Authors:
Jonathan Bean
Herbert Simon famously wrote about design as being the act of devising "courses of action aimed at changing existing situations into preferred ones" [1]. In the world of technology, and in particular AI, it feels more like what's underway is the wholesale creation of a nearly infinite set of possible futures without regard to the consequences—human, environmental, social, or political. It's as though we are cooking pot after pot of AI spaghetti with the single-minded goal of throwing it against the wall to see what sticks. So quickly have we become busy cooking up one batch of spaghetti after another, indeed building vast machines, themselves powered by AI, to cook and throw the spaghetti for us, that we haven't noticed we're not throwing it against a wall, but instead onto other human beings and into a precarious natural world. The boiling water and steaming mess is only partly a metaphor; the data centers currently powering the AI revolution are famously rapacious users of energy, but they are also prodigious producers of heat. Some data center cooling centers dump the excess heat into the air; others move the heat in water, dumping the equivalent of millions of pots of boiling water into waterways or evaporating it into the atmosphere. Need I mention climate change?
At this point, I don't know where I'm going to land on AI, and I don't think anyone else can be sure of our AI future. One recurring idea is that AI will release the masses from the drudgery of work, unleashing higher human creativity and potential. That sounds nice, and I hope it works out that way. But how do we get there, and what is the price? We ignore history at our own peril.
We know—or seem to know—that data centers are spiking energy demand. That is hardly news, but what has surprised some is the apparent severity and speed of the spike in the U.S., where a perfect storm of demand has already arrived. Large incentives in the Inflation Reduction Act and Bipartisan Infrastructure Law have spurred companies to build energy-intensive manufacturing plants in the U.S. Those same policy mechanisms, combined with shifting consumer preferences, are reflected in a shift to energy-hungry electric cars. Utilities are struggling not only to keep up with the growth in demand but also simply to predict it. A recent news article sheds light on the severity of the problem. Amazon, the article reports, is building a server farm powered by a small nuclear plant, in part because it can't find enough power on the grid to buy. In Georgia, the most recent projection of electricity use over the next 10 years is 17 times higher than previous expectations:
It's as though we are cooking pot after pot of AI spaghetti with the single-minded goal of throwing it against the wall to see what sticks.
"When you look at the numbers, it is staggering," said Jason Shaw, chairman of the Georgia Public Service Commission, which regulates electricity. "It makes you scratch your head and wonder how we ended up in this situation. How were the projections that far off? This has created a challenge like we have never seen before" [2].
Power utilities in the U.S. occupy a difficult position in an economy otherwise dictated by market forces. They are granted a monopoly to operate in a given geographic area and are responsible for generating, transmitting, and delivering power to customers. But in some regions of the U.S., an overlying independent system operator, known as an ISO, coordinates generation and transmission on regional grids, keeping just the right flow of electrons headed where they are needed. The economic and material operation of the system—as well as the unique characteristic that electricity must be used at the moment it is created—means utilities are sometimes competitors and sometimes collaborators.
While some electric utilities are operated as nonprofit cooperatives and others are owned by municipalities, about a third of U.S. electricity is generated by investor-owned utilities, which operate under a legal obligation to return value to their shareholders. This means either selling electricity at a higher price (which can be complicated by regulation or the coordination of the ISO) or selling more electricity. For these utilities and others that make money by generating power, data centers should be highly attractive customers. They use large quantities of energy in patterns that are easily predicted; imagine someone who visits a drive-through three times a day and always gets the same order, supersized. But things are getting more complicated because the utility or the ISO both have to make the electricity and deliver it.
So what's happening is a kind of gold rush. As the adage goes, in Yukon Territory, it's better to be the person selling shovels than it is to be a prospector. The problem with this comparison is that in the gold rush, everyone was headed to the same place, so if you had the good sense to sell shovels, there were only a few logical locations—Skagway, Alaska, for example. But this is a distributed gold rush, where the "gold"—excess grid capacity—is spread across the entire electric grid without much logic or reason. To make matters worse, the topology of the network is highly time-dependent. What might be an excellent location for a server farm today might be a terrible one next week, especially if it turns out that your competitor has beat you to it.
In this way, the situation is more similar to the build-out of the railroads, but even that is not a great comparison. Yes, there were market forces in play; railroad companies competed against one another to complete vital links between major and emergent metropolitan areas. This is why cities like Chicago and Denver exist in their current forms. But even these efforts had some level of coordination; the federal government kept a firm hand on development around the railroad lines, and federal policy—underpinned by terrible violence, including the destruction of wild buffalo and the mass murder and forced removal of Indigenous peoples—distributed the benefits, however unfairly, between the railroad companies and a subset of homesteaders willing to move west and stake a claim.
