AI’s impact depends on which firms adopt it

When we discuss technological change and its effects on workers, we naturally focus on the newest, most advanced technologies. Headlines about how much better than intelligent humans the latest model from the leading AI labs performs in the mathematical Olympiad are eye-catching and send a slight shiver down the collective spines of policymakers and the white-collar class (or at least this economics professor).

This fear is understandable and, history suggests, warranted. The conventional analysis frames the effects of AI as a race between humans and machines, with advances in technology enabling machines to increasingly encroach on work tasks that were previously the exclusive domain of human workers. Of course, we humans have agency in this process, and we have proven adept at using new technologies to complement rather than replace our skills, and at dreaming up previously unimaginable products and services, employing entirely new classes of workers. In 2025, according to Revelio Labs, there are more than 60,000 social media managers in the United States, comparable to the number of steel and iron workers (about 65,000 according to the BLS).

Two important caveats apply. The first is that, even if it is true that humans have roughly kept up with machines over the long sweep of history, it has been a very bumpy ride at times. For better and worse, the Industrial Revolution shapes the world we live in today. Virtually all per capita improvements in living standards came after the Industrial Revolution, as did the carbon emissions and environmental damage we now confront. The Industrial Revolution also caused considerable inequality and civil strife, contributing significantly to the revolutionary wave sweeping Europe in the 1840s. If not yet on the same scale, ours is also an era in which the effects of technology and globalization have contributed to a significant economic and political backlash.

The second is that there remains enormous uncertainty about what AI will ultimately become. If AI develops into something approaching superintelligence, all bets regarding its impact on humanity are off. However, as Arvind Narayanan and Sayash Kapoor have recently argued, AI might instead represent a "normal" technological revolution, comparable to other episodes such as electrification.

My research can help understand the potential effects of AI in a world in which it is a revolutionary technology that replaces workers in many tasks, but in which we don’t necessarily all end up plugged into the Matrix. In such a world, how do we maximize the productivity benefits of AI while minimizing the amount of redistribution towards shareholders of AI firms?

In a recent paper, I study how automation technologies spread unevenly across firms. This uneven adoption creates a diverse landscape where cutting-edge automated firms operate alongside others using more labour-intensive methods. For instance, in the online grocery sector, highly automated fulfilment centres with advanced robotics operate alongside smaller facilities that rely primarily on human labor. In US manufacturing, there can be a six-fold difference in robotics adoption rates between the highest and lowest adoption firms within the same industry.

This variation in technology adoption matters tremendously for workers and the broader economy. When new entrants or existing firms adopt increasingly advanced automation, workers’ share of total firm income tends to fall, aligning with conventional wisdom. In contrast, when firms that are not operating with the most advanced technology automate more, something interesting happens: industry wages and workers’ share of income actually increase.

Why does this occur? The least technologically advanced firms in an industry are the least productive, and as a result, pay lower wages than more productive firms in the same industry. These unproductive firms set the wage floor for the industry, acting as an anchor on the pay of all other workers. As these firms improve their technology, that wage floor is lifted, so higher wages at the least productive firms benefit workers across all firms, not just those at newly automated companies. In the aggregate, this increases the economy’s productivity, because previously unproductive firms become more productive by upgrading their technology. Importantly, this also tilts the economic balance in workers’ favour, as firms with advanced technology now pay more of their income to their workers.

In contrast, when only frontier firms advance their automation capabilities, they pull further away from the rest, stretching the productivity distribution. This allows them to capture more profits, reducing labour’s share of income without necessarily raising wages for most workers.

To ensure that economies benefit maximally from the potentially amazing productivity benefits of AI, while preventing the balance of income from tilting yet further in favour of firm owners, we should seek to ensure that technology is as widely diffused as possible across firms. Firms, of course, should be incentivized to upgrade technology and become more productive. Still, experience teaches us that the barriers to technological adoption – financial, cultural, or regulatory – are high. There is a role for government in ensuring that policy is designed not just to foster innovation at the top but to make sure that AI is inclusive, in the sense that as many workers as possible are able to work with AI to increase their productivity, rather than confining the fruits of the coming technological revolution to a privileged few.

This article was first published in Forbes magazine on 2nd May 2025.