The image of Chinese robots dancing like professionals has gone viral. It makes us wonder if machines will replace us in many tasks that humans currently perform. Artificial intelligence (AI) should boost productivity growth by automating the production of goods and services and, more surprisingly, the production of ideas and discoveries. The paradigmatic example is AlphaFold2, the AI model that "solved" the problem of determining the 3D structure of more than 200 million proteins from their amino acid sequence. This has transformed structural biology, reducing analysis time from months or years to minutes, and led to the awarding of the 2024 Nobel Prize in Chemistry to Demis Hassabis and John Jumper of DeepMind. In addition to facilitating imitation and learning, AI can improve itself in a feedback loop. It has evolved from "predicting the next word" to ever-increasing sophistication.
The question is whether machines—AI for cognitive tasks and AI combined with robots for physical tasks—will be able to perform many or most of the tasks currently done by humans. If so, it would lead to an extreme scenario where growth could be explosive, and we could say that this general technology would indeed have a different impact than previous ones (electricity, the internal combustion engine, airplanes, antibiotics, transistors and semiconductors, personal computers, and the internet). All these technologies have sustained per capita income growth of 2% since the end of the last century in the US, for example. Without their contribution, per capita income would very likely not have grown so steadily. However, their impact has not been immediate because new technologies need time to spread, and companies need to make the necessary organizational changes to adapt, overcoming production bottlenecks. Will AI and robots succeed in increasing per capita income growth, or will they simply perpetuate the historical trend of 2%?
What does the evidence tell us? Nobel laureate Robert Solow warned us back in 1987 that "you can see the computer age everywhere except in the productivity statistics." We see that AI is generating clear gains in specific tasks, but on a macroeconomic scale, productivity increases are moderate and uneven. AI can deliver large productivity gains in specific tasks: improvements in customer service performance, the speed and quality of professional writing, the output of intellectual work in consulting, and significant accelerations in specific software tasks. Moreover, US sectors where workers report that AI saves them the most time have experienced the greatest productivity growth. Overall, productivity gains are slow to appear, with low usage intensity and limited impact so far, but with higher expectations for future effects. The macroeconomic effects on productivity remain largely projections: OECD and IMF studies suggest that AI could add significant incremental, but not explosive, growth.
Will AI take our jobs? This will depend on whether the worker performs a job where AI is a replacement or a complement. The case of radiologists is interesting. It was predicted in 2016 that they would be replaced by AI, but it has turned out to be complementary, and the number and salaries of radiologists in the US have increased. Overall, it is to be expected that, while replacing certain tasks, AI will also create new ones for humans. In France, there is evidence that, for the moment, employment is not suffering.
AI and robots will bring about profound changes in the productive structure, with winners and losers. We see how the software sector (softwareThe world is in turmoil due to the transformations that AI can bring about. What is clear is that, for this technology to bear fruit, there needs to be sufficient competition among providers to maintain incentives to innovate and offer good service, and established lobbies and trade associations must not hinder its dissemination. Regulation must balance respecting fundamental rights and citizens' privacy with enabling innovation.
