Holden Karnofsky recently posted an article on the Open Philanthropy Project’s blog giving some background about the OPP thinks about AI. One paragraph in particular stuck out at me—one of several examples of what Holden calls “transformative AI”:
AI systems capable of performing tasks that currently (in 2016) account for the majority of full-time jobs worldwide, and/or over 50% of total world wages, unaided and for costs in the same range as what it would cost to employ humans. Aside from the fact that this would likely be sufficient for a major economic transformation relative to today, I also think that an AI with such broad abilities would likely be able to far surpass human abilities in a subset of domains, making it likely to meet one or more of the other criteria laid out here.
My first reaction to this is that this might not be as big of a transformation as it might seem. For example, according to the BLS, between 1910 and 2000, farmers and farm laborers went from 33 percent of the workforce to 1.2 percent. Many “industrial” jobs peaked around 1950, then declined to well below their 1910 level—the awkwardly named “production and other operatives” category (mostly factory workers AFAICT) declined more than 50% in that time period, while “mine operatives and laborers” declined by a whopping 95%. Furthermore, the proportion of clerical jobs in the labor force has greatly increased, since 1910 but improvements in information technology has greatly reduced or even eliminated the need for many sub-categories of clerical jobs.
I think most people realize that the decline in farm jobs is not because we suddenly decided to let huge amounts of land fall fallow, but because we learned to farm the same amount of land with a lot fewer workers, through harvesting machines and so on. Something similar is true in manufacturing. In spite of the widespread assumption that the decline in manufacturing jobs in America is because all the jobs have been shipped to China or Mexico, in fact industrial production in America is as high as it’s ever been. This is thanks to the incredible rise of automation in factories.
Overall, the data I’m seeing from Measuring Worth is that there was more than a seven fold increase in real (i.e. inflation-adjusted) GDP per capita from 1910 to 2000. Increases in GDP per capita must ultimately come from increases in worker productivity. In other words, in the year 2000 workers were getting more than seven times as much done with the same amount of work (probably somewhat less work, actually). Some of that probably came from things like improvements in industrial organization, but a lot of it undoubtedly came from more and better machines, including computers.
If you described the technology of the year 2000 to a person in 1910, it might sound to them like technology was going to eliminate huge numbers of jobs. But we largely haven’t noticed, beyond vague background awareness of harvesting machines and those weird-looking factory robot arms, because the jobs that got eliminated by technology got replaced with other jobs. Engineering, healthcare, law, computers, education, sales, and food service are just some of the occupations that have seen massive growth in employment.
The example of “transformative AI” described above would be a big deal—equivalent to over three decades of economic growth, using the US’s rate of economic growth as a baseline. But I’m not sure it would qualify as being as important as the industrial revolution (another rough definition Holden suggests for “transformative AI”). Holden suggests there’s a 10% chance of humans developing “transformative AI” in the next 20 years. If 50% of all currently-existing jobs were eliminated by automation in the next 20 years, then in 2036 we’d all be talking about the amazing worldwide economic boom of the 2020s―but that would be the main effect.
There’d be fretting about job loss due to automation—but probably various government policies would do an okay job of making it possible for displaced workers to move to new jobs. When I say “government policies”, I don’t just mean feel-good stuff like more funding for job training, I also mean boring issues of central bank policy. Maybe unemployment creeps up, prompting the US Fed to try negative interest rates for the first time, while in Europe anger over unemployment finally causes the Eurozone to break up, which leads to a period of painful turmoil followed by an economic boom one the European labor market is no longer shackled by ECB short-sightedness.
The one catch is the word “unaided” in the quote from Holden. Imagine a scenario where a huge portion of the labor currently done by humans is done by semi-autonomous robot drones, and the drones have remote operators, but they’re autonomous enough that one remote operator can manage a fairly large number of drones. Maybe the AI that Holden is imagining goes beyond even “one drone operator can manage a fleet of 100 mostly-driverless trucks”. But I’m not sure how much the difference between that scenario and true 100% automation vehicle automation matters in terms of the impact on society. (At least when many other jobs are still being done by humans—automating the last few human jobs could be a big deal.)
The “human gets replaced by robot” model of automation is nice and tidy and easy to reason about. But in the real world I’m not sure there’s always a clean line between jobs being eliminated and jobs being radically transformed.