Daron Acemoglu: The Limited Impact of AI on Employment

A few months before receiving the prestigious Nobel Prize in Economics in 2024, Daron Acemoglu published an article that generated little enthusiasm in Silicon Valley. Contrary to the promises of radical transformation of office work by AI put forth by tech giants, Acemoglu expressed doubts about the significant impact of AI on American productivity, estimating that it would only provide a slight boost. According to him, while AI can automate certain tasks, it will not replace the need for human labor. Acemoglu believes that AI is capable of automating some tasks, but that certain jobs will remain perfectly viable.

Two years after these statements, the discourse surrounding AI has evolved, with growing fears of an employment apocalypse. Political figures, such as a gubernatorial candidate in California, are even considering taxing the use of AI by companies to compensate for the layoffs it might cause. However, current data still supports Acemoglu's thesis, showing that AI has not yet caused major upheavals in employment rates. Discussions about an AI-related job apocalypse are emerging everywhere, from rallies with Senator Bernie Sanders to conversations overheard in supermarket lines.

Some economists, previously skeptical, have become more open to the idea that a seismic change could occur with AI. This reflects a shift in perceptions in the face of rapid technological advancements.

AI Agents: Complement or Threat?

Since Acemoglu's article, one notable development in AI has been the emergence of AI agents capable of operating autonomously. These tools are often presented as potential replacements for human workers. Acemoglu remains skeptical, viewing these agents as complements to human tasks rather than substitutes. He emphasizes the complexity of jobs, which often require a multitude of skills and adaptations that AI agents cannot yet replicate effectively.

The ability of agents to seamlessly manage the orchestration of human tasks remains a crucial question. AI companies are striving to demonstrate that their agents can operate autonomously without errors, but Acemoglu believes that as long as these tools do not master the transition between tasks, many jobs will remain safe from total automation. A radiology technician, for example, juggles 30 different tasks, ranging from taking patient medical histories to organizing mammography image archives. A worker can naturally switch from one format to another, from one database to another, and change work styles to accomplish this.

Recruitment of Economists by AI Giants

Another phenomenon observed by Acemoglu is the frenzy of recruitment of economists by AI companies. OpenAI, for example, has hired Ronnie Chatterji as Chief Economist and announced last year that Chatterji would work with Jason Furman, an economist at Harvard and former advisor to Barack Obama, to study AI and jobs. Anthropic, on its part, has gathered a group of 10 leading economists to conduct similar work. Google DeepMind recently recruited Alex Imas to lead the economics of AGI.

This trend reflects a desire among companies to influence the economic discourse surrounding AI, as public skepticism grows regarding its potential impacts on employment. Acemoglu expresses concerns about the objectivity of the research conducted by these economists, fearing that it may serve to promote views favorable to corporate interests. This dynamic raises questions about the integrity of studies on the impact of AI on work. "What I hope is that we won't see," Acemoglu says, "that they are interested in economists only to advance their viewpoints or fuel the hype."

AI Applications: A Challenge of Accessibility

Acemoglu also notes that AI, while ubiquitous in the form of chatbots, has not yet reached the ease of use of software that marked past technological transformations, such as PowerPoint for presentations and Word for documents. The creation of truly accessible and practical AI applications remains a major challenge.

He acknowledges that the impact of AI on the economy and the labor market is still uncertain, with often contradictory evidence. This uncertainty underscores the complexity of the AI economy and the need for ongoing and nuanced analysis. Even though anyone can converse with an AI model, it often takes time for an average worker to derive practical and productive use from it. This is partly why AI has not yet shown a seismic impact on the labor market or the economy.