Machine gaze in online behavioral targeting: The effects of algorithmic human likeness on social presence and social influence

Digital platforms increasingly use online behavioral targeting (OBT) to enhance consumers’ engagement, which involves using algorithms to “gaze” at consumers—tracking their online activities and inferring their prefer­ ences—so as to deliver relevant, personalized messages (e.g., advertisements, recommendations) to consumers. In light of the rising call for algorithmic transparency, this study investigates the effects of algorithmic trans­ parency on consumers’ experience of social presence and OBT effectiveness, when the OBT algorithm has low or high level of similarity to humans’ conscious mental processes. A one-factor, three-level (no transparency, vs. “observer” algorithm, vs. “judge” algorithm) online experiment with 209 participants was conducted. Results show that for individuals with low anthropomorphism tendency, the “observer” algorithm that did not form meaningful representations of consumers (i.e., low cognitive similarity to humans) reduced social presence, thereby compromising OBT effectiveness. The algorithm that “judged” consumers on meaningful dimensions (i. e., high cognitive similarity to humans) had no such effects.

Introduction. Interaction with algorithms is becoming a common part of con­ sumers’ experience with digital media. To boost consumer engagement, digital platforms (e.g., Facebook, Google, and Netflix) increasingly use online behavioral targeting (OBT)—the practice of delivering relevant, personalized messages to consumers based on the analysis of their online behaviors and sometimes other available metrics such as demographics and geographic data (Boerman et al., 2017; Nill & Aalberts, 2014). Essential to this practice is using machine learning algorithms to monitor consumers and make inferences about their characteristics and preferences, subjecting consumers to the algorithms’ “gaze,” an act involving not only seeing but also interpreting (Lacan, 1977). How do consumers respond to a machine’s “gaze”? How might such an experi­ ence further influence their evaluations of the algorithm-selected mes­ sages and persuasion outcomes of OBT?

Discussion / Conclusion. This study set out to understand consumers’ experience when being gazed at (i.e., being seen and interpreted) by OBT algorithms, with a focus on examining the conditions under which machine gaze generated the greatest social influence and why. Findings suggest that the simi­ larity between the algorithmic processes and humans’ social informa­ tion processing and one’s anthropomorphism tendency are key to OBT algorithms’ social influence on consumers. Findings have both theoret­ ical and practical implications. In conclusion, findings in the current study suggest that when the algorithmic processes of OBT are not disclosed, consumers by default use an anthropocentric mental model to guide their experience. When the algorithmic processes are disclosed, consumers’ experience with OBT and OBT effectiveness depend on the human likeness of the algorithms and their trait anthropomorphism tendency.