Experimental Design

Participants were asked to complete a questionnaire (see Supplementary Material) assessing demographic and musical backgrounds and assigned to either an algorithm (ALG) or a fake partner (FP) condition to which they were blind. Questions addressed familiarity with hip-hop music (5-point Likert), preference for social functions of music listening (two items; both 5-point Likert), and quantity of music listened to per day (5-point scale ranging from <1 to ≥7 h/day). Participants were told that they would be creating three playlists that were to be used at a future virtual social event (i.e., “imagine you have plans to e-meet with some friends later this week, and they have asked you to make playlists you all can listen to while you hang out”). Participants in the FP condition were told that they would be making playlists along with another participant (in fact, a fake partner), whose name, age, and place of residence was matched to those of the participant and provided to the participant via a text-only prompt along with a generic non-moving digital avatar (i.e., gender/ethnicity non-specific). Participants in the ALG condition were told that they would be making such playlists on their own, but that they would have the assistance of MusicBot, a recommendation algorithm, to which a generic non-moving digital logo was assigned and provided to the participants.

The experimental structure involved trials each of one Playlist-Making session, one intermediate “series of prompts” step (i.e., intended to elicit either perceived interaction or no perceived interaction for participants in the FP or ALG experimental condition, respectively), and one Playlist-Listening session. Each participant took part in one trial for each of the three playlists. For each Playlist-Making session, participants were told to listen through a list of 10 song clips provided by the experimenter and to select, in any order, three of those clips to add to the playlist (i.e., based on those clips that the participant liked and thought sounded good together). Subsequent to each Playlist-Making session, participants were told via a series of prompts that either another participant (FP) or a song recommendation algorithm (ALG) had added three additional clips to each playlist (in reality, clip additions were random). Finally, participants took part in a Playlist-Listening session, during which they listened to the shuffled playback of their resultant playlist (i.e., the six songs that had resulted from their song clip selections from the most recent playlist-making session along with the FP/ALG's subsequent clip additions); participants in the FP condition were prompted to engage in joint listening behavior (i.e., “listen with [respective fake partner name] to the playlist you both have created”), whereas participants in the ALG condition were not (i.e., “listen to the playlist that has been created”).

Following the experimental trials, a recognition task was used in order to estimate cognitive self-other overlap, in which participants' memory of clips from Playlist-Listening sessions was measured. Participants were played a randomly selected song clip from the Playlist-Making sessions and prompted to identify whether the song clip had also appeared in a playlist during any of the Playlist-Listening Sessions. All clips had previously been heard by participants during the Playlist-Making sessions (see Supplementary Material for a recognition task trial example), but the first and last 500 ms of each clip were removed to control for immediate recognition (Schellenberg et al., 1999; Filipic et al., 2010; Belfi et al., 2018). Participants were prompted to respond using a keypress as to whether the song clip corresponded to one of their own previous playlist additions, one of the FP/ALG's previous playlist additions, or had not occurred in any of the previous playlists. A total of five practice trials and 25 test trials were administered. Finally, participants were prompted to answer self-report items assessing self-other overlap (i.e., “Which picture best describes your relationship with [fake partner name]/Musicbot?”; inclusion of other in self, via IOS) and domain-general indices of trait empathy relevant to the first hypothesis (i.e., perspective-taking subscale of the Interpersonal Reactivity Index; see Supplementary Material for exact items used). The experiment was conducted online and written using PsychoJS (Peirce, 2007) by the first author.