Generating New Musical Preferences From Multilevel Mapping of Predictions to Reward

Abstract

Much of what we know and love about music hinges on our ability to make successful predictions, which appears to be an intrinsically rewarding process. Yet the exact process by which learned predictions become pleasurable is unclear. Here we created novel melodies in an alternative scale different from any established musical culture to show how musical preference is generated de novo. Across nine studies ( n = 1,185), adult participants learned to like more frequently presented items that adhered to this rapidly learned structure, suggesting that exposure and prediction errors both affected self-report liking ratings. Learning trajectories varied by music-reward sensitivity but were similar for U.S. and Chinese participants. Furthermore, functional MRI activity in auditory areas reflected prediction errors, whereas functional connectivity between auditory and medial prefrontal regions reflected both exposure and prediction errors. Collectively, results support predictive coding as a cognitive mechanism by which new musical sounds become rewarding.

Imaginings from an unfamiliar world: Narrative engagement with a new musical system.

“…Musical stimuli for this study consisted of eight recorded excerpts ranging from 20 to 90 s long, that were composed in the BP scale. These included four solo-clarinet compositions that were composed by our lab for another study (Kathios et al, 2022) and excerpts from four recorded music performances that were composed for a music festival on the BP scale in 2010 (Johnson, 2010). The recorded music performances were: Reminiscences by Steven Yi, Bohlen–Pierce Pan Flute World Premiere Folk Tune by Arturo Raffaele Grolimund, Hoquetus by Johannes Kretz, and Beyond the Horizon by Georg Hajdu.…”

The sound of manufactured music: Reviewing the role of artificial stimuli in music cognition research.

“…In their first study, Kathios et al (2023) found evidence supporting both mere exposure and predictive coding models of music preference. Specifically, participants' familiarity and liking ratings increased as a function of exposure and decreased as a function of prediction error (i.e., participants generally preferred melodies they heard more often in the exposure phase, as well as "standard" melodies over their "altered" versions).…”

“…A mini meta-analysis of Experiments 1-4 showed that both familiarity and liking ratings exhibited quadratic relationships with exposure (meaning that familiarity and liking increased as a function of exposure up to a certain point, after which subsequent exposures had less of an impact on these ratings). Again, Kathios et al (2023) took these overall findings as evidence primarily supporting a predictive coding account of music preference formation-the effect of prediction error (i.e., higher familiarity and liking ratings for standard over "altered" melodies) was present (and actually grew in magnitude) at high exposure levels, suggesting that participants' ability to accurately predict BP musical structure played a significant role in familiarity and liking judgments of all melodies they were exposed to, even when they had been previously exposed to some of those melodies more than others. Kathios et al' (2023) mini meta-analysis also allowed them to test another hypothesis relating to music learning, preference, and individual differences-does music reward sensitivity (i.e., the extent to which individuals report feeling rewarded and/or fulfilled through music listening; Martínez-Molina et al, 2016) impact the music learning/liking trajectories described above?…”

“…Again, Kathios et al (2023) took these overall findings as evidence primarily supporting a predictive coding account of music preference formation-the effect of prediction error (i.e., higher familiarity and liking ratings for standard over "altered" melodies) was present (and actually grew in magnitude) at high exposure levels, suggesting that participants' ability to accurately predict BP musical structure played a significant role in familiarity and liking judgments of all melodies they were exposed to, even when they had been previously exposed to some of those melodies more than others. Kathios et al' (2023) mini meta-analysis also allowed them to test another hypothesis relating to music learning, preference, and individual differences-does music reward sensitivity (i.e., the extent to which individuals report feeling rewarded and/or fulfilled through music listening; Martínez-Molina et al, 2016) impact the music learning/liking trajectories described above? As a first attempt in addressing this question, the authors split their meta-analytic sample (N = 667) into three groups (tertiles) based on their responses on the Barcelona Music Reward Questionnaire (BMRQ; Mas-Herrero et al, 2013), a commonly utilized measure of music reward sensitivity.…”

“…For example, a Western C major scale beginning on C4 (261.63 Hz; do in solfège), would progress through D4 (re), E4 (mi), F4 (fa), G4 (sol), A4 (la), and B4 (ti) before repeating C on C5 (523.25 Hz; do). This system creates scales that divide the interval of repetition (the octave) into seven units (in the diatonic version; 12 in the chromatic version, which would also include C#, D#, F#, G#, and A#) 6 and that include both odd-and even-integer frequency ratios 4 Two of these 25 sources are preprints (Kathios et al, 2023; which were brought to the attention of the author by a helpful anonymous reviewer.…”