AI isn’t completely taking away jobs. I’m quietly chipping away at work, one task at a time.
That’s the point of a viral new research paper that refutes the idea that more exposure to AI automatically means fewer jobs. The authors argue that the real question is not how many tasks a model can perform, but whether those tasks can actually be partitioned without breaking roles.
Microsoft executives fear AI will eat up entry-level coding jobs
read more
Analysts have long warned that millions of jobs could be lost to automation. Recent projections indicate that the United States will lose 10.4 million jobs by 2030, about 6 percent of the workforce. The implicit assumption behind these numbers is straightforward. If the AI is good enough to do your job, you’re happy.
The new paper, written by Louis Gallicano, a professor at the London School of Economics, and Jing Li and Yanhui Wu of the University of Hong Kong, suggests that things are not so simple.
They argue that a job is a bundle, not a neat list of tasks. For example, radiologists do more than just read scan images. They interpret edge cases, discuss them with clinicians, and approve decisions for people to act. Replacing the image reading bits does not necessarily mean replacing the job.
The authors therefore draw a line between so-called “weak bundles” and “strong bundles.” Weak things can be divided without much fuss, but strong things cannot be divided without losing value.
“In weak-bundle occupations, AI automates some tasks and narrows job boundaries… In strong-bundle occupations… AI improves within-job performance but does not remove humans from the bundle,” the authors argue.
For weak bundle jobs, consider handling a large number of support tickets or removing predictable pieces of code. AI doesn’t just replace tasks. It restructures work. Humans end up doing things that machines cannot do, often in narrow parts of their original role.
Sounds like a win on paper. Actually not that many.
When AI takes over some of the work, humans will no longer be dividing their time. They do their best with what’s left. That is, output per worker increases rapidly, prices fall, and suddenly we don’t need as many workers as we used to.
In other words, the hit to jobs will come not from AI doing the job perfectly, but from humans becoming too efficient with the leftovers.
It also matches what we’ve seen so far. AI isn’t erasing jobs, it’s re-creating them. Tasks may move and productivity may increase, but employment and working hours have not changed much, at least not yet. Bundles are often still preserved.
It also explains why predictions of doom and technology optimism are both true at the same time. If you have a high-powered job, one that requires a lot of judgment, context, or responsibility, AI will likely help you get the job done faster and at a higher salary. If you are in a vulnerable position, your role can be quietly hollowed out until there is little left to protect. ®
