Source: Bockstedt & Buckman (2026) — “Humans’ Use of AI Assistance: The Effect of Loss Aversion on Willingness to Delegate Decisions,” Management Science
I used to think algorithm aversion was mostly about trust.
People do not trust the machine. The machine feels cold. The machine cannot explain itself properly. Everyone nods, writes “change management” on a slide, and feels productive.
This paper makes the story more specific.
The issue is not only whether people trust AI. It is also how the decision is framed.
When people are rewarded for good performance, they may still prefer human help over AI help. But when the same decision feels like avoiding a loss, the psychology changes. Loss aversion makes people more sensitive to what could go wrong, and that changes whether they are willing to delegate to AI.
This is very human.
We like to pretend we evaluate tools rationally. In reality, the sentence around the decision matters. “AI can help you gain more” feels different from “AI can help you avoid losing money.” Same tool. Different stomach feeling.
That matters for AI adoption at work.
If a company introduces AI as a performance booster, employees may hear opportunity. If it introduces AI in a high-stakes loss context, people may hear risk, blame, and “if this goes wrong, who gets yelled at?”
Suddenly the resistance looks less irrational.
It may be a reasonable response to the emotional framing of the task.
My takeaway: adoption is not just a model-quality problem. It is a decision-design problem. If leaders want people to use AI, they need to understand what the decision feels like from the user’s side. Otherwise, even a good tool can arrive wearing the wrong psychological outfit.