Task-Dependent Algorithm Aversion

Abstract

Research suggests that consumers are averse to relying on algorithms to perform tasks that are typically done by humans, despite the fact that algorithms often perform better. The authors explore when and why this is true in a wide variety of domains. They find that algorithms are trusted and relied on less for tasks that seem subjective (vs. objective) in nature. However, they show that perceived task objectivity is malleable and that increasing a task’s perceived objectivity increases trust in and use of algorithms for that task. Consumers mistakenly believe that algorithms lack the abilities required to perform subjective tasks. Increasing algorithms’ perceived affective human-likeness is therefore effective at increasing the use of algorithms for subjective tasks. These findings are supported by the results of four online lab studies with over 1,400 participants and two online field studies with over 56,000 participants. The results provide insights into when and why consumers are likely to use algorithms and how marketers can increase their use when they outperform humans.

Robo-investment aversion

“…Much evidence suggests that individuals should show an aversion towards computer algorithms, especially if it concerns decisions from the moral domain. When given the possibility to choose between advice provided by a human or an algorithm, people show a preference for the former and thus exhibit algorithm aversion (Castelo et al, 2019;Dietvorst et al, 2015Dietvorst et al, , 2016Longoni et al, 2019). Bigman and Gray (2018) have demonstrated via a series of experiments that decisions with a moral component appear to be the domain of humans and not "machines", showing machine-aversion in the moral domain, a tendency that was also documented by Gogoll and Uhl (2018).…”

Robo-investment aversion

“…However, we believe this is most likely due to the belief of some participants that algorithms produce superior outcomes, in accordance with reality [ 17 ]. Bias against algorithms re-emerges once participants are informed that humans and algorithms (in our case: an artificial neural network) produce similar outcomes, (presumably) taking away the algorithms’ perceived edge [ 16 ].…”

“…Recent evidence suggests that individuals show an aversion towards computer algorithms under certain conditions. When given the possibility to choose between advice provided by a human or an algorithm, people tend to display a preference for the former and thus exhibit algorithm aversion [ 16 – 19 ]. This algorithm aversion in the original sense can emerge in various contexts, but a common component is that people have to witness the algorithm making a mistake.…”

“…In contrast to the definition of algorithm aversion which is widely accepted in the literature [ 17 , 18 ], we do not link aversion to the success or failure of the algorithms in question. Our approach is conceptually related to Bigman and Gray’s [ 20 ] and Castelo and colleagues’ [ 16 ] findings on people exhibiting a generalized aversion to machines making decisions in tasks which require subjective judgments, in our case moral judgments.…”

Slow response times undermine trust in algorithmic (but not human) predictions

“…Some tasks, like image recognition for instance, are extremely easy for humans, but (currently) difficult for algorithms (Krizhevsky, Sutskever, & Hinton, 2012). Conversely, people may think that making a prediction is a relatively easy task for an algorithm, as it is an objective task involving the integration of multiple pieces of information or complex calculations (Castelo, Bos, & Lehmann, 2019). This view leads us to predict that slower response times will lead to lower quality evaluations of algorithm-generated predictions, as they will signal more effort being exerted for an ostensibly easy task.…”