Intelligent machines as social catalysts

Some people excel at facilitating communication between other people. They are social catalysts: they promote engagement within a team, keep the team on point, defuse hostility, make sure that everyone contributes to the best of their capacity, and generally ensure that the team performs better than the sum of its members. Social catalysts need to be subtly aware of social dynamics, mood changes, face-saving strategies, and unspoken rules of communication. Given these requirements, it would seem that being a social catalyst is a quintessentially human role. The research of Traeger et al. (1), however, opens perspectives about the possibility of having machines rather than people perform the role of social catalysts in human groups. Before we can discuss these perspectives, we need to ask the following questions: why and how would we want a machine to perform such a role? Sometimes, we want a machine to act as a social catalyst simply because there is no available human to perform that role or when we want the social catalyst to act on a nonhuman scale, simultaneously monitoring thousands of conversations. In these cases, we may adopt a better than nothing attitude. We program the machine to do as decent as possible an imitation of the behavior of a human social catalyst and accept that its performance will be somewhere below human level. In other cases, we may want a machine to act as a social catalyst because the machine has decidedly nonhuman ways to perform the role, which work better than what humans might do. For example, the machine may be expert at learning patterns that maintain cooperation between humans or expert at detecting patterns that get humans stuck into suboptimal solutions, and it may act on this knowledge in a way that humans would not even consider—similar to the … [↵][1]1To whom correspondence may be addressed. Email: rahwan{at}mpib-berlin.mpg.de, crandall{at}cs.byu.edu, or jean-francois.bonnefon{at}tse-fr.eu. [1]: #xref-corresp-1-1