Magic machines
Plenty of energy for washing and artificial intelligence
In 2010, Hans Rosling gave one of the best speeches ever.
Rosling’s pitch is simple–his worldview depends on washing machines. That is, it depends on people unlocking time with technology. He explains:
I was only four years old when I saw my mother load a washing machine for the very first time in her life. That was a great day for my mother. My mother and father had been saving money for years to be able to buy that machine, and the first day it was going to be used, even Grandma was invited to see the machine.
And Grandma was even more excited. Throughout her life, she had been heating water with firewood, and she had hand-washed laundry for seven children. And now, she was going to watch electricity do that work.
…
My mother explained the magic with this machine the very, very first day. She said, "Now, Hans. We have loaded the laundry. The machine will make the work. And now we can go to the library."
The numbers on population and energy use that Rosling relies on are dated by over a decade now, but the takeaways are not:
Energy is fundamental to flourishing.
Low-energy users want to move up the energy ladder to use more energy.
Rosling is no Pollyanna. His 9-minute speech covers the potential for washing machines to liberate humanity and the challenges of giving everyone washing machines. Rosling emphasizes replacing polluting energy with clean sources and improving efficiency. These are both stories playing out today.
Even more good news for today–new magic machines are on the horizon.
Artificial intelligence as a new magic machine
Rosling’s grandmother watched the washing machine because she saw it for what it represents. If you grew up with one, it’s a boring background piece. If you did the washing for a family of 9, then you see it as liberation. Hours of work vanish into the machine, and out comes opportunity. In Rosling’s case, that was time for reading and libraries. It could just as easily be time at the beach or freedom to pursue other projects, leisure or professional.
Rosling’s story is about the potential for artificial washing to replace the hours of labor required by washing in the past. If this story of artificial washing moves you, then you will be amazed at the potential for what meaningful improvements in artificial intelligence (AI) will bring to the world.
AI technologies can automate reactions and eventually make decisions within the bounds we set. Unlike artificial washing, AI is better at automating cognitive labor and will assist us in navigating the complexities of modern life. Hours of work will disappear again into these machines, creating space for doing more.
Of course, plenty of AI projects will fail, or at least fail to live up to their promises. Yet, those failures do not imply that we should give up on experimenting with AI. There is still revolutionary potential in AI. Every history of technological development includes failures along with successes. That the first planes did not fly did not imply flying was impossible. So we should expect bankruptcies and soaring valuations as markets sort bluster from potential.
There’s a lot of skepticism about this optimistic story of AI’s potential in public conversations and reporting. Unfortunately, a lot of that skepticism is more science fiction than a useful guide to policymaking.
Some of the poorest-founded objections surround the energy use of AI. These fears run headlong into Rosling’s underlying argument. Humanity progresses because we develop new tools that solve problems. The additional energy use those tools may require shouldn’t be seen as a burden nor wielded as a reason to stop technology rollouts.
The energy needs of AI are misunderstood
Take the data center buildout needs that are dominating so many news stories today. Data centers represent the backbone for developing AI systems and tools for all of us to enjoy. They’re also energy-intensive. Facilities regularly require more than 100MW–some sites rival the demand of entire cities. New AI data centers are searching for ten times that–a gigawatt or more instead of megawatts of energy.
Some of this growth is to be expected. After all, AI requires 10 times the electricity to power a use than a simple Google search. So, yes, artificial intelligence is going to increase energy use.
Will that increase overwhelm us? That fear seems incredibly unlikely. First, some historical context. There’s a long history of exaggerated stories of electricity demands from the technology sector. In 1999, computers were predicted to consume half of the US’s electricity supply by 2008. They actually consumed closer to 1% by 2008. These predictions have all turned out to be incorrect. They are orders of magnitude off the target given that computers use, at most, 4% of the country’s electricity demand today.
These failed predictions give us evidence to be as optimistic as Rosling is in his speech. Sure, environmentalists who think we must reduce energy use at all costs have reason to be concerned. But they are the ones that Rosling pokes fun at by accusing them of wanting to deny today’s energy-poor washing machines, flight, and the other accompanying benefits of modernity.
Second, there’s a misunderstanding of how much growth is really on its way. It’s far from unprecedented. Worries about the electricity demands of AI have disconnected from the concrete load projections that started them and given way to hype without context. The electricity load growth rate was revised in December 2023 to be 4.7% over the next five years instead of 2.6% over the next five years. Yes, that’s a doubling (almost) of expected growth. But it’s a projection that expects only a one percentage point growth each year for the next five years. Does that sound unattainable, unrealistic, or unbelievable? As Lynne Kiesling pointed out, historical load growth is not so different from these projections.
Robinson Meyer, editor at the climate-focused Heatmap news outlet, said it best:
There is no question that AI will use more electricity in the years to come. But so will EVs, new factories, and other sources of demand. America is on track to use more electricity. If that becomes a crisis, it will be one of our own making.
How are we going to avoid a crisis of our own making? A combination of getting better at running data centers and building the generation to power the next generation of magic machines.
