Source: Liu et al. (2026) — “Algorithm Aversion: Evidence from Ridesharing Drivers,” Management Science
Algorithm aversion is often described like a human flaw.
The model knows better. The human ignores it. Silly human. Please attend another training session.
But ridesharing makes the story more interesting.
Drivers live inside algorithmic management. The platform suggests, nudges, prices, routes, matches, rewards, and sometimes changes the rules in ways that feel like they were designed by a committee of spreadsheets.
So when drivers resist algorithmic recommendations, is that irrational?
Maybe not.
A driver has local context the platform may not fully capture. They know traffic quirks, customer patterns, neighbourhood risk, waiting spots, event timing, and the emotional cost of being sent somewhere inconvenient for a theoretically optimal trip.
The algorithm sees a lot. The driver sees differently.
That difference matters.
In platform work, algorithm aversion may reflect distrust, yes, but it can also reflect experience. Workers learn when the system is useful and when it is optimising for the platform more than for them. If the recommendation feels misaligned with their own goals, resistance becomes a form of self-protection.
This is why algorithmic management is so tricky.
The platform may think it is providing guidance. The worker may experience it as control. Same recommendation, different politics.
And when people feel controlled by a system they cannot question, they do not simply become obedient. They adapt, evade, second-guess, or build their own informal rules.
My takeaway: algorithm aversion is not always a trust deficit. Sometimes it is a signal that the worker sees incentives the model ignores. If platforms want better cooperation, they need more transparency, alignment, and respect for local human knowledge. Otherwise the algorithm becomes just another boss people learn to manage.