Paper Understanding When Laypeople Adopt Predictive Algorithms

Predictive algorithms have become widespread decision-support tools across domains including marketing, healthcare, and finance. As their adoption has grown, so too has research examining how non-expert users interact with and choose to use these systems — variously described as artificial intelligence, models, or decision aids. Despite a rich and expanding literature, findings can feel fragmented and at times contradictory, with studies documenting both resistance to and enthusiasm for algorithmic tools depending on context. This paper proposes a cohesive framework for understanding the conditions under which laypeople adopt predictive algorithms. The framework is grounded in the premise that predictive algorithms are fundamentally performance tools, adopted primarily to improve decision outcomes, reduce effort, or both. Accordingly, users' expectations about algorithmic performance are identified as the central driver of adoption decisions. A key insight of the framework concerns how laypeople define and evaluate performance — which differs meaningfully from expert standards. Research suggests that non-expert users exhibit diminishing sensitivity to prediction error: they perceive the difference between a perfect prediction and a near-miss as far more significant than an equivalent difference at higher error levels. As a result, laypeople tend to evaluate algorithms by how often they produce near-perfect predictions, whereas experts typically rely on aggregate error metrics that capture average performance across cases. By synthesizing diverse findings around this performance-centered logic, the framework offers a practical foundation for researchers and practitioners seeking to understand, predict, and design for algorithm adoption among general audiences — while acknowledging that no single framework can fully account for the breadth of factors involved.

Get the Paper