AI adoption actually reduces the productivity of novices while making experts significantly more powerful.
April 23, 2026
Original Paper
Limits to Human-AI Collaboration: AI as Skill Multiplier?
SSRN · 6628867
The Takeaway
An expertise threshold exists where people below a certain skill level cannot distinguish between correct and flawed AI outputs. Novices end up wasting time fixing errors or accepting bad advice, which lowers their overall output. Experts use the same tools to automate tedious tasks and multiply their existing knowledge. This contradicts the popular belief that AI will be the great equalizer for the workforce. It suggests that AI will actually widen the gap between the masters of a craft and the beginners.
From the abstract
Does AI democratize expertise or amplify skill gaps? Using a Bayesian information acquisition framework, we show that when AI unreliability is high, there exists an expertise threshold below which AI adoption reduces productivity. Above this threshold, domain knowledge acts as a filter that allows users to extract value from flawed outputs at low cost, while novices face verification barriers that negate efficiency gains. This formalizes conditions under which AI amplifies rather than equalizes