What Will Be Left for Humans to Do After AI?
I’ll get to the future of humanity shortly, but first I want to discuss reaction to a very recent event in the history of artificial intelligence and in
Not long after a bunch of its “agents” escaped and freaked everybody out, OpenAI had some good news to report: A largely autonomous swarm of its “agents,” it announced, had succeeded in solving a math problem I and most other humans are unable even to understand, a math problem the world’s best mathematicians had, to date, proved incapable of solving.
It was known as the “Navier-Stokes problem”: and was one of the “Millennium Problems,” “a collection of important unanswered math questions,” as Cade Metz explained in the New York Times. Those problems were “meant to push the world’s leading mathematicians to new heights.”
In this instance, those “new heights” ended up being attained not by humans but by artificial intelligences.
The fact that AI, not a human, had apparently solved that Millennium Problem did not sit well with Terence Tao—a UCLA professor who is regarded by many, says the Times, as the finest (human) mathematician of his generation. Prof. Tao is quoted in that Times article as comparing AI to “a really clever student who has memorized everything for the test but doesn’t have a deep understanding of the concept.”
Indeed, AI’s solution to that Millennium Problem caused Prof. Tao to use another simile that was more problematic and more revealing than he had probably intended.
I am going to try to outline the problems with that simile and, in the process, both question the logic employed by the finest (human) mathematician of his generation and explain how humans (all of us, not just mathematicians) might be able to carry on in a world that will likely be populated by intelligences greater than our own.
Bear with me. This won’t be easy.
Here is Prof. Tao’s reaction, as quoted by the Times, to the AI solution to the Millennium Problem : “The effort needed to solve problems is often very instructive. It teaches you something,” he explained. “It’s like going to the gym and having a goal to lift a weight a hundred times…Now, AI can solve questions without really getting any value out of them.”
And here is the revealing simile Prof. Tao used: “It’s like having machines that can lift weights for you at the gym.”
In some sense what Prof. Tao is saying here is obvious: Yes, the fact that artificial intelligence has solved a problem will deprive the best non-artificial (a.k.a. human) mathematicians of the invigorating challenge of trying to solve it themselves. They are deprived of the mental workout. And, of course, AI solving problems at this level raises the unsettling possibility that mathematicians, too, could become redundant.
Prof. Tao is noting, also correctly, that putting effort into solving problems like this, without the crutch of artificial intelligence, is a valuable and productive—“instructive”—exercise for a mathematician’s brain. Can’t argue with that, either.
However, things then get more dicey. For lifting weights at the gym would seem an odd choice as an example of the non-artificial, productive way of doing things, since the gym itself is an entirely artificial, unproductive substitute for all the natural and highly useful ways our ancestors had, over time, to stay in shape: hunting and gathering, chopping wood, lugging logs, clearing land, planting and harvesting, constructing abodes or, more recently, walking to a store to buy some groceries.
We don’t get anywhere when we run on a treadmill: no new food sources are tracked down, no bountiful lands are discovered. Lifting weights does not help clear a field or build a wall. The gym seems more about repetition than “deep understanding,” to use the professor’s phrase. The gym’s benefit is entirely restricted to what the exertion does for the person doing the exerting. Nothing else gets accomplished.
So, Prof. Tao’s perfectly reasonable point—that humans get considerable benefits from trying to solve problems—is partially undercut by his gym simile which would seem to reduce humanity’s efforts to solve tough math problems, even before AI entered the picture, to something equivalent to riding a stationary bike.
The late 20th-century philosopher Jacques Derrida made his reputation by “deconstructing” these sorts of self-contradictory similes and metaphors. Derrida might have noticed how in this case Prof. Tao contrasts artificial intelligence with a more natural form of thinking by comparing it unfavorably to going to the gym—itself an unnatural way of staying fit.
Derrida thought such tangles in our thinking are revealing.
And I am going to argue that ace-mathematician Terence Tao’s somewhat self-contradictory gym metaphor reveals something about the future we humans are facing. For there seems to be a pretty good chance humankind’s future will feature—for worse or (possibly?) for better—less productive work and more gym-like exercises.
I want to call as a witness a fellow from the last third of the 19th century, when human beings were just embarking on the road to redundancy. His name: John Henry.
Mr. Henry was a “steel-driving man,” probably African-American, possibly apocryphal, who is the subject of an extraordinarily well-known folk song. John Henry made holes for the explosives used to blast out railroad tunnels by hammering in steel spikes. But steam drills were just starting to replace humans in digging those holes. John Henry, a proud man, decided, according to the song, to compete with a steam drill to prove that he could beat the machine.
And the human won. John Henry created those holes faster than the steam drill. But, as you may remember from the song, the effort killed him.
And, of course, the steam drill would get better and better at digging holes for explosives. Men would not. The machine, soon enough, made off with—automated—that job.
And the parade of technologies that replaced humans continued in the 19th century and picked up speed in the 20th: steam shovels that outperformed men with shovels, for example; or desktop computers that greatly reduced the need for secretaries and typists.
And the great jobs churn has continued in this new century: disrupting all sorts of careers. Computer programmers—somewhat ironically—are among the first workers to find their jobs threatened by artificial intelligence.
Surprisingly but happily, so far, over these centuries—the economy has continued to produce new jobs to replace those lost to automation. These have been, in many cases, less physically demanding jobs. So, John Henry’s great-great-grandchildren—perhaps working as AI engineers or math professors—would probably have joined Planet Fitness to stay in shape.
It is possible, however, to imagine that the new jobs won’t keep coming, in the rapidly approaching age of artificial intelligence—potentially the most skilled and multi-talented job destroyer humankind has yet invented.
Indeed, OpenAI’s announcement that a swarm of its agents had solved one of the “Millennium Problems” would seem to indicate that there may be no job too intellectually difficult for artificial intelligence to swipe.
And let’s not underestimate how painful it is for someone who, like Terence Tao, was really, really good at some productive, challenging and satisfying work, only to be bested at that work—bested by some brand-new thing: an entity that has been able to read up on, more or less, everything you and, more or less, everyone else knows, that therefore seems likely to be very good at, more or less, everything.
John Henry refused to concede to the machines.
It looks like we may have to.
Some form of “universal basic income” might solve the problem of how people might live without enough paid work. It would not solve the problem I am interested in here: how people will find enough productive, challenging and satisfying things to do.
What we may have to settle for—to occupy our time—is unproductive but challenging and sort of satisfying work: the cognitive equivalents of the physical gyms Prof. Tao was touting: mental gyms providing activities that provide mental exercise.
Humans might very well choose to solve problems, even math problems—problems that have already been solved by other humans or by AIs—just to stay in mental shape.
They might practice speaking a foreign language—even if AI is there to translate any language into any other language.
They might try to get good at competitive games like chess or bridge, or individual games like Wordle or Spelling Bee—while remaining aware an AI could beat them at any of these games.
Humans might write essays, stories or poems, even though an AI could write them much faster and, perhaps, better.
Yes, there is something sad about humankind’s great intellect potentially being limited to puzzles, games and literary exercises, just as there is something sad about humans running on a treadmill.
But maybe it will do.
This essay first appeared in Mitchell Stephens’ Substack, Ideas on Ideas.