Ants and robots, parlour games and steam drills 7
As the flip side of the same coin, the accomplishments of computing machinery
will also alter our understanding of what a machine is. Given that advanced tool
use is a uniquely human characteristic, a change in what artefacts can accomplish,
and a change in human perception of those accomplishments, will have an equally
pivotal influence on the human condition. Both changes that Turing had in mind
are happening right before our eyes.
The paradigm of what I have in mind here has been articulated, in related but
slightly different fashion, by two computer scientists who wrote half a century
after Turing inaugurated that very discipline in his 1936 paper: Rodney Brooks
and Mark Weiser. Both authors highlight the importance of the relations to the
respective environments of the systems they were developing while relegating
those intellectual accomplishments which were central to classical AI to secondary importance.
As one of the first proponents of “Nouvelle AI”, Brooks argues that traditional
AI systems were incapable of grasping important aspects of human cognition
because these aspects are not located in human beings’ heads but in the environments in which they act and interact. Thus, inclusion of these environmental aspects in embodied, that is to say robotic, systems has to take precedence
over criteria of human-likeness or mental representation. It is best accomplished
in bottom-up and modular fashion, starting from fairly simple systems, Brooks
continues:
It seemed a reasonable requirement that intelligence be reactive to dynamic
aspects of the environment, that a mobile robot operate on time scales similar
to those of animals and humans, and that intelligence be able to generate
robust behavior in the face of uncertain sensors, an unpredictable environment, and a changing world. [. . .] Internal world models that are complete
representations of the external environment, besides being impossible to
obtain, are not at all necessary for agents to act in a competent manner. Many
of the actions of an agent are quite separable – coherent intelligence can
emerge from independent subcomponents interacting in the world.
(Brooks 1991, 1228)
No AI system will capture the mechanisms responsible for the specific patterns
of organism-environment interaction unless it is able to capture their purpose,
too, which, in turn, can be accommodated only by modelling the organismenvironment interaction in a most direct way. Directness in this sense does
not require the creation of similes of traits or behaviours on a phenomenal
level, nor will it suffice to consider the structures and processes inside the
organism. Instead, one will have to identify, and factor into the equation, specific couplings between variables within organism and environment, and the
emerging patterns of interaction between them. This is the premise on which
the research programme of “behaviour-based AI” was developed (for statements of this programme, see Beer 1995; Brooks 1999; Maes 1993; Steels and
Brooks 1995).
As the flip side of the same coin, the accomplishments of computing machinery
will also alter our understanding of what a machine is. Given that advanced tool
use is a uniquely human characteristic, a change in what artefacts can accomplish,
and a change in human perception of those accomplishments, will have an equally
pivotal influence on the human condition. Both changes that Turing had in mind
are happening right before our eyes.
The paradigm of what I have in mind here has been articulated, in related but
slightly different fashion, by two computer scientists who wrote half a century
after Turing inaugurated that very discipline in his 1936 paper: Rodney Brooks
and Mark Weiser. Both authors highlight the importance of the relations to the
respective environments of the systems they were developing while relegating
those intellectual accomplishments which were central to classical AI to secondary importance.
As one of the first proponents of “Nouvelle AI”, Brooks argues that traditional
AI systems were incapable of grasping important aspects of human cognition
because these aspects are not located in human beings’ heads but in the environments in which they act and interact. Thus, inclusion of these environmental aspects in embodied, that is to say robotic, systems has to take precedence
over criteria of human-likeness or mental representation. It is best accomplished
in bottom-up and modular fashion, starting from fairly simple systems, Brooks
continues:
It seemed a reasonable requirement that intelligence be reactive to dynamic
aspects of the environment, that a mobile robot operate on time scales similar
to those of animals and humans, and that intelligence be able to generate
robust behavior in the face of uncertain sensors, an unpredictable environment, and a changing world. [. . .] Internal world models that are complete
representations of the external environment, besides being impossible to
obtain, are not at all necessary for agents to act in a competent manner. Many
of the actions of an agent are quite separable – coherent intelligence can
emerge from independent subcomponents interacting in the world.
(Brooks 1991, 1228)
No AI system will capture the mechanisms responsible for the specific patterns
of organism-environment interaction unless it is able to capture their purpose,
too, which, in turn, can be accommodated only by modelling the organismenvironment interaction in a most direct way. Directness in this sense does
not require the creation of similes of traits or behaviours on a phenomenal
level, nor will it suffice to consider the structures and processes inside the
organism. Instead, one will have to identify, and factor into the equation, specific couplings between variables within organism and environment, and the
emerging patterns of interaction between them. This is the premise on which
the research programme of “behaviour-based AI” was developed (for statements of this programme, see Beer 1995; Brooks 1999; Maes 1993; Steels and
Brooks 1995).