Christine Bauer and Markus Schedl, researchers at the Department of Computational Perception, Johannes Kepler University Linz, propose an approach to expanding the utility of recommender systems by developing capabilities to incorporate surprise and opposition, that is, opposing perspectives, in recommendations. These are nontrivial challenges if not wicked problems in design thinking [1] that have some appeal, if for no other reason than to counter the original definition of the Netflix effect: scanning long lists of movie recommendations, but not finding anything to watch.

Part of the challenge addressed here is defining surprise and opposition, a complex task that seems much like efforts to define beauty (“in the eye of the beholder”) and obscenity (“I know it when I see it”) [2].

Another part of the challenge tackled by Bauer and Schedl is conceptual: how to get the job of augmenting recommendation systems with surprise and opposition done. They propose using the priming technique from psychology [3] and nudge theory from behavioral economics [4] for surprise and opposition design. The 2015 movie Focus, starring Will Smith and Margot Robbie, includes elements of priming. The Prestige (2006), starring Scarlett Johansson, Christian Bale, and Hugh Jackman, portrays surprise or plot twists. The paper illustrates these concepts using a music recommender system.

A third part of the challenge is practical. How will augmented recommenders be accepted? Are these features even desirable to users? I would have liked to see this area explored further. IBM’s Chef Watson, a recipe recommender system, was tuned to be less surprising [5]. Many gamers felt that Navi, the AI fairy navigation recommendation feature in the video game The Legend of Zelda: Ocarina of Time, was annoying and wished it could be turned off. Clippit/Clippy, the infamous Microsoft Office assistant, turned off users before it was eventually turned off [6]. My experience using Inspire for Faculty from Civitas Learning to address the problem of student retention and persistence [7] indicates that nudges need to be contextual, timely, short, and precise. Not easy.

Additionally, I would have liked to see Bauer and Schedl discuss the possible combination of human and machine hybrid activities for surprise and opposition--what Daugherty and Wilson call fusion skills in their recent book [8]. This could lead to better fidelity and acceptance of recommender systems, at least for now.

This relatively short position paper may not be a “sleeping beauty” [9], but could have significant influence as it expands our thinking about what is possible and provokes thought about what is desirable in recommender systems. One such thought is to remember the nudger’s credo: nudge for good.