Source: Greiner et al. (2026) — “Incentives, Framing, and Reliance on Algorithmic Advice: An Experimental Study,” Management Science
Algorithmic advice sounds like a simple thing.
The model gives a recommendation. The human looks at it. The human decides whether to follow it. Very tidy. Very unrealistic.
In real life, people do not evaluate advice in a vacuum. They evaluate it inside a messy little room filled with incentives, accountability, fear, habits, and the possibility of looking foolish.
This paper studies how incentives and framing affect reliance on algorithmic advice. That is important because so many organisations assume adoption is mostly about accuracy. Make the model better, and people will follow it.
Maybe. But not always.
If the reward structure punishes visible mistakes more than quiet missed opportunities, people may avoid algorithmic advice even when it is useful. If the advice is framed as replacing judgment, people may resist. If it is framed as supporting judgment, the same recommendation may feel less threatening.
This is why two teams can receive the same AI tool and behave completely differently. One sees helpful support. The other sees future blame arriving early in a very polished interface.
This is very human.
People are not just asking, “Is the algorithm right?”
They are asking, “If I follow this and it fails, who owns the failure?” They are asking, “Will I look lazy?” They are asking, “Does using this make me smarter, or does it make me replaceable?”
Those questions do not appear in the model output, but they shape behaviour.
My takeaway: algorithmic advice needs incentive design, not just interface design. If leaders want people to use AI well, they need to align the social and organisational context around it. Otherwise the advice may be accurate, available, and politely ignored.