26 The New Ecology
with issues of whether machines will ever be able to mimic humans
enough to fool us. Claims were made for the test being passed in 2014
by software called “Eugene”, developed in Saint Petersburg and able
to emulate a 13-year-old boy: before contemplating that we are on
the verge of a singularity, it should also be pointed out that this was
an adapted version of Turing’s rules that allowed for a success rate of
the machine fooling human judges more than 30 per cent of the time
(Eugene managed 33 per cent).
59
Turing’s observation regarding mimicry offers a route out of the impasse of AI and AGI, which is, after W. Grey Walter’s “mimicry of life”
via simple robots in the late 1940s, to consider intelligence as simply that
which exhibits interesting behaviour.
60
Based in Bristol, Walter was able to
construct mechanical tortoises that were phototropic (able to follow light)
and had a bump mechanism to allow them to change direction. His first
robots, Elmie and Elsie, were thus able to simulate autonomous motion
without human intervention, leading to a branch of robotics development
that did not need to concern itself with consciousness and the difficult
tasks of cognitivism. As Russell and Norvig point out in their guide to
modern thinking about AI, the standard approach to artificial intelligence
that is considered and evaluated by the Turing Test (if not actually by Turing himself), ”thinking humanly”, is only one of four approaches, including acting humanly, thinking rationally, and acting rationally. In practical
terms, it is probably the latter, also known as the ”rational agent”, which
tends to concentrate on intelligent behaviour in hardware and software,
that offers the most immediate benefits for future development.
61
The fact that this book is largely concerned with automation and
applied AI rather than strong or general artificial intelligence is by no
means to assert that AGI is impossible: complexity theory and notions
of biological emergence frequently demonstrate that complex systems
can arise from situations where there is not enough information, as in
termite colonies or traffic patterns. Applied AI has often resulted in more
useful models and theories, for example, the modular approach based
on Fodor’s observations in the 1980s that the human mind is largely
composed of task-specific modules,
62
the cognitive processes that allow
us to concentrate on winning (or losing) a game of chess. Fodor’s approach has been criticised, but it does have practical applications in the
field of AI research and robotics. Simulated micro worlds are easier to
understand (and program) than the entire universe of cognition, and this
is particularly evident in areas such as game playing where computers
are becoming better than humans. Rather than simple brute force – an
attempt to compute every possible move recursively – such controlled environment simulations work better via heuristics to build up the limited
range of moves that are possible in a given situation. As with playing
chess, we shall see in a later chapter that this approach to AI is also especially pertinent to certain forms of journalism.
with issues of whether machines will ever be able to mimic humans
enough to fool us. Claims were made for the test being passed in 2014
by software called “Eugene”, developed in Saint Petersburg and able
to emulate a 13-year-old boy: before contemplating that we are on
the verge of a singularity, it should also be pointed out that this was
an adapted version of Turing’s rules that allowed for a success rate of
the machine fooling human judges more than 30 per cent of the time
(Eugene managed 33 per cent).
59
Turing’s observation regarding mimicry offers a route out of the impasse of AI and AGI, which is, after W. Grey Walter’s “mimicry of life”
via simple robots in the late 1940s, to consider intelligence as simply that
which exhibits interesting behaviour.
60
Based in Bristol, Walter was able to
construct mechanical tortoises that were phototropic (able to follow light)
and had a bump mechanism to allow them to change direction. His first
robots, Elmie and Elsie, were thus able to simulate autonomous motion
without human intervention, leading to a branch of robotics development
that did not need to concern itself with consciousness and the difficult
tasks of cognitivism. As Russell and Norvig point out in their guide to
modern thinking about AI, the standard approach to artificial intelligence
that is considered and evaluated by the Turing Test (if not actually by Turing himself), ”thinking humanly”, is only one of four approaches, including acting humanly, thinking rationally, and acting rationally. In practical
terms, it is probably the latter, also known as the ”rational agent”, which
tends to concentrate on intelligent behaviour in hardware and software,
that offers the most immediate benefits for future development.
61
The fact that this book is largely concerned with automation and
applied AI rather than strong or general artificial intelligence is by no
means to assert that AGI is impossible: complexity theory and notions
of biological emergence frequently demonstrate that complex systems
can arise from situations where there is not enough information, as in
termite colonies or traffic patterns. Applied AI has often resulted in more
useful models and theories, for example, the modular approach based
on Fodor’s observations in the 1980s that the human mind is largely
composed of task-specific modules,
62
the cognitive processes that allow
us to concentrate on winning (or losing) a game of chess. Fodor’s approach has been criticised, but it does have practical applications in the
field of AI research and robotics. Simulated micro worlds are easier to
understand (and program) than the entire universe of cognition, and this
is particularly evident in areas such as game playing where computers
are becoming better than humans. Rather than simple brute force – an
attempt to compute every possible move recursively – such controlled environment simulations work better via heuristics to build up the limited
range of moves that are possible in a given situation. As with playing
chess, we shall see in a later chapter that this approach to AI is also especially pertinent to certain forms of journalism.