AI and the Labor Market: What IS Research Can Contribute

The economics literature on AI and employment operates at a level of abstraction that makes it hard to know what to do with. Acemoglu and Restrepo have built formal models showing how automation displaces tasks and whether new tasks emerge fast enough to compensate. Brynjolfsson and McAfee argued in "The Second Machine Age" that digital technology drives productivity but also widens inequality because the gains accrue unevenly. These are serious contributions. But when I try to use them to understand what is happening inside a specific hospital, or a law firm, or a logistics company when AI gets deployed, I keep hitting a wall. The macro framing does not tell me what happens to the nurse's role when a predictive deterioration algorithm takes over the monitoring task. It does not tell me how the radiologist's professional identity shifts when an AI reads the first pass of every scan. It does not tell me who in the organization gains power and who loses it when a dispatch algorithm replaces a human dispatcher.

That gap is where IS research lives.

The occupational vulnerability literature tried to bring this down to the task level. Research I have read on this, most closely associated with Frey and Osborne's 2013 work at Oxford, attempted to classify occupations by their susceptibility to computerization. My reading of that literature suggests that a substantial share of current occupations may be partially automatable, though the specific numbers from that study have been contested and I would not cite them as settled facts. What the task-level framing did accomplish, even with its methodological critics, was to move the conversation from "will AI take jobs" to "which tasks within jobs are exposed." That shift matters. It implies that most workers will not see their whole job eliminated. They will see their job reshaped, with some tasks automated and others remaining human. The interesting research question is not whether displacement happens but how organizations and workers navigate the reshaping.

IS researchers are positioned to study this directly. We study how organizations implement and use technology. We study how technology changes work routines, communication patterns, and decision-making structures. We have methods, ethnographic, survey-based, experimental, longitudinal, that can get inside an organization and trace what actually happens when an AI system goes live. That is not something a macro labor economist can easily do with national employment data.

The central theoretical question I keep coming back to is the complementarity question. Does AI augment human work or substitute for it? Economists frame this at the aggregate level: do AI investments create enough complementary demand for human labor to offset the tasks displaced? IS researchers can test complementarity at a much finer grain. In a specific organizational context, for a specific set of workers and tasks, does the AI deployment create new higher-order work that the workers can move into, or does it reduce headcount and compress the remaining roles? Benbya et al. (2021), writing in MIS Quarterly Executive, argued that AI in organizations is unprecedented in complexity and demands a thoughtful human-AI configuration. That is a research agenda, not just a finding. The configuration question, who does what, how human and AI contributions combine, how roles get redefined, is exactly what IS methods can answer.

The "missing middle" in the public debate is what bothers me most. The popular framing collapses into two camps. Either AI will automate most jobs and produce mass unemployment, or AI will create new jobs faster than it destroys old ones and everything will be fine, as happened with previous waves of automation. Both claims operate at a level of generality that obscures what is actually happening in specific organizations, industries, and labor markets right now. The missing middle is the organizational level: how AI deployments change skill requirements, reshape career ladders, redistribute decision authority, and alter the day-to-day experience of work. That is not the kind of thing you can read off national employment statistics.

Rai et al. (2019) framed the next generation of digital platforms as human-AI hybrids where AI complements rather than replaces humans, and where platform value emerges from the joint configuration. I find that framing more useful than the displacement frame, because it redirects attention from counting jobs to understanding configurations. But it is still a high-level framing. The IS research that would actually move this forward would study specific configurations in specific contexts. What tasks does the AI handle, what does the human handle, how do they hand off to each other, who has authority over the AI's outputs, what happens when the AI is wrong? These are IS questions. They require fieldwork or experiments inside organizations, not macro datasets.

There is also a power dimension that the labor economics literature underweights and that IS research is better equipped to address. When an AI system takes over a monitoring or scheduling or evaluation task that a human previously performed, the humans who designed and control the AI system gain influence over work that they did not previously touch. The worker whose judgment was previously authoritative becomes subject to algorithmic evaluation. This is not just a labor economics question about wages and employment. It is a question about power, professional identity, and organizational authority. IS research on algorithmic management, which I will write about separately, is beginning to take this seriously. But there is a lot of empirical ground still to cover.

The practical question IS researchers should be asking is not "will AI take jobs overall?" It is "in this organization, with this AI deployment, for these workers, what is the actual mechanism by which work changes?" That question has a different answer in every organizational context, and that variation is data. Studying it is what IS researchers do best.