Research Insights: Algorithmic Ethnography with Ritwik Banerji

25 Apr, 2018

Research Insights: Algorithmic Ethnography with Ritwik Banerji

Ritwik Banerji's paper, "Resembling Repugnance: Beyond the 'Uncanny Valley' in the Algorithmic Ethnography of Human Sociality," expands upon the age-old idea that the contempt for humanlike technology stems from its failure to emulate humanity through real-time interactions. He proposes, however, that subjects feel repulsed because its behavior simultaneously reminds them of familiar, unfavorable human qualities.

In his own words:

The (paper) openly questions the frequent assumption that our human aversions to the humanlike qualities of mechanical social interactants is entirely explained by the uncanny valley hypothesis. Additionally, it also pushes for further attention to the fact that for all our talk of uncanny valleys, we remain quite unsure what exactly causes these aversions in the first place.

Drawing on an experimental ethnography in which performers of free improvisation (in music) have been asked to play with a virtual performer of this practice and then compare it to a human performer, the paper reveals that the aversions to such humanlike mechanical interlocutors is both due to their "uncanny" resemblance to human interactants but also do to the fact that they often remind us of people who are simply irritating. For example, for one player it was indeed the case that he found the experience uncanny. What this means is that he found it odd that I had found a way to simulate the violent intensity of an improvising electric guitarist, and yet all the physicality and theatricality that he might expect from such a player was absent since all this loud energy was coming out of an immotile speaker box. For others, however, the behavior was in fact all too human. One player was reminded by the system's awkward silences of players who lack the confidence to take a risk and keep the piece going with a new idea while another was reminded of improvisers who are "bad" listeners and play too much.

Thus, the overall point was that the uncanny valley is only at best a name for a phenomenon, but not quite an explanation of its cause. Moreover, all these various reactions to these systems illustrate that the encounter with such simulated human characters is really just an exercise in eliciting each human interlocutor's conception of "humanness" and ultimately indicates that this concept is actually quite a lot more variable and culturally-specific than universalizing discourses on this topic might suggest. Lastly, these encounters also suggest that when we say that a machine's behavior is "not human," this denial of human performativity might actually be referring the fact that the system may behaving like a person, but a person who is not fully formed (i.e. a child) or a person-type with whom we are not culturally familiar (i.e. the classic "Other" of much anthropological research.)