Can AI Make the Middle Class Great Again?

The robots are coming for our jobs! Well, that’s what everyone seems to think. Whether it’s Elon Musk warning about the end of human work or Geoffrey “godfather of AI” Hinton suggesting we all learn plumbing, the fear of a jobless future in the age of artificial intelligence is pervasive. In a 2023 Gallup poll, 75% of U.S. adults believed AI would lead to fewer jobs.

Ironically, this anxiety about employment comes at a time when the US has created an economy with one of the lowest unemployment rates amongst developed nations (and the lowest overall since modern times). The real issue isn’t the availability of work; it’s about the type and quality of those jobs.

AI might hold the key to improving job satisfaction, but only if it’s used in the right way. A new research paper by David Autor, titled “Applying AI to Rebuild Middle Class Jobs”, explores how Artificial Intelligence can do exactly that: restore the kind of middle-class, middle-skill jobs that have been hollowed out by decades of automation and globalization. Drawing on collaborative work with economists and technologists, Autor offers a new perspective: AI isn’t here to kill our jobs, but rather to democratize access to higher-value work, especially for those who have been left behind by previous waves of technological change. A different outlook, given the current discourse.

AI’s potential to democratize expertise

The paper starts with a simple but striking observation: the promise of the Information Age was that computers would democratize information and flatten economic hierarchies. Instead, digitalization has concentrated economic power and decision-making in the hands of The Elite, leaving (too) many workers in low-paid service jobs (and 1 in 20 in “useless” jobs)… with little chance of that changing any time soon.

Autor the author claims that AI presents an opportunity to create a new paradigm. Instead of replacing experts, AI can augment the abilities of non-experts with foundational skills, enabling them to perform tasks typically reserved for highly trained professionals like doctors, lawyers, or software engineers. Imagine a world where a nurse, guided by AI, can safely make decisions previously reserved for a physician (likely lowering health costs). This is the kind of future Autor envisions.

The methodology behind this claim involves both theoretical analysis and real-world case studies, like GitHub Copilot’s effect on programmer productivity (55% coding speed increase) and the impact of AI on customer service representatives. The paper argues that AI, when used well, can help workers perform higher-stakes decision-making, essentially restoring the middle-class jobs that automation eroded.

What AI could mean for the middle class

1. AI can be (should be!) an equalizer. Autor argues that “AI, if used well, can assist with restoring the middle-skill, middle-class heart of the U.S. labor market that has been hollowed out by automation and globalization.” Instead of creating more DoorDash “workers”, AI could create more middle-income jobs with decision-making responsibilities.

   
   

2. AI can extend the relevance and reach of human expertise, allowing workers with some foundational training to take on more complex roles. This essentially would bring roles traditionally monopolized by highly educated professionals to the masses (to some extent — we don’t want to get over-excited here).

   
   

3. AI complements human expertise. It does not replace it. In experiments with tools like GitHub Copilot, AI didn’t eliminate programmers’ jobs; it made less experienced developers nearly as productive as more experienced ones, closing the productivity gap. Interestingly, it did not fully bridge that gap, so real experts should have nothing to worry about.

   
   

4. The problem isn’t the number of jobs, it’s their quality. Autor notes that “The industrialized world is awash in jobs, and it’s going to stay that way.” But AI could help improve the quality and dignity of jobs, particularly for those without a college degree. No mention, however, of how AI would improve the dignity of an Uber driver…

How to harness AI for good

1. Let the government intervene and support training programs that help workers understand how to effectively use AI tools. A study by McKinsey found that Generative AI could enable labor productivity growth of 0.1 to 0.6 percent annually through 2040. But, as the paper suggests, AI is dangerous in the hands of those without the right foundation, like giving a pneumatic nail gun to an untrained DIYer. Proper training ensures workers can safely leverage AI to enhance their roles.

   
   

2. Regulate the AI market to avoid monopolies. The technology has the potential to democratize expertise, but it can also concentrate power if cornered by a few tech giants. Today, a majority of the AI market is controlled by just a few companies (you know the ones), which threatens the equitable distribution of AI benefits. Governments should ensure AI tools are accessible to all and not just to large corporations. For example, the European Union’s AI Act aims to regulate AI in a way that promotes transparency and prevents monopolization, creating a framework that could be replicated globally. Better yet… actively break up the likes of Google.

   
   

3. Support the creation of AI-augmented roles without burying everything in red tape. Policies should support the definition of roles where AI can complement human skills to perform tasks previously reserved for elite professionals. By adjusting regulations to allow mid-skill professionals to take on more responsibilities with AI support, we can increase access to good jobs. The American Medical Association has already suggested that the integration of AI into healthcare could improve patient care and allow healthcare workers to expand their roles, potentially addressing the shortage of medical professionals.

Where the paper falls (really) short

While the paper offers a hopeful vision, there are a few limitations.

First, there is an overreliance on the assumption that AI will always be used in a complementary manner. History shows that technology often leads to job polarization rather than creating new middle-class roles. The research could benefit from more empirical data on how exactly AI-driven roles can avoid the pitfalls of past technological shifts. Again, the poor soul delivering my food in the cold isn’t getting “upgraded” anytime soon (tip your delivery person y’all).

Moreover, Autor’s argument hinges on policy intervention. This is easier said than done in practice, especially in a polarized political environment.

One final element: companies that upgrade their workers with AI may very well not give them a raise and choose to reward shareholders instead. Instead of holding our breath for technology (which has systematically worsened inequalities), we could do the one thing that has improved inequality: higher taxes on the super-rich (100M$ is fine, but nobody needs 1B$).

A hopeful vision for AI and the middle-class

If we play our cards right, AI could enable more workers to do meaningful, high-quality jobs and reduce income inequality. This is not an inevitable future but one that requires active choices. Choices about how we train workers, regulate AI, and distribute the benefits of technological progress. Whenever and wherever you vote… remember that.

Good luck out there.