Overcoming Algorithm Aversion: People Will Use Imperfect Algorithms If They Can (Even Slightly) Modify Them

By Berkeley J. Dietvorst, Joseph P. Simmons, Cade Massey. 2018

1,011 citations recorded by the source corpus.

Although evidence-based algorithms consistently outperform human forecasters, people often fail to use them after learning that they are imperfect, a phenomenon known as algorithm aversion. In this paper, we present three studies investigating how to reduce algorithm aversion. In incentivized forecasting tasks, participants chose between using their own forecasts or those of an algorithm that was built by experts. Participants were considerably more likely to choose to use an imperfect algorithm when they could modify its forecasts, and they performed better as a result. Notably, the preference for modifiable algorithms held even when participants were severely restricted in the modifications they could make Studies 1-3. In fact, our results suggest that participants' preference for modifiable algorithms was indicative of a desire for some control over the forecasting outcome, and not for a desire for greater control over the forecasting outcome, as participants' preference for modifiable algorithms was relatively insensitive to the magnitude of the modifications they were able to make Study 2. Additionally, we found that giving participants the freedom to modify an imperfect algorithm made them feel more satisfied with the forecasting process, more likely to believe that the algorithm was superior, and more likely to choose to use an algorithm to make subsequent forecasts Study 3. This research suggests that one can reduce algorithm aversion by giving people some control-even a slight amount-over an imperfect algorithm's forecast. Data, as supplemental material, are available at https://doi.org/10.1287/mnsc.2016.2643 . This paper was accepted by Yuval Rottenstreich, judgment and decision making.

Subfields: Artificial Intelligence, General Decision Sciences, Management Science and Operations Research

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In General Decision Sciences

In Management Science and Operations Research

Citations in the corpus

Automated rationale generation: a technique for explainable AI and its effects on human perceptions (2019, background): “For instance, contestability, the ability to either reject a reasoning or ask for another one, which has shown to improve user satisfactions [16, 22] can be incorporated in the future.”
User Perceptions and Trust of Explainable Machine Learning Fake News Detectors (2022, background): “When AI appears as a collaborator or with more agency, users are more likely to trust it and more willing to accept it (Dietvorst, Simmons, & Massey, 2016; Nguyen et al., 2018).”
Human Trust in Artificial Intelligence: Review of Empirical Research (2020, background): “For instance, Dzindolet et al. (2003) tested an automated decision aid and found that errors significantly decreased trust and reliance on the aid.”
Consumers and Artificial Intelligence: An Experiential Perspective (2020, positive): “For instance, offering users the possibility to correct an algorithm’s output, even if only slightly, is enough to increase their likelihood of using the superior, although imperfect, algorithm rather than the preferred, inferior human forecast (Dietvorst, Simmons, and Massey 2016).”
A systematic review of algorithm aversion in augmented decision making (2019, background): “Could the reduction in algorithm aversion caused by allowing human users to modify the algorithmic aid's output (e.g., Dietvorst et al., 2016) be fur-thered by allowing them to also modify the algorithmic aid's input?”
Artificial Intelligence in Utilitarian vs. Hedonic Contexts: The “Word-of-Machine” Effect (2020, background): “…circumstances resistance to algorithmic advice is attenuated: if people have the opportunity to modify algorithms and thus exert control over them (Dietvorst, Simmons, and Massey 2016), if the human likeness of algorithms is increased (Castelo, Bos, and Lehman 2019), if the task entails a numeric…”
Augmenting Medical Diagnosis Decisions? An Investigation into Physicians' Decision-Making Process with Artificial Intelligence (2021, background): “A second research stream in IS has suggested that distinct heuristic processes of human decision makers prevent adequate advice utilization (Xiao and Benbasat 2011, 2015, Dietvorst et al. 2015, 2018).”
Algorithmic bias: review, synthesis, and future research directions (2021, background): “These courses of actions or behavioural responses include refraining from accepting the algorithmic recommendation (Ebrahimi & Hassanein, 2019; Grgić-Hlača et al., 2019), engaging in algorithm aversion (Dietvorst et al., 2018), and refusing to adopt an algorithmic system (Lee, Jain et al., 2019).”
Transparency and trust in artificial intelligence systems (2020, background): “In follow-up work, B. Dietvorst et al. (2018) showed that people were more than twice as likely to rely on algorithmic forecasts if they were allowed to change a fraction of the AIs predictions compared to when they were bound what the AI predicted.”
Economics, Fairness and Algorithmic Bias (2020, positive): “Psychology has in fact moved towards documenting and understanding biases against using algorithms (or “algorithm aversion,” Dietvorst et al., 2015, Dietvorst et al., 2016) despite their superior performance.”