Access to electricity is still missing for about 860 million people worldwide, including 15,000 people within the U.S.
This is where market mechanisms fall short: We naturalize the existence of the market, but too often forget how and, just as critically, why, particular market mechanisms came into being [3]. The form of the electric market is a response to regulation that recognized access to inexpensive, affordable, and reliable electric power as critical to improving quality of life. Electricity brought light, so people could read, learn, and work at night; refrigeration, so food could be kept fresh; and health and hygiene, in the form of pumped water, cleaner clothes, and cleaner homes. Yet access to electricity is still missing for about 860 million people worldwide, including 15,000 people within the U.S., many of those on the Navajo Nation [4]. And regardless of whether a utility is run as a for-profit or a nonprofit entity, the more electricity it sells, the more money it can potentially invest in infrastructure or reliability. Potentially is the key word in that sentence. Many believe that the California utility PG&E succumbed to the temptation to hand out big dividends to investors at the expense of basic maintenance, including fire prevention.
There are two questions that need to be examined. The first: How can we tease apart the provision of electric energy for basic needs (e.g., light, heating and cooling, and health) from AI-powered stuff that's nice to have, but, strictly speaking, not entirely necessary (e.g., asking Alexa to fart)? The costs of bumping the power grid up to power AI should not be borne by those who are already struggling to cover basic expenses. It may be shocking to learn that 20 percent of households in the U.S. have trouble paying their utility bill on time [5]. The second question: Are the demand predictions realistic? Over the past 30 years, electricity use in the U.S. has remained relatively flat, even with population growth, people living in bigger homes, and a proliferation of consumer technology. So the electric companies may be crying wolf. Why would they do this? Even though wind and solar generation is cheaper to build than fossil fuel infrastructure, the unpredictability of generation from renewable sources makes the utility's job more difficult. On the other hand, a natural gas plant can be switched on or off and its output dialed up or down as needed. This, combined with the mandate to provide reliable power, gives utilities an incentive to overestimate the growth in demand.
Technologists should be part of these conversations. Here are some questions for us to consider: Is it okay for cryptocurrency to be gobbling up as much energy as the entire country of Denmark [6]? How much electricity is it reasonable and responsible to devote to AI? What is the responsibility of the technology industry to pay for new generation and distribution upgrades, given that industry is driving demand for all of this new power? How can governance structures for the grid be tweaked to reduce the incentives for arbitrage?
The best we seem to be doing right now is playing a game of electric grid whack-a-mole. Regulation is slow to change by design, but that is no reason the tech industry cannot act proactively. As the wild frontier of AI continues to engulf us, big data's big impact on access to affordable power must be a central focus of conversations about ethics and equity.
References
1. Simon, H.A. The science of design: Creating the artificial. Design Issues 4, 1/2 (1988), 67–82.
2. Halper, E. Amid explosive demand, America is running out of power. Washington Post. Mar. 7, 2024; https://www.washingtonpost.com/business/2024/03/07/ai-data-centers-power/
3. Cochoy, F., Trompette, P., and Araujo, L. From market agencements to market agencing: An introduction. Consumption Markets & Culture 19, 1 (2016), 3–16.
4. Larson, A. Did you know there are 60,000 U.S. citizens who lack access to electricity? Power. Oct. 1, 2020; https://www.powermag.com/did-you-know-there-are-60000-u-s-citizens-who-lack-access-to-electricity/.
5. U.S. Department of Energy Office of Energy Efficiency & Renewable Energy. Decarbonizing the U.S. Economy by 2050: A National Blueprint for the Buildings Sector. Apr. 2024; https://www.energy.gov/eere/decarbonizing-us-economy-2050-national-blueprint-buildings-sector
6. Kohli, V., Chakravarty, S., Chamola, V., Sangwan, K.S., and Zeadally, S. An analysis of energy consumption and carbon footprints of cryptocurrencies and possible solutions. Digital Communications and Networks 9, 1 (2023), 79–89.
Author
Jonathan Bean is an associate professor of architecture, sustainable built environments, and marketing at the University of Arizona and codirector of the Institute for Energy Solutions. He studies taste, technology, and market transformation. [email protected]
Copyright held by author
The Digital Library is published by the Association for Computing Machinery. Copyright © 2024 ACM, Inc.
Post Comment
No Comments Found