The intelligence of environments 155
A humanoid robot built according to the cognitive robotics approach will
inhabit an informational environment that is analogously shaped to that of a
human being in a number of relevant aspects. Prevailing disanalogies are to only
some extent a matter of the limitations of the subset of trackable information chosen for the model. They are to a more significant extent a matter of the underlying
ecological relevance of the conditions to be thus tracked. This problem haunts all
approaches to embodied AI bar evolutionary robotics: unless a genuine functional
equivalence with organic traits can be established for a robotic trait, the make-up
of the robot’s informational environment will be based on a presumed, putative
relevance of the conditions to be tracked. Human sensu-motoric integration has
the function it has because human beings need to track ambient information in
a certain integrated way, as a matter of biological necessity. Equally, an object
affords what it affords to a human being on the grounds of his or her constitution
and abilities, which, again, to a large part are a matter of biological necessity.
One may design analogues of these necessities in similar but tremendously more
complex fashion as in evolutionary robotics, but this is not part of the approach
under discussion here. Still, a cognitive robot will be in the business of tracking
some of the same informational domains that human beings are in the business
of tracking, even in the absence of a genuine biological, somatic, evolutionary
foundation – which will always be the ultimate sceptical resource for an AI critic
of the Searlean or Dreyfusean slant.
Experiments in evolutionary and cognitive robotics have been feeding into
an image of the human mind according to which perception and action are coupled in direct and integrated fashion, and which has gained much prominence,
within and outside computer science and engineering, over the representationalist
and logicist image that travelled with classical AI and its artefacts. By different
but nonetheless behaviour-based means, other approaches in Nouvelle AI, most
prominently ‘sociable’ humanoid robotics, can be found to feed into the same
image. Unlike evolving and cognitive robots, sociable robots and, as their closest kin, embodied conversational agents are designed to interact with humans in
shared environments, and not only shape their own but also have a chance to alter
human informational environments.
Embodied conversational agents and social robotics
Embodied conversational agents are human-like virtual agents, normally displayed on a screen, with whom some form of natural spoken conversation is
possible, including the display of a number of natural behavioural cues. These
include adequate intonation and prosody in spoken expressions as well as gestures
and facial expressions from the agent’s side. However, the mode of interaction
remains mostly asymmetrical to date, as input to the system in most current agent
technologies can only be typed, so that the agent is not in command of vision
or other sensory modalities by which to register behavioural cues of its human
counterparts. In response to the typed input, the conversational agent articulates
complex, syntactically and semantically well-formed sentences with gestures,
A humanoid robot built according to the cognitive robotics approach will
inhabit an informational environment that is analogously shaped to that of a
human being in a number of relevant aspects. Prevailing disanalogies are to only
some extent a matter of the limitations of the subset of trackable information chosen for the model. They are to a more significant extent a matter of the underlying
ecological relevance of the conditions to be thus tracked. This problem haunts all
approaches to embodied AI bar evolutionary robotics: unless a genuine functional
equivalence with organic traits can be established for a robotic trait, the make-up
of the robot’s informational environment will be based on a presumed, putative
relevance of the conditions to be tracked. Human sensu-motoric integration has
the function it has because human beings need to track ambient information in
a certain integrated way, as a matter of biological necessity. Equally, an object
affords what it affords to a human being on the grounds of his or her constitution
and abilities, which, again, to a large part are a matter of biological necessity.
One may design analogues of these necessities in similar but tremendously more
complex fashion as in evolutionary robotics, but this is not part of the approach
under discussion here. Still, a cognitive robot will be in the business of tracking
some of the same informational domains that human beings are in the business
of tracking, even in the absence of a genuine biological, somatic, evolutionary
foundation – which will always be the ultimate sceptical resource for an AI critic
of the Searlean or Dreyfusean slant.
Experiments in evolutionary and cognitive robotics have been feeding into
an image of the human mind according to which perception and action are coupled in direct and integrated fashion, and which has gained much prominence,
within and outside computer science and engineering, over the representationalist
and logicist image that travelled with classical AI and its artefacts. By different
but nonetheless behaviour-based means, other approaches in Nouvelle AI, most
prominently ‘sociable’ humanoid robotics, can be found to feed into the same
image. Unlike evolving and cognitive robots, sociable robots and, as their closest kin, embodied conversational agents are designed to interact with humans in
shared environments, and not only shape their own but also have a chance to alter
human informational environments.
Embodied conversational agents and social robotics
Embodied conversational agents are human-like virtual agents, normally displayed on a screen, with whom some form of natural spoken conversation is
possible, including the display of a number of natural behavioural cues. These
include adequate intonation and prosody in spoken expressions as well as gestures
and facial expressions from the agent’s side. However, the mode of interaction
remains mostly asymmetrical to date, as input to the system in most current agent
technologies can only be typed, so that the agent is not in command of vision
or other sensory modalities by which to register behavioural cues of its human
counterparts. In response to the typed input, the conversational agent articulates
complex, syntactically and semantically well-formed sentences with gestures,