158 Environments of intelligence
whereas others are deliberately omitted or even distorted, arguably even if a faithful match could be accomplished for them.
The second limitation with which the strategy of human-likeness of agents is
confronted lies in the observable irritations in the course of interaction that are
induced by the asymmetries of the transmission channels available to either side,
and by the human interrogator’s very knowledge of the machine nature of his or
her counterpart. Instead of interrogators trying to reveal the agents’ nature as a
machine, as in Turing’s imitation game, human users can be observed to deliberately and playfully test the limits of the agents’ conversational capabilities – for
example by teasing them (for the following observations, see Krummheuer 2008;
2009): in human communication, teasing is a means of applying criticism in a
playful, tongue-in-cheek fashion that tests for the counterpart’s sense of humour.
It is a practice that relies on a degree of mutual familiarity, and it is not normally
applied on first encounter. Otherwise, teasing behaviour will appear inappropriate or even transgressive. Teasing a conversational agent on first encounter thus
appears as an awkwardly hybrid kind of interaction. It implies a wilful transgression, while generally remaining within the bounds of interpersonal communication and the mutual expectations that come with it: to which extent does
the counterpart cooperate in the teasing game, and at which point will irritation,
embarrassment or annoyance ensue? It can be observed that, first, the agent will
appear irritated quite quickly, obviously not ‘getting the point’ of the tease and,
second, that the interrogators are likely to exploit the aforementioned communicative asymmetries of the interaction to achieve precisely this end. Neither is likely
to happen in normal interpersonal communication.
In the social robotics case, one will not find instances of teasing sensu strictu, as
this practice will require an elaborate degree of verbal behaviour from both sides.
What one will find, however, is that, in the set-up of the interaction experiments,
behaviours towards KISMET, the paradigm-defining social robot, were actually
designed to resemble the behaviours one would normally display towards a small
child, with utterances of whatever content being produced in a characteristically
pointed praising, scolding, comforting or other tone that is typically reserved
to parent-child communication. In effect, however, the subjects’ behaviours go
beyond this mode of communication, in being marked by a lack of distance and a
sense of not taking the counterpart seriously that would be unusual in any kind of
interpersonal interaction, whereas the basic patterns of interpersonal interaction
remain intact. Hence, a partial analogy to teasing can be observed. For example
in a video recording of an interaction in the robotics laboratory, one subject will
be found asking KISMET in a scolding tone: “Where did you put your body?”
Although obviously not expecting a verbal response, the subject will be seen eliciting an apparently embarrassed expression from the robot. Another subject can
be observed turning towards the laboratory staff and jokingly grimacing towards
them after having successfully elicited a similarly embarrassed reaction on scolding the robot.
5
In either example, no degree of naturalness, familiarity and human-likeness in
the agent’s design will ensure that it is ultimately interacted with as a person, yet
whereas others are deliberately omitted or even distorted, arguably even if a faithful match could be accomplished for them.
The second limitation with which the strategy of human-likeness of agents is
confronted lies in the observable irritations in the course of interaction that are
induced by the asymmetries of the transmission channels available to either side,
and by the human interrogator’s very knowledge of the machine nature of his or
her counterpart. Instead of interrogators trying to reveal the agents’ nature as a
machine, as in Turing’s imitation game, human users can be observed to deliberately and playfully test the limits of the agents’ conversational capabilities – for
example by teasing them (for the following observations, see Krummheuer 2008;
2009): in human communication, teasing is a means of applying criticism in a
playful, tongue-in-cheek fashion that tests for the counterpart’s sense of humour.
It is a practice that relies on a degree of mutual familiarity, and it is not normally
applied on first encounter. Otherwise, teasing behaviour will appear inappropriate or even transgressive. Teasing a conversational agent on first encounter thus
appears as an awkwardly hybrid kind of interaction. It implies a wilful transgression, while generally remaining within the bounds of interpersonal communication and the mutual expectations that come with it: to which extent does
the counterpart cooperate in the teasing game, and at which point will irritation,
embarrassment or annoyance ensue? It can be observed that, first, the agent will
appear irritated quite quickly, obviously not ‘getting the point’ of the tease and,
second, that the interrogators are likely to exploit the aforementioned communicative asymmetries of the interaction to achieve precisely this end. Neither is likely
to happen in normal interpersonal communication.
In the social robotics case, one will not find instances of teasing sensu strictu, as
this practice will require an elaborate degree of verbal behaviour from both sides.
What one will find, however, is that, in the set-up of the interaction experiments,
behaviours towards KISMET, the paradigm-defining social robot, were actually
designed to resemble the behaviours one would normally display towards a small
child, with utterances of whatever content being produced in a characteristically
pointed praising, scolding, comforting or other tone that is typically reserved
to parent-child communication. In effect, however, the subjects’ behaviours go
beyond this mode of communication, in being marked by a lack of distance and a
sense of not taking the counterpart seriously that would be unusual in any kind of
interpersonal interaction, whereas the basic patterns of interpersonal interaction
remain intact. Hence, a partial analogy to teasing can be observed. For example
in a video recording of an interaction in the robotics laboratory, one subject will
be found asking KISMET in a scolding tone: “Where did you put your body?”
Although obviously not expecting a verbal response, the subject will be seen eliciting an apparently embarrassed expression from the robot. Another subject can
be observed turning towards the laboratory staff and jokingly grimacing towards
them after having successfully elicited a similarly embarrassed reaction on scolding the robot.
5
In either example, no degree of naturalness, familiarity and human-likeness in
the agent’s design will ensure that it is ultimately interacted with as a person, yet