization. Its lighter environmental impact makes it a practical choice for businesses, educators, and developers looking to balance AI capabilities with sustainability.
[This one is a bit more of a nuanced about the tradeoffs rather than straightforward benefits, but it was a super interesting read!]
Takeaway Tip
Remember you're talking to a computer! While AI interfaces allow us to talk to computers without coding knowledge, it still processes things differently than a human conversational partner. Drop “please” and “thank you” from your query to save a little bit of computational power.
Copyright (C) *|CURRENT_YEAR|* *|LIST:COMPANY|*. All rights reserved. *|IFNOT:ARCHIVE_PAGE|**|LIST:DESCRIPTION|**|END:IF|*
Our mailing address is: 533 Palisades Rd. Wildwood, MO 63021
