Seeing an algorithm perform makes people significantly less likely to choose it over a human forecaster¶

Claim · All claims
Evidence · 1 study · causal r2 · q3
Currency · newest study 2014 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q3 i? Across four experiments manipulating experience with the model, participants who saw the algorithm's forecasts were significantly less likely to tie their bonuses to the algorithm's forecasts than participants who did not see it perform. → Berkeley J. Dietvorst 2014

Evidence¶

Berkeley J. Dietvorst 2014¶

Berkeley J. Dietvorst, Joseph P. Simmons, and Cade Massey. (2014). Algorithm Aversion: People Erroneously Avoid Algorithms After Seeing Them Err. Journal of Experimental Psychology: General. https://doi.org/10.1037/xge0000033

q3 · i? · causal · r2

Five lab and MTurk experiments in which participants forecast MBA success or state airline-passenger ranks and chose between betting on a statistical model or a human forecaster after seeing either, both, or neither perform. The article reports the effect was "highly significant, in all four studies in which we manipulated experience with the model."

"In every experiment, participants in the model-and-human condition were significantly less likely to tie their bonuses to the model than were participants who did not see the model perform."

Discussion¶

Related Claims¶

• Algorithm aversion persists even among participants who saw the algorithm outperform the human forecaster — a narrower finding that bears on this claim
• People lose confidence more quickly in algorithms than in humans after seeing the same kind of mistakes — related
• Confidence in the algorithm's forecasts mediates the effect of seeing it perform on choosing it — related
• A belief that the algorithm is less likely to be perfect does not explain experience-driven algorithm aversion — a narrower finding that bears on this claim
• Prediction algorithms forecast student achievement from small windows of data, more accurately when data are enriched with instructional design and learning-process detail — related
• Slight control over an algorithm's output reduces algorithm aversion, with process control producing the strongest calibration effects — related