Hey Reader,

This week, my students and I have been talking a lot about the climate effects of AI.

The footprint of AI usage has been a big issue for years – and it's become an even more common question because of viral headlines claiming that AI "drinks a bottle of water" to generate even a small amount of content.

And it's about more than just quantifying resource usage. In our new post in Innovating with AI Magazine, author Amy Smith explores how grocery stores can use AI to optimize their resource usage – while also risking of accidental discrimination and unfairness.

Amy's article is a fascinating look at how grocery stores are using AI across pretty much every aspect of their businesses, from sourcing to aisle design. Here's a sampling:

To read more, click here for the full, free, shareable article from Innovating with AI Magazine.

•••

For the second half of today's newsletter, I've got a sneak peek into what I talked about with students in our most recent live training session, which was all about celebrating the great things about AI while being honest about its negative impacts.

By far the most popular headline of the last few months is this one:

This is a very catchy headline 🙂. I did some digging, though, and it seems like the data is a mix of:

1. Outdated (models are more efficient now than a few years ago)
2. Incorrect about how modern data center cooling works (less water is needed than the original study assumed)

To support the contention that the original study is misleading, I shared Andy Masley's detailed post about comparing ChatGPT environmental impacts on to other common behaviors. His takeaways are pretty striking – particularly this comparison of water usage between ChatGPT, watching television, and eating a hamburger.

(If you ever wondered why "stop eating beef" is one of the cornerstones of climate policy, this chart shows you why. Chicken and turkey are much less resource-intensive.)

The thing is, we could debate until we are blue in the face about whether people "should" use AI or not. Ultimately, every action we take that uses resources is something that we could choose to modify or omit, reducing our environmental footprint. So every decision – whether eating or driving or using AI – has tradeoffs.

I think the good news in all this is that the AI industry's incentives are generally well-aligned with the goal of reducing the environmental footprint of AI. For example:

• Building solar-powered datacenters makes everything cheaper, since solar is rapidly becoming the cheapest form of energy.
• Making datacenter cooling more efficient also makes everything cheaper, and locating datacenters in places with plenty of water, like the Midwest and Southeastern US, makes "water scarcity" a moot point.
• Making models and computer chips more efficient saves money for everyone.
• Cheaper per-token pricing is better for everyone.

In other words, AI companies and datacenter companies benefit from cheaper per-unit pricing, and consumers do as well. So consumers and companies are aligned here.

Contrast that to the oil and gas industries, where companies are absolutely not in alignment with consumers. For example, because of the expenditures required to drill for oil, prices need to be around $60 a barrel to even make it economical to drill in the United States. That means that lower oil prices actually make production impossible – pretty much the opposite dynamic that happens with lower AI token prices. On top of that, there is crazy cartel behavior among foreign oil producers, so the big players are always manipulating prices in ways that are unrelated to consumer benefits.

In other words, the incentives for Google and Meta's data centers are not the same as the incentives for notorious polluters, like the Big Oil companies.

The Big Oil companies benefit when prices go up (harming consumers). Their public image is already pretty much as low as it can go, so there's minimal benefit to making themselves seem more "green." And they are working hard to lock consumers in to more oil purchases – for example, by lobbying against electric vehicles.

By contrast, the companies building AI data centers benefit from higher efficiency and lower pricing, since there's a nearly limitless demand for more AI usage. No one needs to lock consumers in – we'd use AI even more if it were cheaper and more plentiful. And these companies care deeply about their reputations and operate in a highly competitive space where any of their rivals could steal market share at any moment.

To sum it up – AI is not harmless by a long shot, but over time, it seems like the combination of greater efficiency and cleaner power solutions will make it a clear win when you consider the costs against the benefits.

•••

📚 Here's what else I'm reading this week:

• ​Can Pittsburgh’s Old Steel Mills Be Turned Into an AI Hub?​ (WSJ, gift link)
• ​A.I. Is Making Sure You Pay for That Ding on Your Rental Car​ (NYT, gift link)
• ​Elon Musk Updated Grok. Guess What It Said?​ (The Atlantic, gift link)
• ​Indeed and Glassdoor to lay off 1,300 workers as AI shakes up job search business​ (CBS News)
• ​Google hires Windsurf CEO Varun Mohan, others in latest AI talent deal​ (CNBC)

•••

🙋‍♀️ Want to be part of our next survey?

Until next time,

– Rob