essay

AI will go quiet the way electricity did

Jeff Reynar
AI will go quiet the way electricity did

Kevin Kelly has compared AI to electricity for more than a decade. In a 2014 Wired essay, he wrote that AI “will enliven inert objects, much as electricity did more than a century ago. Everything that we formerly electrified we will now cognitize.” He predicted it would be “almost invisible except when it blinks off,” and that “like all utilities, AI will be supremely boring.”

His latest essay, Swarm Intelligence, puts a date on it: 50 years from now, most AI will be invisible. He pictures “zillions of tiny interacting AIs” forming something like an atmosphere, and writes that “AI is so prevalent we become blind to it.” Electricity’s history backs him up, and it also shows what companies had to change before the new power paid off.

When electricity was the topic

A hundred years ago, people talked about electricity the way we talk about AI now. They wired their homes and wanted the whole country wired, then bought electric irons, washing machines and vacuum cleaners while factories rebuilt around electric motors. Today electricity comes up at a party when the power goes out, or when someone’s arguing about a data center going up near their town. Electricity is now so boring and taken for granted that people would think you odd for bringing it up for almost any other reason.

AI is everyone’s topic right now: personal agents, what it’s doing to jobs, this week’s launches and startups, the power those data centers draw. In 50 years, AI will be just as banal as electricity is today, and cocktail party conversation will have moved on.

The payoff took forty years

The economist Paul David told electricity’s business story in a 1990 paper, The Dynamo and the Computer. In 1899, electric motors supplied under 5% of the mechanical power in American factories. They passed half in the early 1920s, four decades after the first central power station opened.

Early electrified factories used what David calls “group drive.” One big steam engine turned a long overhead shaft, with belts running down from it to every machine. Owners put electric motors where the steam engine had been and kept the shafts and belts. The gains came with “unit drive,” a motor on each machine. With the shafting gone, factories could be single-story buildings with skylights, and owners could lay out machines in the order the work moved. David found that growth in factory motor capacity accounts statistically for about half of the faster productivity growth in manufacturing during the 1920s.

AI on group drive

Kelly compares AI to electricity itself. Inside a company, AI works more like the motor, and where you put the motor matters.

A chat window off to the side, where someone pastes in a question, is group drive: one big motor, with a person acting as the belt. The work moves the way it always has. Someone reads the email, decides it matters, retypes the useful part into the CRM or a ticket, and assigns it.

Unit drive moves AI out of the chat window and into the work, with an agent at each point where work arrives. That’s Kelly’s swarm inside one company, many small AIs each doing one job, like reading a customer’s message, updating the CRM or scheduling the follow-up.

Every motor needs a power station

Factory electrification picked up after 1914 to 1917, David writes, when utility rates fell and central power stations came to supply more than factories’ own generators. He ties the spread of unit drive to new factories built where they could buy their electricity. Each machine had its own motor, and every motor ran on the same supply.

A swarm of agents needs a supply too, and for agents it’s what the company knows. An agent shipping a customer’s order needs their new office address, which showed up in last week’s email signature. An agent planning deliveries needs the date a supplier moved in a Slack thread. Those facts often show up in a message first, well before anyone updates a system of record. Kelly writes that these AIs will “store memory between themselves.” Inside a company, that memory is the power station: a brain every agent draws on, kept current from the messages the company already gets.

The analogy breaks in one useful place. Power flows one way, and a power station learns nothing from the machines it runs. A brain gets better as agents write back what they learn.

Kelly also writes that “No single institution, corporation, or government will be able to control the swarm environment.” Inside a company, the work arrives by email and in chat, through more than one vendor’s tools, so the brain has to draw from all of them.

What we’re building

this+that reads the messages a company already gets across Gmail, Outlook, Slack, Microsoft Teams, and Google Chat, as each one arrives. It turns them into tasks, keeps contacts current, and holds what it learns in a brain that workflows read from and write back to, with a person approving each update.

Look at where your company uses AI today. If it’s a window people paste into, you’re running group drive.