This was cross-posted by the Forum team after the time that it was published.
In casual conversations about the future of AI—particularly among people who don’t go in for wild, sci-fi stuff—there seems to be a lot of attention given to the problem of technological unemployment: AI systems outcompeting humans at enough jobs to create a drastic, sustained rise in the unemployment rate.
This tends to be seen as a “near-term” problem, whereas the world-transforming impacts of AI I’ve laid out tend to be seen as more “long-term.”
This could be right. But here I’ll try to convey an intuition that it’s overstated: that the kind of AI that could power a massive productivity explosion and threaten humanity’s very existence could come pretty soon after—or even before! - the kind of AI that could lead to significant, long-lasting technological unemployment.
“Technological unemployment” AI would need to be extraordinarily powerful and versatile
The first key point is that I think people underestimate how powerful and versatile AI would have to be to create significant, long-lasting technological unemployment.
For example, imagine that AI advances to the point where truck drivers are no longer needed. Would this add over 3 million Americans to the ranks of the unemployed? Of course not—they’d get other jobs. We’ve had centuries of progress in automation, yet today’s unemployment rate is similar to where it was 50 years ago, around 5-6%.
(Temporary unemployment/displacement is a potential issue as well. But I don’t think it is usually what people are picturing when they talk about technological unemployment, and I don’t see a case that there’s anything in that category that would be importantly different from the daily job destruction and creation that has been part of the economy for a long time.)
In order to leave these 3 million people durably unemployed, AI systems would have to outperform them at essentially every economically valuable task.
When imagining a world of increasing automation, it’s not hard to picture a lot of job options for relatively low-skilled workers that seem very hard to automate away. Examples might include:
Caregiver roles, where it’s important for people to feel that they’re connecting with other humans (so it’s hard for AI to fully fill in).
Roles doing intricate physical tasks that are well-suited to human hands, and/or unusually challenging for robots. (My general sense is that AI software is improving more rapidly than robot hardware.)
Providing training data for AIs, focused on cases where they struggle.
Surveying and interviewing neighbors and community members, in order to collect data that would otherwise be hard to get.
Perhaps a return to agricultural employment, if rising wealth leads to increasing demand for food from small, humane and/or picturesque farms (and if it turns out that AI-driven robots have trouble with all the tasks these farms require—or it turns out that AI-run farms are just hard to market).
Many more possibilities that I’m not immediately thinking of.
And these roles could end up paying quite well, if automation elsewhere in the economy greatly raises productivity (leading to more total wealth chasing the people in these roles).
In my view, a world where automation has made low-skill workers fully unemployable is a world with extremely powerful, well-developed, versatile AI systems and robots—capable of doing everything that, say, 10% of humans can do. This could require AI with human-level capabilities at language, logic, fine motor control, interpersonal interaction, and more.
Powerful, versatile AI could quickly become transformative (“most important century”) AI
And then the question is, how far is that from a world with AI systems that can make higher-skilled workers fully unemployable? For example, AI systems that could do absolutely everything that today’s successful scientists and engineers can do? Because that sounds to me like PASTA (my term for a type of AI that I’ve argued could make this century the most important of all time for humanity), and at that point I think we have bigger things to worry about.
In fact, I think there’s a solid chance that PASTA will come before the kind of AI that can make lower-skilled workers unemployable. This is because PASTA might not have to match humans at certain kinds of motor control and social interaction. So it might not make anyone totally unemployable (in the sense of having zero skills with economic value), even as it leads to a productivity explosion, wild technologies like digital people, and maybe even human extinction.
The idea that we might see AIs fully outcompete low-skill humans in the next few decades, but not fully outcompete higher-skill humans until decades after that, seems intuitively a bit weird to me. It could certainly end up being right, but I worry that it is fundamentally coming from a place of anthropomorphizing AI and assuming it will find the same things easy and challenging that we do.
Bottom line: I think it’s too quick to think of technological unemployment as the next problem we’ll be dealing with, and wilder issues as being much further down the line. By the time (or even before) we have AI that can truly replace every facet of what low-skill humans do, the “wild sci-fi” AI impacts could be the bigger concern.
It seems possible that the reverse will happen. Frey and Osborne is probably outdated now, but it predicted that e.g. insurance underwriters are substantially more likely to be computerized than makeup artists.[1]
Arguably you can interpret this to mean that makeup artists are higher-skill than insurance underwriters, of course