Horse Manure, Transistors, and AI
The data centers that sustain ChatGPT, Claude, Gemini, and other generative artificial intelligences are increasingly hungry for electricity. A recent analysis by the Electric Power Research Institute estimated that data centers’ usage of US power could leap from 4.5% today to as much as 17% by 2030. Reading that figure, I wondered if AI will be so constrained by its demand for energy that its development stalls.
The impending AI bottleneck prompts two questions in my nonartificial mind. How were similar bottlenecks overcome in the past? History offers one guide to the future. Science fiction may offer another. Can we find answers in futures imagined by writers?
As the industrial revolution played out, cities became bigger, and the need for horse-powered transport burgeoned. Paradoxically, overground and underground railways boosted the demand for horsepower because of the so-called last-mile problem: Whereas crates of ice-packed fish could be rushed by train from fishing ports to cities, horse-drawn vehicles conveyed the fish from stations to markets and on to fishmongers.
The Great Horse-Manure Crisis encapsulated the problem. The name comes from a purported prediction in 1894 in the Times of London, which stated that “in 50 years, every street in London will be buried under nine feet of manure.” The prediction turned out to be an urban myth invented in 2004. Yet, it captured the anxiety about a genuine problem. A single draft horse produces astonishing amounts of waste every day.
The rapid and widespread adoption of the internal combustion engine saved the day. By 1911, the number of cars, buses, and trucks on London’s streets had surpassed the number of horse-drawn vehicles.
Three decades later, telephone companies faced a different bottleneck. Each exchange relied on thousands of electromechanical relays to connect and route calls. And long-distance calls required fragile, bulky, inefficient vacuum tubes to amplify signals to offset transmission losses. The American Telephone and Telegraph Company tasked its research arm, Bell Laboratories, to develop a compact solid-state alternative. The result was the transistor.
Science fiction may offer a complementary perspective on overcoming technological bottlenecks. Scottish writer Iain Banks set most of his science fiction novels in the “Culture,” an egalitarian society of humanoids coexisting with benevolent, hugely capable artificial intelligences called Minds. The Culture is what mid-20th-century futurists termed a postscarcity society: One in which technology has become so advanced that goods, services, and energy are virtually free. To sustain the Culture, Banks concocted hyperspace substrates, infinite quantum swapping, energy-grid tethering, and sentient electromagnetic fields.
Among the Culture’s artificial intelligences are drones. Despite their modest size—the smallest are no larger than a tennis ball—their processing power far exceeds that of their humanoid companions. They do not need data centers to perform complex calculations for split-second decisions: In Use of Weapons (1990), one drone rapidly computes and deploys force fields and plasma charges to destroy a squad of attackers threatening its humanoid companion.
With a few taps on his keyboard, Banks leapfrogged over bottlenecks that might have challenged the Culture’s scientists and engineers, allowing him to explore how humans may coexist with stupendously capable artificial intelligences. His fictional solutions remind us that today’s engineering constraints need not be tomorrow’s.
So what technologies might make AI’s energy bottleneck seem as quaint as London’s horse-manure problem? New energy sources such as fusion power are contenders. But they don’t address the manure-scale problem: Current generative-AI systems are remarkably inefficient. The answer clearly won’t be hyperspace substrates or infinite quantum swapping, but it could—and should—be a new form of AI that does far more with far less energy, perhaps by borrowing tricks from the human brain.
The energy I consumed to write this article came from a sandwich.
—Charles Day
Charles Day is a Senior Editor for Physics Magazine.