The intelligence of environments 159
neither is it unequivocally not being treated as a person. The agent is treated like a
human being to a very peculiarly circumscribed extent that has no genuine counterpart in interpersonal communication in everyday, real-world contexts. It will
be thus treated on the grounds of a selection of specifically presented elementary
human-like features. The domain of signals that typically pertain to interpersonal
interaction has become ambiguous in its extension. Although the human beings
involved will readily distinguish between an agent or robot and another human
being, there will be some relevant cues emanating from the agents and robots
that cannot be disambiguated by the human counterparts to a degree sufficient to
confidently place them within either realm. The agents and robots are somewhat
precariously situated in human informational environments, where the boundaries
of the domains of signals of being human are being tested, and may be modified,
in interaction.
With respect to the convergence/isomorphism distinction introduced in the last
section of Chapter 8, both conditions apply to an – instructively – limited extent:
to the extent that naturalness of interaction between human being and robot or
agent is successfully achieved, and to the extent that the agent’s or robot’s behaviours are governed by regularities, that is programmes, that have such naturalness as their aim, and to the extent that the robot or agent’s expressions provide
recognisable and reliable analogues of what a human counterpart would display
in the same kind of situation, the condition of isomorphism holds. Under this proviso, an agent or robot will be able to provide a general model of the elementary
cues by which human beings recognise a human counterpart’s surprise, sadness,
embarrassment or other affective state, without implying a claim that such states
are actually implemented in the robot. For example a robot model was used to
identify the cues by which a person’s trustworthiness is recognised (DeSteno et al.
2012; Lee et al. 2013). However, to the extent that irritation and ambiguity inadvertently arise in interaction, the isomorphism will remain incomplete.
Conversely, if convergence were to hold on the level of human-likeness, the
human-like cues emitted by the agent or robot would, first, have to map onto traits
with human-like functions of the agent or robot itself. Second, they would have
to be detectable for the human counterpart with sufficient reliability, for being
governed by the same regularities. Hence, convergence would require a functional
analogue of surprise, sadness, embarrassment or other affective state to hold for
the robot or agent, and hence a partly human-like cognitive architecture, which
is not necessarily part of the design of these systems – but see Breazeal et al.
(2009) for a study in pursuit of such a design within the social robotics paradigm.
To the extent that the cues involved are tailor-made to leave a partly human-like
impression, and to the extent that irritation and ambiguity may arise in interaction
precisely for the partialness of that impression and the non-identity of underlying
functions in the robot or agent, convergence is not achieved.
A different kind of convergence-requiring situation arises when human and
robot are supposed to work in coordinated fashion on a shared, environmentbound task, where it is the information on that task that has to match in the first
place, and where any verbal or behavioural cues exchanged stand in the service
neither is it unequivocally not being treated as a person. The agent is treated like a
human being to a very peculiarly circumscribed extent that has no genuine counterpart in interpersonal communication in everyday, real-world contexts. It will
be thus treated on the grounds of a selection of specifically presented elementary
human-like features. The domain of signals that typically pertain to interpersonal
interaction has become ambiguous in its extension. Although the human beings
involved will readily distinguish between an agent or robot and another human
being, there will be some relevant cues emanating from the agents and robots
that cannot be disambiguated by the human counterparts to a degree sufficient to
confidently place them within either realm. The agents and robots are somewhat
precariously situated in human informational environments, where the boundaries
of the domains of signals of being human are being tested, and may be modified,
in interaction.
With respect to the convergence/isomorphism distinction introduced in the last
section of Chapter 8, both conditions apply to an – instructively – limited extent:
to the extent that naturalness of interaction between human being and robot or
agent is successfully achieved, and to the extent that the agent’s or robot’s behaviours are governed by regularities, that is programmes, that have such naturalness as their aim, and to the extent that the robot or agent’s expressions provide
recognisable and reliable analogues of what a human counterpart would display
in the same kind of situation, the condition of isomorphism holds. Under this proviso, an agent or robot will be able to provide a general model of the elementary
cues by which human beings recognise a human counterpart’s surprise, sadness,
embarrassment or other affective state, without implying a claim that such states
are actually implemented in the robot. For example a robot model was used to
identify the cues by which a person’s trustworthiness is recognised (DeSteno et al.
2012; Lee et al. 2013). However, to the extent that irritation and ambiguity inadvertently arise in interaction, the isomorphism will remain incomplete.
Conversely, if convergence were to hold on the level of human-likeness, the
human-like cues emitted by the agent or robot would, first, have to map onto traits
with human-like functions of the agent or robot itself. Second, they would have
to be detectable for the human counterpart with sufficient reliability, for being
governed by the same regularities. Hence, convergence would require a functional
analogue of surprise, sadness, embarrassment or other affective state to hold for
the robot or agent, and hence a partly human-like cognitive architecture, which
is not necessarily part of the design of these systems – but see Breazeal et al.
(2009) for a study in pursuit of such a design within the social robotics paradigm.
To the extent that the cues involved are tailor-made to leave a partly human-like
impression, and to the extent that irritation and ambiguity may arise in interaction
precisely for the partialness of that impression and the non-identity of underlying
functions in the robot or agent, convergence is not achieved.
A different kind of convergence-requiring situation arises when human and
robot are supposed to work in coordinated fashion on a shared, environmentbound task, where it is the information on that task that has to match in the first
place, and where any verbal or behavioural cues exchanged stand in the service