4 Preliminaries
classic definition conceives of a machine as “ ‘a closed kinematic chain’ or ‘a
combination of resistant bodies so arranged that by their means the mechanical
forces of nature can be compelled to do work accompanied by certain determinate
motions’ ” (Mitcham 1994, 170f, quoting Franz Reuleaux’s 1876 book Kinematics of Machinery). In seeking to extend the domain of what machines can do
beyond kinematics (and, by extension, the effects of electrical energy), Turing
aligned himself with a small tradition of thinkers and inventors who conceived of
machines that could serve logico-mathematical purposes beyond basic calculating
functions, most notably Charles Babbage (1864) and Gottfried Wilhelm Leibniz
(1685). In abstraction from what cog-belt-and-pulley machines do, a machine was
now conceived of as any arrangement of discrete elements and the discrete operational steps associated with them which, with determinate regularity, transforms a
certain input into a certain output.
With respect to comparing human and machine accomplishments on the basis of
this theoretical concept, the key novelty introduced by Turing’s theory of computability was twofold: first, his theoretical machines, in analogy to human computers, could switch between different sets of instructions, and hence between different
logico-mathematical tasks, without necessitating a change to the structure of the
machine proper. The input-output relations and a certain subset of the machine’s
operations were now conceived of as modifiable. Second, the same type of logical
operations could be realised in a variety of different systems, from human beings to
specific machines, in a variety of ways, and thus in abstraction from their physical
characteristics. The first part of this twofold idea is entailed by Turing’s description
of his theoretical “Universal Machines”, whereas the second has become known as
“Turing Machine functionalism”. It includes opening up the conceptual possibility of
artefacts to accomplish tasks that would count as thinking in a human. However, this
is an implication (not a corollary) of the larger argument, which is not about human
beings or machines in particular but about a meta-mathematical method. Above all,
the second part of Turing’s idea waives any requirement that those accomplishments
would have to be reached by the same means in the machine analogue.
Hence, a superficial reading of the imitation game is prone to misguiding the
reader into seeking for an analogy between human and machine accomplishments
that is closer than intended. In one of the first philosophical essays that took
Turing’s thought experiment seriously, Keith Gunderson (1964) captures a point
that seems to have been lost in the long-lasting, and arguably not always fruitful,
philosophical debates around the possibilities and limitations of AI:
One might well contend that machines can’t think, for they do much better than that. [. . .] Machines can almost instantaneously and infallibly produce accurate and sometimes original answers to many complex and difficult
mathematical problems with which they are presented. They do not need to
“think out” the answers. In the end the steam drill outlasted John Henry as a
digger of railway tunnels, but that didn’t prove the machine had muscles; it
proved that muscles were not needed for digging railway tunnels.
(Gunderson 1964, 244f )
classic definition conceives of a machine as “ ‘a closed kinematic chain’ or ‘a
combination of resistant bodies so arranged that by their means the mechanical
forces of nature can be compelled to do work accompanied by certain determinate
motions’ ” (Mitcham 1994, 170f, quoting Franz Reuleaux’s 1876 book Kinematics of Machinery). In seeking to extend the domain of what machines can do
beyond kinematics (and, by extension, the effects of electrical energy), Turing
aligned himself with a small tradition of thinkers and inventors who conceived of
machines that could serve logico-mathematical purposes beyond basic calculating
functions, most notably Charles Babbage (1864) and Gottfried Wilhelm Leibniz
(1685). In abstraction from what cog-belt-and-pulley machines do, a machine was
now conceived of as any arrangement of discrete elements and the discrete operational steps associated with them which, with determinate regularity, transforms a
certain input into a certain output.
With respect to comparing human and machine accomplishments on the basis of
this theoretical concept, the key novelty introduced by Turing’s theory of computability was twofold: first, his theoretical machines, in analogy to human computers, could switch between different sets of instructions, and hence between different
logico-mathematical tasks, without necessitating a change to the structure of the
machine proper. The input-output relations and a certain subset of the machine’s
operations were now conceived of as modifiable. Second, the same type of logical
operations could be realised in a variety of different systems, from human beings to
specific machines, in a variety of ways, and thus in abstraction from their physical
characteristics. The first part of this twofold idea is entailed by Turing’s description
of his theoretical “Universal Machines”, whereas the second has become known as
“Turing Machine functionalism”. It includes opening up the conceptual possibility of
artefacts to accomplish tasks that would count as thinking in a human. However, this
is an implication (not a corollary) of the larger argument, which is not about human
beings or machines in particular but about a meta-mathematical method. Above all,
the second part of Turing’s idea waives any requirement that those accomplishments
would have to be reached by the same means in the machine analogue.
Hence, a superficial reading of the imitation game is prone to misguiding the
reader into seeking for an analogy between human and machine accomplishments
that is closer than intended. In one of the first philosophical essays that took
Turing’s thought experiment seriously, Keith Gunderson (1964) captures a point
that seems to have been lost in the long-lasting, and arguably not always fruitful,
philosophical debates around the possibilities and limitations of AI:
One might well contend that machines can’t think, for they do much better than that. [. . .] Machines can almost instantaneously and infallibly produce accurate and sometimes original answers to many complex and difficult
mathematical problems with which they are presented. They do not need to
“think out” the answers. In the end the steam drill outlasted John Henry as a
digger of railway tunnels, but that didn’t prove the machine had muscles; it
proved that muscles were not needed for digging railway tunnels.
(Gunderson 1964, 244f )