Environmental Impacts – Kevin’s Meandering Thoughts
2022-02-11_21-13-24_ILCE-7C_DSCAG3501_DxO flickr picture by Miguel Discart & Kiri Karma shared underneath a Artistic Commons (BY-SA) license
Oh, the irony of ironies – I used Google’s NotebookLM to assist me with this activity of trying on the environmental prices of AI by means of the lens of three completely different articles. What a slacker I’m! And in trying on the environmental influence, I added to it on this seventh day of the 12 Days of AI.
BUT the NotebookLM AI was in a position to summarize and collate and thematically string collectively info and insights from throughout the three articles in a really environment friendly approach. The articles have been:
- Through Kleinman Heart for Power Coverage: https://kleinmanenergy.upenn.edu/commentary/podcast/why-ai-consumes-so-much-energy-and-what-might-be-done-about-it/
- Through Tech Goal: https://www.techtarget.com/searchenterpriseai/function/Power-consumption-of-AI-poses-environmental-problems
- by way of Tom’s Information: https://www.tomsguide.com/ai/chatgpt-energy-emergency-heres-how-much-electricity-openai-and-others-are-sucking-up-per-week
Right here is the abstract by AI:
Two stories reveal the substantial environmental influence of AI, significantly giant language fashions like ChatGPT. The vitality consumption for a single ChatGPT question is considerably larger than a regular Google search, resulting in elevated demand for information facilities powered by energy-intensive sources like pure gasoline. This surge in demand additionally ends in larger water utilization for cooling servers, with various consumption ranges relying on geographic location. Specialists predict that the issue might worsen tenfold by 2030 except effectivity enhancements are made.
And I discovered this factoid alarming and engaging:
A single ChatGPT-4 Question can use the equal electrical energy of powering a house for 20 days and eat as much as 3 bottles of water for cooling.
I keep in mind trying into this subject for ETMOOC2 again in April, when it was clear that the advances in AI was resulting in a surge in electrical energy and cooling. This diagram from Environmental Science & Expertise continues to be value inspecting.
And naturally, I had to make use of the AI podcast function in NotebookLM (which I discover has some latest updates — an possibility to present the AI audio software guiding factors to speak about). I requested it to look at how a lot of an influence on the atmosphere went in to creating ITS personal podcast. On the 7:27 mark, it does study its personal footprint, if solely briefly.
Right here is the 15 minute podcast.
I’ll finish by sharing what my AI inquiry says could be finished to mitigate the influence on the planet.
A number of methods can assist mitigate AI’s vitality footprint:
- Bettering mannequin effectivity: Growing AI fashions that require much less information and computational energy for coaching and inference is essential. This will contain strategies like mannequin compression and quantization to scale back vitality consumption throughout inference.
- Optimizing information heart infrastructure: Investing in energy-efficient information heart infrastructure, together with superior cooling methods and clever energy distribution, can considerably scale back vitality waste. Moreover, strategically finding information facilities close to renewable vitality sources or in areas with decrease environmental influence can assist decrease reliance on fossil fuels.
- Accountable information administration: Specializing in producing high-quality information as an alternative of accumulating large quantities of low-quality information can scale back storage wants and related vitality consumption. Often deleting unused information and selling information reuse and recycling can additional contribute to vitality financial savings.
So I assume I grew to become a part of the issue by trying into the issue with AI. Sigh.
Peace (and pondering the planet),
Kevin
