What will be AI's impact on human work? Luohan community offers insights.
While there is much buzz on AI's impact following ChatGPT, Luohan community members and presenters have long led research into technology's effects on labor
While AI has been the rage in the news recently, the impact of this powerful technology on jobs and on inequality has been an area of long-standing study for the Luohan Community, including some of the world’s leading scholars on technology and labor markets. Many have offered insights at our forums, reprised here. Featuring Avi Goldfarb, Anton Korinek, Chris Pissarides, Michael Spence, David Autor, Hanming Fang, and Joshua Gans.
Sir Chris Pissarides (LSE) and Anton Korinek (UVA, Brookings)
Keynes's big mistake is that he completely missed the development of the service economy… They were employing a tiny fraction of people. Today, those are employing most people. So if you look at the evidence, it's obviously against you. — Chris Pissarides
I think we are so used to the perspective, I want to call it the anthropocentric perspective, that only our human brains can do certain things. — Anton Korinek
Nobel laureate Chris Pissarides and Anton Korinek debated whether AI can replace most human labor within a reasonable timeline, during Korinek’s presentation to Luohan “Preparing for the (non-existent?) future of work” in November 2022.
Korinek believes the human brain is essentially an information processing device. He believes that advances in AI will mean that machines will have sufficient information processing power, and more importantly, the abstract models necessary to perform many current human cognitive functions. These may even include services that require human emotional understanding.
In contrast, Pissarides is skeptical that humans can be displaced. He cites the precedence of Keynes in 1931, who feared industrial automation would render humans unemployed only to miss the growth of the service industries. Pissarides notes the continued need for service jobs, and the human preference for creations and services by other humans. He does not imagine we can get to a jobless future within a reasonable time.
Avi Goldfarb (University of Toronto)
I think with AI, more people are going to win than lose… I'm optimistic about high-skilled versus low-skilled, more optimistic than you'd think, given what happened in computing over the past 40 years in the US labor market and Europe. — Avi Goldfarb
During a presentation on the big ideas in his book Power and Prediction in February 2023, Avi Goldfarb expressed cautious optimism about AI’s impact for most workers. Goldfarb believes that the skills bias will not be negative for average workers with AI, as it was with previous automation. In fact, AI may be able to upskill many lower to medium-range workers. For example, AI tools might allow nurses to do what doctors currently do in terms of diagnosis and treatment.
He does acknowledge possible risks of AI allowing a small number of people to make decisions at scale, concentrating income and power.
Ultimately, he feels we are at a very early stage of the AI revolution, we have yet to see the systems changes that are truly disruptive, and we lack good data or analysis to answer the questions on AI’s ultimate impact.
David Autor (MIT) and Hanming Fang (University of Pennsylvania)
Is AI going to do to a lot of professional work what office automation has done to a lot of clerical work? We don't know. — David Autor
Eventually society has become much more productive as a whole, but the distribution of that output is concentrated to a very small set of capital owners a small set of workers, then I think there might be more appetite for large-scale redistribution. — Hanming Fang
David Autor and Hanming Fang discussed the potential for AI disrupting professional jobs following Autor’s presentation at Luohan on The Origins and Content of New Work in July 2021. Autor’s rigorous work studying the labor markets have shown us both the labor market impact of automation and the evolution of new types of jobs. His broad conclusion is that, in the US, the last forty years have seen automation negatively impact a number of middle-income workers. Meanwhile, new jobs are being created as rapidly as in the past, but have often been either the low or high ends of the labor market, adding to economic inequality.
Both Autor and Fang acknowledge that AI will potentially have a large impact, but agree there is too much unknown about the nature of AI’s disruption. For Autor, the history of middle-income job displacement gives concern, and Autor and Fang both wondered if AI will displace high-income labor in professional jobs.
Fang mused about the need for a universal basic income (UBI). But Autor hopes the scenario requiring a significant UBI does not happen, because the political economy of that system will have negative consequences, one where “we're rich, but we don't have a natural means to distribute without forced redistribution.” Autor notes that despite some losers, we’ve been relatively lucky the past 80 years. Despite automation and efficiency gains, human labor has remained scarce enough for many people to enjoy jobs and rising incomes. Autor is cautious about the future, but believes one factor that helps is demographics. There are going to be fewer workers needing jobs going forward. And those fewer workers will be better educated, and perhaps can still find themselves scarce enough in the AI era.
Michael Spence (Dean Emeritus of Stanford GSB)
I think it's almost certain to produce a burst of productivity growth and reverse most of the recent trends we saw. It will show up well outside the bounds of conventional measures of economic performance in healthcare, biomedical science, the energy transition, education, and in inclusive growth patterns. I'm pretty optimistic especially for societies that invest heavily in human capital. — Michael Spence
In Luohan’s Frontier Dialogue #7 in September 2021, Nobel laureate Michael Spence offered his structured approach of looking at the impact of new technologies to give a sense of the impact of AI.
In Spence’s framework, digital automation is novel from prior rounds of mechanization. This automation has the potential to replace non-routine tasks.
Spence sees two phases: Automation 1 and Automation 2. Automation 1 targets tasks that cannot be easily “routine or codifiable” tasks, and much research such as David Autor’s showed this affected many middle-income jobs. Automation 2 involves much more advanced tasks, that may even include tasks impossible for humans to do. This are the AI technologies now being developed, which Spence see as “super powerful general purpose technologies across the economy.” Spence gives an example in DeepMind’s systems that can reasonably predict the 3D structure of proteins for medical research.
According to Spence, we’re not very far along Automation 2 yet, so it’s too early to tell how it plays out. We currently don’t know if this technology will have the same distributional impact on middle-income jobs as Automation 1. But what Spence is sure is that it will boost productivity across the economy. Societies that invest in human capital and other policies may be best positioned to capture this.
Joshua Gans (University of Toronto)
Everything we've talked about now is how you take a person trying to do their job. And they can do it 10 times better, or 10 times faster as a result of what these can do. Well, that makes people more useful, more valuable. There's always more stuff for us to do. — Joshua Gans
During a DiTalks chat with Luohan’s Randy Xu in January 2023, Joshua Gans shares his reasons for optimism that AI will not lead to a jobless future.
What underlies his views are several factors. First, Gans sees AI as a great equalizer for many. As an example, while ChatGPT may dilute the value of some writers, it will boost the productivity of others who have trouble with writing, including those working in non-native languages. So the net distributional effects may not be negative. Second, Gans sees human workers boosted with AI, which will makes humans more productive and thus, more valuable. “There’s always more stuff for us to do,” says Gans who uses the precedent of the spreadsheet which did not put financial analysts and accountants out of work. Finally, the current generation of AI technologies can process statistical information and take over some tasks, but cannot provide the intuition and drive that humans have. Humans will certainly need to adapt to and learn to use AI, but will retain irreplaceable advantages.