Lav Varshney Discusses AI and the Physical Economy of Intelligence

Della Pietra Infinity Professor and Inaugural Director of Stony Brook University's AI Innovation Institute, Lav Varshney, led a discussion of the impact of AI at the Laufer Center. Photos by John Griffin.

Artificial intelligence transforms daily life a little more every day. Healthcare, business, science, the arts and every other discipline are facing new ethical and environmental challenges as they grapple with the new capabilities and possibilities that come with artificial intelligence.

In a far-reaching discussion titled "The Physical Economy of Intelligence: Marketcraft, Data Centers and Climate," Lav Varshney, Della Pietra Infinity Professor and Inaugural Director of Stony Brook University's AI Innovation Institute, discussed the puzzles, pitfalls and possibilities of an increasingly AI-reliant world.

The talk took place September 23 in the Laufer Center and was sponsored by the Collaborative for the Earth, the Department of Political Science, and the AI Innovation Institute.

Varshney said that despite the current national focus on data centers, the future of AI won't be determined by the infrastructure and neural networks themselves. The institutions and markets built around the technology will determine where AI goes, and whether it's a force for good or for ill.

"It will also determine whether we can sustain these efforts in building infrastructure and running models," said Varshney. "As we build markets for intelligence, we need to consider the economic and environmental repercussions and make sure that we can actually measure what's happening."

Varshney also cautioned about considering only the cost side of the economic equation.

Reuben Kline, associate professor of political science at Stony Brook and director of the Center for Behavioral Political Economy, with Lav Varshney.

"The premise of the economic argument is that the cost of AI is not going to be sustainable," he said. "Many are already working on more efficient approaches and more efficient models. But there's also a sort of Jevons paradox that perhaps usage will also increase, and that might be a counterbalance."

Jevons paradox is an economic theory stating that as technology improves the efficiency of using a resource, total consumption of that resource increases rather than decreases.

"As we look at large-scale AI, very large data centers are being built," he said. "If you look inside of those data centers, they have racks and racks of chips and graphical processing units that consume a lot of energy and require cooling. AI is very physical, it's not just a disembodied brain in the cloud."

While AI adds more load to the grid, it's also being used to make the power grid more efficient and more resilient, and has impacted virtually all industries.

"There's physicality on both sides, both in running these algorithms and also on their downstream impacts," said Varshney. "It's worthwhile to think about that balance."

Varshney made a parallel to the automobile and the automobile industry. "We don't want to just have 'tank-to- wheel' assessments," he said. "We're not just measuring the amount of gas that goes into using a car and the emissions that result. We need to think about all the prior steps upstream."

For cars, those steps include things like the extraction of petroleum, the refining process, the distribution and storage to fueling stations, and emissions and future disposal - all big industries unto themselves.

"There's a lot more to think about than just the car, and this is the kind of picture we want to keep in mind for AI, as well," he said. "If the downstream impacts are significant, they might be able to counteract the input requirements, so there might be an interesting balance from a sustainability perspective."

In the Q&A that followed, Varshney was asked about the paranoia of AI replacing not only jobs, but humans. He said he thought the discourse and questions around the labor impacts of AI are often too simplified.

"When you think of jobs, they're a collection of tasks. I probably do 30 different things in a day, and of those, maybe six are appropriate for AI," he said. "But counseling a student is not something I would want an AI to do. Or mentoring. There's some human element of externalizing knowledge that's required. So I think there will be a refactoring of jobs. The automatable parts will become automated, but the other 24 things that I do today will still be the professor's job, and maybe there'll be three new things as well."

Varshney acknowledged recent graduates have had trouble finding jobs, calling it a "temporary blip while organizations adjust."

"To use electricity as an example, it took 50 years for organizational and infrastructure changes to be made," he said. "That process is happening much more quickly with AI. I think within the next few years, we'll have those adjustments."

Attendee Marianne Reinhardt, managing director of Accenture, a leading global professional services company that helps organizations build digital foundations and use artificial intelligence to improve their operations, seconded Varshney's assessment.

"We are no longer hiring for the traditional skill sets," said Reinhardt. "We're hiring for deployed engineering, and all these other types of jobs are popping up based on the economy that is rearing because of AI and quantum and other technologies that are emerging. It's just fundamentally different. It's like when the personal computer came out and everybody's skill set had to be bumped up a bit. AI is quite revolutionary and different from that, but the same exact things are happening. Undergrads are coming out with certain skill sets, and smart people will adapt."

Varshney highlighted the importance of seeing the bigger picture as we build and invest in markets for intelligence.

"We want prices to reflect the scarcity of water, and we want to account for the effects on land beyond the purchase parcel - and as we're doing this, we want to make sure that we can actually measure what's happening," he said. "Market design is not just about prices and structures, but also information flow and predicting the response of participants. So all of these are the questions we're starting to struggle with, and I encourage all of you to help that struggle. Think about the cost of AI in producing and deploying intelligence, but also think about the returns in how AI changes the physical world. We can think about both sides of the ledger."

- Robert Emproto