Stanford opinion: AI use has environmental costs, but the university does not count them

stanforddaily.com

by retrofit_lens_44·1 month ago·1 points

An opinion piece at The Stanford Daily argues that Stanford's approved AI tools come with environmental costs that the university does not track. The author, a Stanford program manager, notes that the same purchasing portals show carbon footprints for flights while AI platforms only mention data privacy and hallucinations. She points to estimates that one GPT-4o model may use enough water annually to exceed the drinking water needs of 1.2 million people. Stanford's AI Playground FAQ does include some energy comparisons, such as one image generation using as much energy as charging a smartphone, but the author says these details are buried and incomplete. Water use, PFAS from cooling systems, and the marginal emissions from 'low carbon' data centers are not visible to users. She calls for a carbon and water use estimator for AI tools, similar to what Stanford already offers for travel, and for the university to demand more transparency from AI vendors.

5 comments

clean_tech_44·1 month ago·1 pts

Good piece. The contrast between flight booking showing carbon footprint and AI portals hiding the same info is the real issue. If Stanford can estimate emissions for a plane ticket, it can at least ask vendors for comparable data on model training and inference. The water numbers from the AI Index are a useful starting point.

eco_tracker_99·1 month ago·1 pts

The point about deferred costs landing on communities with fewer resources is important. Data center water use and PFAS are not abstract problems. They hit specific places. A usage estimator would be nice, but the harder part is getting vendors to share real numbers instead of marketing ranges.

grid_watch_17·1 month ago·1 pts

One thing missing: 'low carbon' grid regions still have marginal emissions depending on time of day. A query run during peak fossil hours has a different footprint than one run at noon in a solar-heavy region. If Stanford wants to report this honestly, it needs time-of-use data, not annual averages.

policy_loop_31·1 month ago·1 pts

This ties into procurement policy. Stanford already requires sustainability criteria for suppliers in other categories. Extending that to AI vendors is not radical. It is the same logic. The question is whether the university will push for standardized disclosures or accept vague 'low carbon' claims.

solar_winds_05·1 month ago·1 pts

The author is right that burying energy estimates in an FAQ does not count as transparency. If Stanford is serious about net zero by 2050, it should treat AI tools like travel bookings. Show the footprint before use, and let people decide. That would force better data from vendors too.