The Jevons paradox is good, actually
Over the last few decades, we’ve gotten a lot better at running data centers. A 2020 paper in Science points out that data center electricity use grew only a tiny amount from 2010 to 2018. As the authors explain, electricity use for data centers globally increased by just 6%, but “...global data center compute instances increased by 550% over the same time period.” That’s a huge improvement in our access to computers without much of an increase in energy demands.
The study’s graphic is a bit clearer and illustrates how much better we’re getting. The three rows indicate different cases–the historical cases of 2010, 2018, and a hypothetical doubling of the use of data centers relative to 2018 electricity use.
As shown, that doubling of 2018 demand only looks like a significant change in the blue graph–the chart representing compute doubling. The electricity use barely budges.
How did we do this? Most gains come from getting more efficient in the computing itself along with getting better at cooling and running those facilities efficiently. For what it’s worth, the projection shown above predates today’s AI boom. That may put the projection on the optimistic side. It also assumes that current efficiency trends continue. There has been some reported stalling in these improvements that may affect that projection. Yet there have been dramatic improvements, like Nvidia’s Blackwell system.
Some environmentalists worry that these kinds of efficiency improvements will trigger the Jevons paradox. In short, the paradox points out that lowering the costs of something often means that you do more of that thing. More specifically, the paradox suggests that when we become more efficient at some activity, say, by using fewer inputs or lesser amounts of those inputs, we do more of that thing. In the case of energy and technology, lowering the energy use requirements of a technology encourages people to use that technology more often. This is often called a rebound effect as an energy-saving measure increases energy use by lowering costs.
The first response to this concern should be a shrug. What’s so bad about energy use? It’s great that we’ll use more energy. As Rosling suggests, don’t you think people without washing machines should have them? That people without easy transportion should be allowed to get moving? That someone who goes from pen and paper to a laptop gains much more opportunity because they use more energy? To some degree, the paradox gives rise to concerns that miss the point–doing more is doing better. Doing more with less is even better!
The second response should be pointing out how ridiculous the concern is in practice. We used more energy in 2018 than in 2010, and we’ll use more in 2030 than today. But the graphics from Science show just how much we can get even factoring in the rebound effect. Put another way, there is no widespread threat to your home’s power supply because of the onward march of data centers. Both of these statements can be true:
Yes, the rebound effect from efficiency is real.
No, the rebound effect in computing is unlikely to pose a meaningful environmental problem or risk crashing our energy system in the long run.
It’s great that we’ll use more energy
The use of Jevons paradox as a hammer to oppose developing new energy uses borders on nonsensical. Exploiting energy efficiency opportunities and building additional generation to power the data centers that serve as the backbone for artificial intelligence is great for humanity.
Despite the risks of triggering the Jevons paradox, data centers have seen massive increases in use with only a blip in actual energy consumption increasing. This kind of efficiency is great. We should be excited about the ways that we can do more with less because it enables us to do much more over time. Combining efficiency improvements with building additional generation makes powering data centers essentially a solved problem.
Given the improvements in efficiency, the ample supply of generation waiting to be added to the grid, and the expanding supply of low-carbon electricity, we have little to worry about from continued data center buildouts. Providing the fuel for artificial intelligence’s potential to blossom will be an insurmountable challenge only for those who object to additional energy use. Balancing those books will be impossible because their questionable argument is that the modern world has already far outstripped itself.
In a few decades, AI will be as unnoticed as your home’s washing machine
Hans Rosling’s optimism and clear thinking are infectious. As he plays out his grandmother’s excitement by sitting down in front of the washing machine on stage, it’s impossible not to be entertained and thrilled about the potential for new technologies to liberate people from drudgery. She was in awe that her grandchildren would never lose the same extent of their lives to washing as she had.
Before watching Rosling’s speech, I had never thought about washing machines as a tool for liberation. They had been a constant element in the background not worth my attention. In the same way, artificial intelligence is in the news today but will disappear into the background tomorrow.
AI shares the same fate. Humanity’s inventions first engender panic, then acceptance, and finally fade out of the conscious mind. After unlocking new levels of prosperity and giving us hours of our time back, they are forgotten. Like a washing machine or dishwasher, you’ll only notice them when they’re absent or broken.
AI, the next wave of magic machines, are more often the supervillains in science fiction tales than washing machines. Still, they will experience the same move from public awe and consternation into everyday items of no particular note.
Worth your time on AI, data centers, and load growth
A skeptic’s take on AI and energy growth, Jesse Jenkins and Robinson Meyer
Overblown data center energy concerns, Dustin Cote
Lynne Kiesling’s ongoing series on data center electricity use rolls on at Knowledge Problem
How to productively worry about AI energy use, Michael Giberson
Purpose, Pleasure, and Meaning in a World Without Work, Nick Bostrom and Russ Roberts
A Warning About AI from 1863, Pessimists Archive
A while ago, I gave a cell phone tour of my apartment to a Philippina gal I was courting. Of the items that I showed her, she was most taken by the washing machine. I had not really appreciated its importance before. I also found it interesting that she had a smartphone for our conversation but not a washing machine.