If you ever find yourself being ignored by a bartender while trying to order a drink, it probably means you haven’t been tipping enough. However, if that bartender happens to be a robot, it could be that it simply hasn’t picked up on the social signals that indicate your interest in placing an order. As the field of robotics increasingly looks to breach the human-machine communication barrier, a team of researchers have created a robotic barman named James, which has helped them decipher the unspoken exchanges that occur between customers and bar staff.
Human social interactions are extremely complex, and often involve subtle hints and signals that convey a great deal of information not contained within our explicit verbal communication. However, while these are often taken for granted by living people, they pose a major challenge for robotics developers, whose creations often lack the social intelligence to perceive these human gestures. For this reason, researchers at Bielefeld University’s Cluster of Excellence Cognitive Interaction Technology devised an experiment to determine what sorts of cues robotic bartenders should be programmed to look out for.
To conduct their study, the team used a method known as Ghost-in-the-Machine, whereby human participants (or "ghosts") were allowed to “observe the scene through the eyes and ears of the robot,” but not interact with customers. As each scene unfolded, ghosts were instructed to indicate the appropriate action from James’s repertoire – such as asking what a customer wanted, serving them a drink, or ignoring them. By comparing the choices made by the ghosts and those made by the robot itself, researchers were able to decipher which social signals are most indicative of a person’s intentions to order a drink at a bar.
For instance, results indicated that ghosts shifted their focus as the drink-serving process progressed. Initially, they paid attention to visual cues such as the body position of punters, choosing to interact with those who made eye contact or faced them directly. Once this initial contact had been made, speech became the key social modality, with ghosts basing their decisions on what customers said, even if they were no longer within their sight of vision. In contrast, James processed all data equally, and was therefore unable to determine which signals were the most important and which it could afford to ignore. As a result, it tended to break off interactions when a customer was no longer visible (such as when the robot turned its back to start preparing a drink).
In a statement, study co-author Sebastian Loth explained that “if a customer was not visible, the robot assumed that it could not serve a drink or speak into thin air.” As such, the results of the experiment indicate that “a robotic bartender should [be programmed to] sometimes ignore data.”
The implications of this research – which has been published in the journal Frontiers in Psychiatry – could lead to “substantial improvements for human-robot interaction policies,” which may well drive the development of superior service robots in a number of different settings. With robots already being used to staff a hotel in Japan and guide visitors around a museum, a new age of human-machine interactions could soon be upon us.
2
