What Do People Want From Algorithms? Public Perceptions of Algorithms in Government

Objective: This study examined how specific attributes of algorithmic decision-making tools (ADTs), related to algorithm design and institutional governance, affect the public’s perceptions of implementing ADTs in government programs. Hypotheses: We hypothesized that acceptability varies systematically by policy domain. Regarding algorithm design, we predicted that higher accuracy, transparency, and government in-house development will enhance acceptability. Institutional features were also expected to shape perceptions: Explanations, stakeholder engagement, oversight mechanisms, and human involvement are anticipated to increase public perceptions. Method: This study employed a conjoint experimental design with 1,213 U.S. adults. Participants evaluated five policy proposals, each featuring a proposal to implement an ADT. Each proposal included randomly generated attributes across nine dimensions. Participants decided on the ADT’s acceptability, fairness, and efficiency for each proposal. The analysis focused on the average marginal component effects of ADT attributes. Results: A combination of attributes related to process individualization significantly enhanced the perceived acceptability of the use of algorithms by government. Participants preferred ADTs that elevate the agency of the stakeholder (decision explanations, hearing options, notices, and human involvement in the decision-making process). The policy domain mattered most for fairness and acceptability, whereas accuracy mattered most for efficiency perceptions. Conclusion: Explaining decisions made using an algorithm, giving appropriate notice, providing a hearing option, and maintaining the supervision of a human agent are key components for public support when algorithmic systems are being implemented.