Theory of Machine: When Do People Rely on Algorithms?

Abstract

Algorithms--scripts for mathematical calculations--are powerful. Even though algorithms often outperform human judgment, people resist allowing a numerical formula to make decisions for them (Dawes, 1979). Nevertheless, people increasingly depend on algorithms to inform their decisions. Eight experiments examined trust in algorithms. Experiments 1A and 1B found that advice influenced participants more when they thought it came from an algorithm than when they thought it came from other people. This effect was robust to presenting the advisor jointly or separately (Experiment 2). Experiment 3 tested a moderator; excessive confidence in one’s own knowledge attenuated reliance on algorithms. These tests are important because participants can improve their accuracy by relying more on algorithms (Experiment 4). Experiments 5 and 6 tested a mechanism for reliance: subjectivity of the decision. For objective decisions, participants preferred algorithmic advice and for subjective decisions, participants preferred advice from people. Experiment 6 tested the interaction of subjectivity and the availability of expert advice. Participants preferred an expert to an algorithm, regardless of the domain (Experiment 6). Experiment 7 examined how decision makers’ own expertise influenced reliance on algorithms. Experts in national security, who regularly make forecasts, relied less on algorithmic advice than lay people did. These results shed light on the important question of when people rely on algorithmic advice over advice from people and have implications for the use of technological algorithms.