AI Is Coming for Your Job! Maybe

We have heard all about it. Generative AI (GenAI) tools such as ChatGPT, Gemini, Claude and the like have become so powerful that they can take over some of tasks that firms now employ people to do. They are a new kind of industrial revolution, except this time it is not craftspeople being replaced by machines, it is office workers being replaced by computing cycles. Unemployment will skyrocket and some occupations will be decimated or worse.

How true is this? This was the question Matteo Devigili, Erdem DogukanYilmaz, Vibha Gaba, and I investigated using data on hiring by major US firms and reported in a Management Science research paper. To answer the question accurately, we thought it was necessary to not just count jobs, but also look at the content of the jobs. Or rather, the skills that each job requires. This can be done by analyzing the text of the job postings firms use when they hire, which can be done automatically by using AI text analysis (yes, we appreciate the irony).

What were we looking for? Because we work for a business school, we care about the blend of activities that managers have and the skills they apply. These can be divided into broad categories: (1) organizing by creating and supervising the structure of the firm, (2) information processing by moving information to where it is needed, and (3) negotiating by handling conflicts and allocating rewards. Of these skills, the last is obviously the one requiring the greatest human touch, so it should be relatively human-centered. The first two might not be, or at least so we thought. What did the evidence say?

We were nearly right, and in ways that should worry managers. There is now lower demand for jobs doing managerial work involving information processing such as coordination and communication. Similarly, jobs engaged in monitoring employee productivity and rewarding high performers are now less needed. The managerial work that has seen no change is structuring the firm through staffing positions and mapping their interrelation, designing incentive systems, and handling conflicts within the firm.

This adds up to a significant reduction in firm hiring of managers. It also means a shift from the more mechanical parts of management to the more human parts: there will be proportionally fewer number crunchers and communicators and more leaders. How consequential will this be for the job market? Truth is, we don’t yet know. We found that firm adjustments of their hiring were so immediate that they made all these changes based on how they thought that AI would benefit the firm. If they are wrong about the AI benefits, they will adjust back. If they are right, then a new management job market has been created.

Devigili, Matteo, Erdem Dogukan Yilmaz, Vibha Gaba, and Henrich R. Greve. 2026. Skill Deprioritization: Reorganizing in the Age of Generative AI. Management Science, forthcoming.

Conspiracy Theory Life and Politics: Fear and Joy

Conspiracy theories have always existed, but we now live in world with more conspiracy theory believers and faster conspiracy theory spread than ever before. We know some of the reasons. Social media spread credible lies fast, even faster than they spread facts. Conspiracy theories are very effective political tools that can be used by dictators to influence uneducated voters in other nations. But it is not just the internet and politics – more is at play.

To find out, we set out to examine the spread of conspiracy theories related to the COVID-19 pandemic. The main questions were what made individuals start propagating conspiracy theories and whether they would get stuck in a conspiracy theory mindset or would move on. We used data from Twitter, which was a hotbed for conspiracy theories even before it became Xified. Our evidence was solid and not very encouraging.

First, we noted that many of the conspiracy theory spreaders were not regular people, but instead were bots programmed to manipulate others. This is normal in social media, and we used a strong tool for separating those from actual humans. Next, we found that regular people were driven by fear – the greater the threat from COVID-19, the more conspiracy theories they spouted. That included conspiracy theories saying that there was no such thing as COVID-19, it was all a lie. Believing in such a conspiracy theory and acting on it is exactly wrong when the infection rate is high. We are confident that this conspiracy theory killed people.

Equally disturbing, people were encouraged to continue propagating new conspiracy theories, or the same, by seeing their statements retweeted by others. In other words, the joy of seeing one’s conspiracy theory of COVID-19 becoming popular among others drove people towards more conspiracy theorizing.

It also drove them towards a variety of conspiracy theories, including some theories that cannot both be true. COVID-19 cannot both be a lie designed to keep people at home and a bioweapon designed by China, but the same people said both of these, often within a week of each other. It is true that they generally preferred similar conspiracy theories (see the graph above), but inconsistent conspiracy theories by the same people is a deeply troubling behavior.

In our data, conspiracy theories look like a form of reality denial. If the world presents people with information – real information – that is troubling to them, they can escape into conspiracy theories.

This should concern us greatly because there are many sources of fear, and many ways of manipulating conspiracy theories, including political reasons. For example, does the US have an affordability problem? Let’s find a conspiracy theory explaining it. And the conspiracy theory will distract people from addressing the problem, allowing it to persist.

Greve HR, Rao H, Vicinanza P, Zhou EY. 2022. Online Conspiracy Groups: Micro-Bloggers, Bots, and Coronavirus Conspiracy Talk on Twitter. American Sociological Review 87(6): 919–948.