192 Environments of intelligence
More positively, there might be a historically grounded, indirect but still systematic justification for commencing from an account of artefacts, so as to arrive
at some conclusions with respect to cognitive functions more generally. After all,
the concept of machine played a crucial role in the emergence of the cognitive
sciences, as two of its founding disciplines strongly relied on machine models:
AI as inspired by Turing’s work, especially (1950), and Cybernetics as inspired
partly by Ashby’s non-computational, material machine model of adaptive processes known as the “homeostat” (1960). These two different but related types of
models – Turing and Ashby were concurrent and corresponding members of the
“Ratio Club” and knew each other’s work (Husbands and Holland 2008) – were
crucial to the establishment of the notion of “mind as machine” that defined the
cognitive sciences ever since (Boden 2006; Gardner 1985). One of the reasons
why Turing considered the original question of whether machines can think “too
meaningless to deserve discussion” (1950, 442, see Chapter 1) is that the “normal
use of the words” (1950, 433) is so parochial and historically contingent that
the very conjunction of “machine” and “thinking” seemed to be a conceptual
impossibility on the normal use of the words in 1950. Part of Turing and Ashby’s
endeavours was to demonstrate that what first seemed a conceptual impossibility could become a matter of serious consideration and empirical investigation.
If such an empirical investigation is possible, even if it has its limitations, and
if such investigation informed the design of real-world computing artefacts, and
even if these artefacts are not thinking machines, these artefacts will have made a
contribution to redefining not only the concepts of “machines” and “thinking” but
also our perceptions and practical treatment of what the nature of machines and
human thinking actually is.
In one of the best-defined and most carefully argued sceptical accounts of AI,
Harry Collins and Martin Kusch (1998) propose a conceptual distinction between
human beings and machines that is very intriguing in this respect: on their account,
a machine is distinguished from a human being by its inability to perform or “do”
actions of one kind in particular, namely those actions which can be accomplished
in manifold, materially dissimilar ways and can be identified only as belonging
to the same type by reference to an understanding of their social context. To the
observer, the behavioural patterns involved may look rather different on different
occasions, but they are tied together by the socially sanctioned purposes that they
serve. One can pay a bill in several ways, by handing over cash or by swiping a
credit card through a card reader. It would be the same action in the light of the
purpose of paying a bill – but only in a society where credit card payment is an
established practice.
In contrast to these parochially human “polimorphic” actions (playing on the
words “polymorphic” and “polis”), Collins and Kusch continue, human beings
and machines seem to share an ability of performing “mimeomorphic” actions,
which are defined as those actions which are tied to one and the same type of
behavioural sequence, with no or rather minute degrees of variance permitted.
There are only so many ways of performing a certain dance move or golf swing.
Doing them different is likely to amount to doing them incorrectly. For their
More positively, there might be a historically grounded, indirect but still systematic justification for commencing from an account of artefacts, so as to arrive
at some conclusions with respect to cognitive functions more generally. After all,
the concept of machine played a crucial role in the emergence of the cognitive
sciences, as two of its founding disciplines strongly relied on machine models:
AI as inspired by Turing’s work, especially (1950), and Cybernetics as inspired
partly by Ashby’s non-computational, material machine model of adaptive processes known as the “homeostat” (1960). These two different but related types of
models – Turing and Ashby were concurrent and corresponding members of the
“Ratio Club” and knew each other’s work (Husbands and Holland 2008) – were
crucial to the establishment of the notion of “mind as machine” that defined the
cognitive sciences ever since (Boden 2006; Gardner 1985). One of the reasons
why Turing considered the original question of whether machines can think “too
meaningless to deserve discussion” (1950, 442, see Chapter 1) is that the “normal
use of the words” (1950, 433) is so parochial and historically contingent that
the very conjunction of “machine” and “thinking” seemed to be a conceptual
impossibility on the normal use of the words in 1950. Part of Turing and Ashby’s
endeavours was to demonstrate that what first seemed a conceptual impossibility could become a matter of serious consideration and empirical investigation.
If such an empirical investigation is possible, even if it has its limitations, and
if such investigation informed the design of real-world computing artefacts, and
even if these artefacts are not thinking machines, these artefacts will have made a
contribution to redefining not only the concepts of “machines” and “thinking” but
also our perceptions and practical treatment of what the nature of machines and
human thinking actually is.
In one of the best-defined and most carefully argued sceptical accounts of AI,
Harry Collins and Martin Kusch (1998) propose a conceptual distinction between
human beings and machines that is very intriguing in this respect: on their account,
a machine is distinguished from a human being by its inability to perform or “do”
actions of one kind in particular, namely those actions which can be accomplished
in manifold, materially dissimilar ways and can be identified only as belonging
to the same type by reference to an understanding of their social context. To the
observer, the behavioural patterns involved may look rather different on different
occasions, but they are tied together by the socially sanctioned purposes that they
serve. One can pay a bill in several ways, by handing over cash or by swiping a
credit card through a card reader. It would be the same action in the light of the
purpose of paying a bill – but only in a society where credit card payment is an
established practice.
In contrast to these parochially human “polimorphic” actions (playing on the
words “polymorphic” and “polis”), Collins and Kusch continue, human beings
and machines seem to share an ability of performing “mimeomorphic” actions,
which are defined as those actions which are tied to one and the same type of
behavioural sequence, with no or rather minute degrees of variance permitted.
There are only so many ways of performing a certain dance move or golf swing.
Doing them different is likely to amount to doing them incorrectly. For their