I asked ChatGPT to draw an image of the factory of the future:
This doesn’t look promising. I see at least two problems. The first is that the man in the picture is standing idle because all tasks have been automated. Although there is ample evidence that AI will not be able to automate all jobs anytime soon, it will make some workers’ jobs redundant while also creating new jobs. Here’s a figure from a recent OECD survey among firms adopting AI that shows the fraction of firms in which employment decreased, increased, or didn’t change:
To be clear, these are just correlations and not necessarily the causal impacts of AI on firm-level employment. But let’s assume that AI is decreasing employment in some firms while increasing it in others. This means that displaced workers in firms that decrease employment must find other jobs, perhaps in the firms that increase employment because of AI. Transitioning into new jobs takes time and can be very taxing financially as well as mentally. Economists have tried to quantify these transition costs from automation, and estimates are about 1% of GDP per year. That’s a huge number. The best way to minimize this welfare loss is to maximize the fraction of firms that increase employment when adopting AI. The second problem that the ChatGPT image reveals is that, despite all the equipment, it seems that nothing is being produced. I’ve stared at the picture for a very long time, but I couldn’t figure out what is being made here. So not only is the marginal product of labor zero (i.e. the worker standing idle), but average labor productivity (i.e. turnover divided by employment) is also zero. This figure shows annual productivity growth rates for several advanced economies since 1900 in reality:
In Europe, productivity growth has declined steadily since 1960. Why this is happening is somewhat of an enigma. Why doesn’t the 1990’s technology boom or all the inventions since then show up in the productivity statistics? The famous Draghi report about European competitiveness points to the same problem. Can AI bring back productivity growth? There is causal evidence from some experiments that it could. On the other hand, the figure above shows a long-run decline in economy-wide productivity growth, and it seems unlikely that AI will be able to revert this trend. If our main challenge is to use AI to create goods and increase productivity growth, what can we do? First, we can invest in workers. This figure shows that the biggest change for firms when adopting AI is the need to train workers:
Learning workers how to use AI is likely to increase labor demand and productivity. But if this is true, why do we see persistent skills shortages? Why do we underinvest in workers’ skills? One reason is that it is unclear what skills exactly are needed. We can’t expect every worker to learn code, but what other skills are complementary to AI? Another reason is that it is unclear who should pay for investing in workers’ skills. Employers might not want to invest if workers can leave and apply their skills elsewhere, and workers might not want to invest if skills are firm-specific and labor contracts are temporary.
Second, given the pervasive nature of generative AI, merely investing in workers’ skills will not be enough to create good jobs and increase productivity growth. We should also rely on Europe’s strong social dialogue. A crucial difference between European and US labor markets is that Europe has a stronger tradition of consultation between employers and workers through, among others, work councils and collective agreements. This figure shows that in firms that adopt AI in consultation with workers do better in terms of productivity and job quality than firms that adopt AI without consulting their workers:
To be clear, this is just correlational not necessarily causal. However, there seems to be a consensus among European policy makers that the social dialogue can be a means to adopt pro-worker AI. The challenge will be to support the social dialogue with a useful framework to do so, without overregulating firms to support productivity growth.