Computing Machinery and Intelligence

Download Report

Transcript Computing Machinery and Intelligence

Computing Machinery and
Intelligence
Alan Turing
Brief Bio on Alan Turing:
• Born 1912, an Englishman, often called
the father of digital computing and artificial
intelligence.
• One of the most prominent allied code
breakers in WWII.
• Pioneered many early computing
techniques.
• Was convicted of homosexuality in 1952,
committed suicide at the age of 41 in 1954
under the influence of the chemical
castration medications that he was courtordered to take.
• In 2009, the British gov’t issued a formal
apology to Turing.
What’s at stake:
• What is at stake here is a claim about
what is possible in principle rather than
what is currently the case or what may or
may not be the case soon.
• Turing proposes a test which (he claims)
would provide us with reason to say that
any machine that passed it would be a
thinking thing.
The Test
• Turing proposes that the machine play an
imitation game. The machine is in one
room, and a human is in another. The
judges get to ask questions by means of
something like text messages or IM. The
machine’s responses to questions are
supposed to be able to be such that the
judges aren’t able to tell the difference
between the machine and the human
more than half the time.
Caveats against cheating:
• The kinds of “machines” that are disallowed from
the competition include:
– Humans born in the usual way
– Clones of humans
– Humans made in something other than the usual way
• The difficulty is in making a rule about the kinds
of machines disallowed that provides a definition
of ‘machine’ that doesn’t include people (Recall
the problems of vagueness that are intrinsic to
language).
Functionalism
• While Turing does not use this word, this is the term that
philosophers have applied to the kind of position that
Turing holds.
• Functionalism is a school of thought that is above all
committed to multiple realizability.
• Multiple realizability is the idea that the same functions
can be realized by multiple physical setups. Consider a
vacuum tube radio versus a solid state radio or a digital
versus an analog watch.
• According to the functionalist in this context, thinking is
something that can be done by persons as well as
(someday) machines.
• So Turing’s Test is supposed to determine whether
some given machine is the functional equivalent of a
human mind.
Counterarguments
• Be sure to consider which of these Searle
subscribes to and which he does not, and keep
track of how Turing answers them.
–
–
–
–
–
–
–
–
Theological Argument
The “Heads in the Sand” Objection
The Mathematical Objection
The Argument from Consciousness
Arguments from various disabilities
Lady Lovelace's Objection
Argument from Continuity in the Nervous System
Argument from Informality of Behavior
The Theological Argument
• Turing is not generally impressed with
theological arguments, but argues that the
second premise below is dubious (God
can do whatever God wants)
• The argument looks something like:
– Thinking requires a soul
– Mechanical things cannot have souls
– .: Mechanical things cannot think
The “Heads in the Sand” Objection
• This is an objection that many hold but
that does not survive much scrutiny.
• Turing does not specifically refute this
objection aside from pointing out that no
valid argument follows from this view:
– It would be scary if computers could think
– .: computers cannot think
The Mathematical Objection
• Goedel’s incompleteness theorems show that no
axiomatic system is self-proving (at least on axiom must
always be assumed)
• This is a limitation on any discrete state machine (what
remains to be shown is that such a limitation does not
apply to humans)
• Turing points out that while the proponents of the prior
two objections would not be interested in ANY proposed
criteria for machine intelligence, while the proponent of
this argument might entertain the Turing Test.
The Argument from Consciousness
• In order to be thinking, a being must be
“conscious” where consciousness requires some
form of self-awareness.
• Turing Responds:
– In that case the only way to perform an adequate test
would be to BE the machine and see if you’re
conscious, which applies not only to machines but
other people, too. (leads to solipsism)
– If a behavioral test is allowed though, it seems the
Turing Test could well satisfy this condition (see
example p. 530)
Argument from Various Disabilities
• Machines could never X (list of examples p. 530), and
various forms of thinking are presumably on the list.
• This springs from a set of inductive generalizations (all
machines encountered so far are “ugly, designed for a
limited purpose, when required for a different purpose
they are useless, the variety of behaviors of any one of
them are small, etc.)
• Such generalizations (because they are inductive) do not
preclude any exceptions to them.
Lady Lovelace’s Objection:
• Thinking requires either 1) indeterminism
or 2) originality.
• Turing Responds:
– 1) if indeterminism violates some law of
nature, then it remains to be shown how
humans violate those laws of nature
– 2) the Turing test would establish if machines
could ever “surprise us”.
Continuity in the Nervous System
• This is a rather sophisticated objection.
– The human nervous system is not a discrete state machine
– No discrete state machine can mimic human nervous structures
– .: Since computers are discrete state machines, no computer can
mimic human nervous structures.
• Turing points out that any discrete state machine can
approximate the output of a non-discrete state machine
(like a differential analyzer) such that nobody could tell
the difference only from the output. This is the point of
the Turing Test; it falsifies p2 above.
Informality of Behavior
• Turing summarizes the argument: “If each man
had a definite set of rules of conduct by which he
regulated his life he would be no better than a
machine. But there are no such rules, so men
cannot be machines”
• Convert this to categorical form and:
– All those with definite rules of conduct are machines
– No men are those with definite rules of conduct
– .: No men are machines
(continued)
• Turing points out that the above argument is
invalid due to an undistributed middle term.
• In any case, If one substitutes “laws of behavior”
for “principles of conduct” the argument comes
out alright, and must either show that there are
no such laws of behavior for humans or that
human are (at some level) machines.
• Turing points out that there can be no basis for
concluding that there are no laws of behavior.
Learning Machines
• Turning concedes that programming
something to be able to pass the Turing
test “out of the box” may not be the most
likely approach. Instead building
something with an ability to learn and
teaching it like a human child might be
more effective.
Overall approach:
• Turing’s argumentative approach has two
parts:
1. Present a positive account of what criteria
are sufficient for thinking
2. Defend the idea that machines could be
thinking things against a set of general
principles that would exclude machines from
the category of thinking things.