In a recent blog, I discussed Stanford University economist Erik Brynjolfsson’s new study showing that young college graduates are struggling to gain a foothold in a job market shaped by artificial intelligence (AI). His analysis found that, since 2022, early-career workers in AI-exposed roles have seen employment growth lag 13 percent behind peers in less-exposed fields. At the same time, experienced workers in the same jobs have held steady or even gained ground. The conclusion: AI isn’t eliminating work outright, but it is affecting the entry-level rungs that young workers depend on as they begin climbing career ladders.
The potential consequences of these findings, assuming they bear out, become clearer when read alongside Enrique Ide’s recent paper, Automation, AI, and the Intergenerational Transmission of Knowledge. Ide argues that when firms automate entry-level tasks, the opportunity for new workers to gain the tacit knowledge—the kind of workplace norms and rhythms of team-based work that aren’t necessarily written down—isn’t passed on. Thus, productivity gains accrue to seasoned workers while would-be novices lose the hands-on training they need to build the foundation for career progress.
This short-circuiting of early career experiences, Ide says, has macro-economic consequences. He estimates that automating even five percent of entry-level tasks reduces long-run US output growth by an estimated 0.05 percentage points per year; at 30 percent automation, growth slows by more than 0.3 points. Over a hundred year timeline, this would reduce total output by 20 percent relative to a world without AI automation. In other words: automating the bottom rungs might lift firms’ quarterly performance, but at the cost of generational growth.
This is where we need to pause and take a breath. While Ide’s results sound dramatic, it is critical to remember that the dynamics and consequences of AI adoption are unpredictable, and that a century is a very long time. For instance, who would have said in 2022 that one of the first effects of AI automation would be to benefit less tech-savvy boomer and Gen-X managers and harm freshly minted Gen-Z coders?
Given the history of positive, automation-induced wealth and employment effects, why would this time be different?
Finally, it’s important to remember that in a dynamic market-driven economy, skill requirements are always changing and firms are always searching for ways to improve their efficiency relative to competitors. This is doubly true as we enter the era of cognitive, as opposed to physical, automation. AI-driven automation is part of the pathway to a more prosperous economy and society for ourselves and for future generations. As my AEI colleague Jim Pethokoukis recently said, “A supposedly powerful general-purpose technology that left every firm’s labor demand utterly unchanged wouldn’t be much of a GPT.” Said another way, unless AI disrupts our economy and lives, it cannot deliver its promised benefits.
What then should we do? I believe the most important step we can take right now is to begin “stress-testing” our current workforce development policies and programs and building scenarios for how industry and government will respond should significant AI-related job disruptions occur. Such scenario planning could be shaped into a flexible “playbook” of options to guide policymakers geared to the types and numbers of affected workers. Such planning didn’t occur prior to the automation and trade shocks of the 1990s and 2000s with lasting consequences for factory workers and American society. We should try to make sure this doesn’t happen again with AI.
Pessimism is easy and cheap. We should resist the lure of social media-monetized AI doomerism and focus on building the future we want to see by preparing for and embracing change.
