Slight control over an algorithm's output reduces algorithm aversion, with process control producing the strongest calibration effects¶

Claim · All claims
Evidence · 1 study (2 entries) · theoretical r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? The article reports, citing Dietvorst et al., that giving users even a slight degree of control over an algorithm's output was sufficient to reduce algorithm aversion substantially. → Pierre Haritz 2026 q2 i? The article reports, citing Cheng and Chouldechova, that both outcome control and process control can mitigate aversion, with process control producing the strongest calibration effects. → Pierre Haritz 2026 (2)

Evidence¶

Pierre Haritz 2026¶

Pierre Haritz, Hendrik Krone, and Thomas Liebig. (2026). Addressing Trust in AI Systems through Education: A Didactic Perspective. https://arxiv.org/abs/2609.02453 doi:10.48550/arxiv.2609.02453

q2 · i? · theoretical · r2

Narrative review attributing this experimental finding to Dietvorst et al.'s subsequent study on overcoming algorithm aversion. The article reports slight modification ability "was sufficient to reduce algorithm aversion substantially"; no effect sizes are printed.

"giving users even a slight degree of control over an algorithm's output, namely the ability to modify its predictions by a small amount, was sufficient to reduce algorithm aversion substantially"

Pierre Haritz 2026 (2)¶

Pierre Haritz, Hendrik Krone, and Thomas Liebig. (2026). Addressing Trust in AI Systems through Education: A Didactic Perspective. https://arxiv.org/abs/2609.02453

q2 · i? · theoretical · r2

Narrative review attributing this experimental finding to Cheng and Chouldechova's CHI 2023 comparison of outcome and process control. The article reports process control produced "the strongest calibration effects"; no effect sizes are printed.

"Cheng and Chouldechova extended this finding by distinguishing between outcome control (the ability to adjust outputs) and process control (the ability to influence how the algorithm works), and showed that both forms of control can mitigate aversion, with process control producing the strongest calibration effects"

Discussion¶

Related Claims¶

• Participants more averse to thinking are more susceptible to the speedup illusion; AI familiarity measures do not predict calibration error — related
• Seeing an algorithm perform makes people significantly less likely to choose it over a human forecaster — related
• Algorithm aversion persists even among participants who saw the algorithm outperform the human forecaster — related
• Confidence in the algorithm's forecasts mediates the effect of seeing it perform on choosing it — related
• People lose confidence more quickly in algorithms than in humans after seeing the same kind of mistakes — related