The intelligence of environments 157
Despite the human-likeness and naturalness of interaction involved, the aim of
the systems under investigation here is not indiscernibility of the agent or robot
from human beings but, either proximately (for social robots) or ultimately (for conversational agents), a design that allows human beings to engage in an anxiety-free,
in some selected respects, natural communication with a computer system – which
is an issue of increasing societal relevance, given that demographic developments
in Western societies are such that caring for the elderly will increasingly have to
be delegated to partly human-like robots and agents, acceptance of whom by their
users will be a design issue (LFE-b-2015, MIMED-a-2015).
On the accounts provided by Joseph Bates (1994) and Robbert-Jan Beun
et al. (2003), embodied conversational agents in particular are one experimental
approach to the recurrent problem of mismatching metaphors in human-computer
interaction (HCI): one of the key variables in the choice of metaphors of interaction is the degree of human-likeness, visual, textual or other, with which the
system is going to appear to the user. The metaphors intended in design and the
ones experienced in use are often found to diverge, thus leading to difficult or
failed interaction. In this context, an additional cue for the human-likeness of
one’s machine counterpart may enable a better matching, in that the expectation
of being able to converse with the system as with a human being is reinforced.
However, there are at least two instructive limitations to this strategy.
The first limitation concerns the choice of the aspects and the degree of humanlikeness of the agent. Although it might seem that a high degree of perceived
human-likeness would facilitate interaction, that degree does not necessarily correlate with the perceived degree of familiarity – a problem that has been discussed
in HCI research under the imaginative heading of the “uncanny valley” (Mori
1970): although perceptions and attitudes towards an industrial robot normally
remain indifferent, as it neither displays a notable degree of human-likeness nor
is expected to look familiar in any way, humanoid robots and embodied conversational agents will be both more human-like and more familiar in appearance –
until a certain point: modest similarity is conducive to familiarity if the appropriate
human features are chosen for imitation, and if the degree of approximation is
well-calibrated. Robots and agents may remain clearly distinct from human beings
but still appear fairly familiar in interaction, if a limited but well-chosen set of cues
used in human communication is implemented in a pointed fashion – very much
in the sense of a good caricature. If, however, the degree of similarity increases
further, yet without attaining absolute perfection, the robot’s or agent’s appearance
is likely to become less familiar, to the point of being irritating or even eerie. In
fact, Ho and MacDorman (2010) present empirical evidence for the assumption
that the familiarity variable in Mori’s argument is most accurately described as the
inverse degree of eeriness (i.e. as reassurance) in the appearance of the agent. Thus,
the selection of features to be imitated by an agent system, and the degree to which
they actually resemble the human model, turn out to be crucially important variables.
4
Hence, convergence or isomorphism between the cues of human-likeness
of artificial agents and the features of natural human beings are best conceived
as obtaining in relation to certain key variables that shall be faithfully matched,
Despite the human-likeness and naturalness of interaction involved, the aim of
the systems under investigation here is not indiscernibility of the agent or robot
from human beings but, either proximately (for social robots) or ultimately (for conversational agents), a design that allows human beings to engage in an anxiety-free,
in some selected respects, natural communication with a computer system – which
is an issue of increasing societal relevance, given that demographic developments
in Western societies are such that caring for the elderly will increasingly have to
be delegated to partly human-like robots and agents, acceptance of whom by their
users will be a design issue (LFE-b-2015, MIMED-a-2015).
On the accounts provided by Joseph Bates (1994) and Robbert-Jan Beun
et al. (2003), embodied conversational agents in particular are one experimental
approach to the recurrent problem of mismatching metaphors in human-computer
interaction (HCI): one of the key variables in the choice of metaphors of interaction is the degree of human-likeness, visual, textual or other, with which the
system is going to appear to the user. The metaphors intended in design and the
ones experienced in use are often found to diverge, thus leading to difficult or
failed interaction. In this context, an additional cue for the human-likeness of
one’s machine counterpart may enable a better matching, in that the expectation
of being able to converse with the system as with a human being is reinforced.
However, there are at least two instructive limitations to this strategy.
The first limitation concerns the choice of the aspects and the degree of humanlikeness of the agent. Although it might seem that a high degree of perceived
human-likeness would facilitate interaction, that degree does not necessarily correlate with the perceived degree of familiarity – a problem that has been discussed
in HCI research under the imaginative heading of the “uncanny valley” (Mori
1970): although perceptions and attitudes towards an industrial robot normally
remain indifferent, as it neither displays a notable degree of human-likeness nor
is expected to look familiar in any way, humanoid robots and embodied conversational agents will be both more human-like and more familiar in appearance –
until a certain point: modest similarity is conducive to familiarity if the appropriate
human features are chosen for imitation, and if the degree of approximation is
well-calibrated. Robots and agents may remain clearly distinct from human beings
but still appear fairly familiar in interaction, if a limited but well-chosen set of cues
used in human communication is implemented in a pointed fashion – very much
in the sense of a good caricature. If, however, the degree of similarity increases
further, yet without attaining absolute perfection, the robot’s or agent’s appearance
is likely to become less familiar, to the point of being irritating or even eerie. In
fact, Ho and MacDorman (2010) present empirical evidence for the assumption
that the familiarity variable in Mori’s argument is most accurately described as the
inverse degree of eeriness (i.e. as reassurance) in the appearance of the agent. Thus,
the selection of features to be imitated by an agent system, and the degree to which
they actually resemble the human model, turn out to be crucially important variables.
4
Hence, convergence or isomorphism between the cues of human-likeness
of artificial agents and the features of natural human beings are best conceived
as obtaining in relation to certain key variables that shall be faithfully matched,