Why trust an algorithm? Performance, cognition, and neurophysiology

A systematic review of algorithm aversion in augmented decision making

“…Though this too is a nuanced finding. In comparing the influence of social information (i.e., information about social norms) with that of algorithm‐related information, Alexander et al () used statistical information (“The algorithm is 75% accurate.” p. 281) as their algorithm‐related information. This assumes participants understand the need to tolerate probabilistic error, whereas using functional algorithm‐related information might have had a different effect if, say, it was to explain how the algorithm works in layperson terms.…”

“…This means approaching algorithm aversion as a project of behavior change in which hardwired organizational routines and social norms pose as major obstacles. A number of suggestions have been made along these lines: Choragwicka and Janta () suggest framing the benefits of algorithm utilization in relatable terminology, Alexander et al () propose manipulating the perceived social consensus, and Fisher (), Klimoski and Jones (), and Kuncel () advocate for localized reward schemes that apply to specific decision‐making roles in organizations. Each of these holds promise, and the most effective incentivization program is likely to vary by environment.…”

“…What manifests from these pre-existing expectations is a paradigm in which human decision makers perceive and respond to advice generated by algorithms differently than advice generated by humans, even if the advice itself is identical. Various mechanisms underlying this difference in response are demonstrated throughout the literature, such as the tendency for humans to seek a social or parasocial relationship with the source of advice (Alexander, Blinder, & Zak, 2018;Önkal, Goodwin, Thomson, Gonul, & Pollock, 2009;Prahl & Van Swol, 2017), the persistent belief that human error is random and repairable whereas algorithmic error is systematic (Dietvorst et al, 2015;Dietvorst, Simmons, & Massey, 2016;Highhouse, 2008b), experts' domain confidence leading to underutilization of seemingly unnecessary algorithmic aids (Arkes, Dawes, & Christensen, 1986;Ashton, Ashton, & Davis, 1994), or a lack of training preventing a human user from properly utilizing an algorithmic aid (Mackay & Elam, 1992; cf. Green & Hughes, 1986).…”

“…Existing research points out that organizational and social structures favor the expert intuiter over a cold algorithmic decision maker and incentivize accordingly (e.g., Alexander et al, 2018;Brown, 2015;Eastwood, Snook, & Luther, 2012;Highhouse, 2008b;Klimoski & Jones, 2008;Kuncel, 2008;Önkal et al, 2009). Brown (2015) and Hafenbrädl, Waeger, Marewski, and Gigerenzer (2016) argue that augmented decision making requires extra motivation because it involves combining multiple judgements rather than the acceptance of a single calculation.…”

“…Because of this, decision makers are motivated to conform to social norms, be it a professional maintaining an aura of omniscience in front of clients/ patients (Arkes, Shaffer, & Medow, 2007;Eastwood et al, 2012) or an employee seeking the support of management (Sanders & Courtney, 1985). In fact, information about others' algorithm utilization (i.e., social incentivization) is shown to have a greater influence than information about the algorithm itself (i.e., algorithmic literacy) on decision makers' engagement and performance with algorithmic aids (Alexander et al, 2018). Though this too is a nuanced finding.…”