Top 10 Things I Disagreed With in the US Senate AI Insight Forum on the Workforce
I had the honor of speaking at the recent bipartisan U.S. Senate listening session on AI and employment, led by Majority Leader Chuck Schumer (D-NY) and Sens. Martin Heinrich (D-NM), Todd Young (R-IN), and Mike Rounds (R-ND). I commend them for using this format to solicit a wide range of views on the important issue of AI and AI policy. It was a pleasure to participate alongside an ideologically diverse group of representatives from organized labor, business, academia, and other think tanks.
My written remarks focused on why the job dislocation impacts from AI are likely to be very manageable and why AI-based automation will be critical to boosting workers’ wages. In listening to the other speakers, there were at least two broad themes I strongly agreed with:
▪ We need better policies and programs to help workers manage technological dislocation. The U.S. worker-training and adjustment system, to the extent it can even be called a system, underserves those who need it. ITIF has laid out a comprehensive agenda to improve it.
▪ Employer engagement with workers when AI is deployed can improve AI effectiveness. Workers often bring valuable insights to the deployment of new process technologies like AI. But after listening to some of the union reps, I worry that organized labor will oppose AI applications that eliminate jobs. This would not be in workers’ best interests. I am a big fan of Swedish labor unions. As one leader told me: “In Sweden, unions don’t fear new technology, we fear old technology.” Exactly.
On the other hand, there were many comments I did not agree with. Here are my top 10:
1. Automation lowers wages. Neither logic nor good academic literature support this assertion. It’s based on two myths. One is that automation will reduce the number of jobs, so workers will have less bargaining power. (See number 4 below.) The other—echoed by another panelist—is that if capital is used to eliminate work, then the price of labor will be less than the price of capital. No. Wages are not set by employers; they are set by labor markets. If they were set by employers, I’d cut ITIF policy analysts’ wages in half and use the savings to hire twice as many analysts. But if I did that, they’d all go to Brookings or AEI.
2. Technology has been the cause of inequality. This has been a common refrain over the last two decades, but it isn’t true. If automation was the big driver of inequality, then why did inequality decline during the 1950s and 1960s when automation-based productivity was at its peak, then grow after 1980 when productivity increased at less than half the rate? The Economic Policy Institute has found that inequality did not increase as a result of jobs in middle-wage occupations being eliminated by productivity gains. The fact that the top 25 U.S. hedge fund managers earned an obscene $26 billion in 2021 had nothing to do with tech-driven automation and everything to do with the institutional structure of the U.S. economy.
3. The pace of AI innovation is like a tsunami. AI has been around since the 1950s and has gone through regular periods of growth and stagnation. We are in a growth period, but past technologies (electricity, steel, internal combustion engines) have triggered more rapid change. Moreover, the rapid acceptance of ChatGPT has had much more to do with its fascinating customer experience and ease of use than any change in the overall rate of technology adoption. To be sure, AI and large language models are ushering in a new era of computing that could define the next decade or so. But that’s a pattern we’ve seen many times before, roughly every decade.
4. We run the risk of Great Depression-like unemployment rates. Wow! The reality is that the long stretch of high unemployment in the Great Depression was caused by terrible fiscal and monetary policy. In addition, nearly every study on technology-based productivity growth has found that it produces no increase in unemployment, for the simple reason that productivity lowers costs, which boosts wages and lowers prices, which in turn leads to more consumption, and hence more jobs.
5. There has been rapid adoption of automation technologies. The prevailing assumption is that since automation has been rapid and wage growth has been slow, AI-based automation must have no benefits. Actually, productivity growth rates over the last 15 years have been abysmal. This is in part because capital expenditures on automation have been relatively anemic.
6. The tax code favors capital, not workers. The notion that the tax code is stacked in favor of excessive automation is promulgated to get policies to slow automation. The reality is that while companies can immediately write off for tax purposes wages and expenditures on worker training, they must amortize over a number of years investments in machinery and equipment.
7. Employers have strong incentives to use AI to harm workers. With the unemployment rate near an all-time low, employers have at least some incentives to ensure their workers are not unhappy and to use AI to improve job quality, such as by reducing unpleasant, stressful, or dangerous tasks. This may be why the OECD recently found that most workers reported an improvement in job quality from the introduction of AI into the workplace.
8. Luddites were not antitechnology. As long as you don’t define destroying automatic looms with sledgehammers as antitechnology, we are in complete agreement.
9. Artificial general intelligence could be here in five years. Let me just quote MIT professor and Rethink Robotics CEO Rodney Brooks, who observed with respect to AI: “Misled by suitcase words, people are making category errors in fungibility of capabilities—category errors comparable to seeing the rise of more efficient internal combustion engines and jumping to the conclusion that warp drives are just around the corner.”
10. AI should be used to complement workers, not eliminate jobs. Unfortunately, this appears to have become the consensus view in Washington. But if this had been the historical American ethos, then we’d still have elevator operators, telephone operators, newspaper typesetters, and lots more gas station attendants, travel agents, and bank tellers. America became the richest nation in the world because we not only accepted automation; we embraced it. It would be a tragedy if the notion that it’s wrong for organizations to automate work becomes the norm, because it would be profoundly anti-worker for productivity and wages to stagnate.