Title: The Birth of the Computer
1The Birth of the Computer
2Foundations
- The theory of computation and the practical
application it made possible the computer was
developed by an Englishman called Alan Turing.
3Alan Turing
- 1912 (23 June) Birth, Paddington, London
- 1931-34 Undergraduate at King's College,
Cambridge University
- 1932-35 Quantum mechanics, probability, logic
- 1936 The Turing machine, computability,
universal machine
- 1936-38 Princeton University. Ph.D. Logic,
algebra, number theory
- 1938-39 Return to Cambridge. Introduced to
German Enigma cipher machine
- 1939-40 The Bombe, machine for Enigma decryption
- 1939-42 Breaking of U-boat Enigma, saving battle
of the Atlantic
- 1946 Computer and software design leading the
world.
- 1948 Manchester University
- 1949 First serious mathematical use of a
computer
- 1950 The Turing Test for machine intelligence
- 1952 Arrested as a homosexual, loss of security
clearance
- 1954 (7 June) Death (suicide) by cyanide
poisoning, Wilmslow, Cheshire.
- from Andrew Hodges http//www.turing.org.uk
/turing/
4The Decision Problem
- In 1928 the German mathematician, David Hilbert
(1862-1943), asked whether there could be a
mechanical way (i.e. by means of a fully
specifiable set of instructions) of determining
whether some statement in a formal system like
arithmetic was provable or not. - In 1936 Turing published a paper the aim of which
was to show that there was no such method.
- On computable numbers, with an application to
the Entscheidungs problem. Proceedings of the
London Mathematical Society, 2(42)230-265).
5The Turing Machine
- In order to argue for this claim, he needed a
clear concept of mechanical procedure.
- His idea which came to be called the Turing
machine was this
- (1) A tape of infinite length
- (2) Finitely many squares of the tape have a
single symbol from a finite language.
- (3) Someone (or something) that can read the
squares and write in them.
- (4) At any time, the machine is in one of a
finite number of internal states.
- (5) The machine has instructions that determine
what it does given its internal state and the
symbol it encounters on the tape. It can
- ? change its internal state
- ? change the symbol on the square
- ? move forward
- ? move backward
- ? halt (i.e. stop).
6Current state 1 If current state 1 and curre
nt symbol 0 then new state 10 new symbol 1
move right
0
1
0
1
1
1
7Current state 10 If current state 1 and curr
ent symbol 0 then new state 10 new symbol
1
move right
1
1
1
1
1
81
1
1
1
1
9Functions
- It is essential to the idea of a Turing machine
that it is not a physical machine, but an
abstract one a set of procedures.
- It makes no difference whether the machine is
embodied by a person in a boxcar on a track, or a
person with a paper and pencil, or
- a smart and well-
- trained flamingo.
10Turings Theorem
- In the 1936 paper Turing proved that there are
general-purpose Turing machines that can
compute whatever any other Turing machine.
- This is done by coding the function of the
special-purpose machine as instructions of the
other machine that is by programming it. This
is called Turings theorem. - These are universal Turing machines, and the idea
of a coding for a particular function fed into a
universal Turing machine is basically our
conception of a computer and a stored program. - The concept of the universal Turing machine is
just the concept of the computer as we know it.
11Computers and the Mind
12Can Machines Think?
- In Computing machinery and intelligence,
written in 1950, Turing asks whether machines can
think.
- He claims that this question is too vague, and
proposes, instead, to replace it with a different
one.
- That question is Can machines pass the
imitation game (now called the Turing test)? If
they can, they are intelligent.
- Turing is thus the first to have offered a
rigorous test for the determination of
intelligence quite generally.
13The Turing Test
- The game runs as follows. You sit at a computer
terminal and have an electronic conversation. You
dont know who is on the other end it could be a
person or a computer responding as it has been
programmed to do. - If you cant distinguish between a human being
and a computer from your interactions, then the
computer is intelligent.
- Note that this is meant to be a sufficient
condition of intelligence only. There may be
other ways to be intelligent.
14Functionalism
- In 1960 Hilary Putnam (b. 1926) used the idea of
a Turing machine to develop a view about the
mind.
- This view is called functionalism.
15Functionalist Analysis
- Functionalism holds that a mental state can be
exhaustively characterized by describing
- (a) the input that typically brings it about
- (b) its relations to other mental states and
- (c) the output it tends to produce.
- So, for example, we can characterize pain as the
mental state
- (a) typically brought about by damage to the
body
- (b) that is closely related to anger, regret, and
so on and
- (c) that tends to produce shouting, crying, and
the like.
16Computationalism
- An extension of this idea, due primarily to Jerry
Fodor (b. 1935) and one that Fodor traces back
to Turing as well is the view that thought is
essentially a series of moves in a Turing
machine. - It is, in effect, the manipulation of symbols on
an abstract tape the performing of
computations.
- This view is called computationalism, and it is
the only real theory weve got about what thought
is.
Central to this view is the idea that the form of
thought (its syntax) e.g., the 0s and 1s on
the tape determines its content or meaning (its
semantics).
17Artificial Intelligence
18The Church-Turning Thesis
- Turing, and a logician called Alonzo Church
(1903-1995), independently developed the idea
(not yet proven by widely accepted) that whatever
can be computed by a mechanical procedure can be
computed by a Turing machine. - This is known as the Church-Turing thesis.
19AI The Argument
- Weve now got the materials to show that AI is
possible
- P1 Any function that can be computed by a
mechanical procedure can be computed by a Turing
machine. (Church-Turing thesis)
- P2 Thinking is nothing more than the computing
of functions by mechanical procedures (i.e.,
thinking is symbol manipulation).
(Functionalist-Computationalist thesis) - C1 Therefore, thinking can be performed by a
Turing machine.
- P3 Turing machines are multiply realizable. In
particular, they can be realized by computers,
robots, etc.
- ? It is possible to build a computer, robot,
etc. that can think. That is, AI is possible.
20Searles Challenge The Chinese Room Argument
21John Searle (b. 1932)
22A Distinction
- Strong AI The research program whose aim is to
produce entities with real intelligence.
- Weak AI The research program whose aim is to
produce entities that simulate real
intelligence.
- Compare the weather a computer that models
weather patterns is simulating the weather rather
than recreating it.
- Searles argument is directed at strong AI.
23Intentionality
- Searle claims that it is not possible (with
standard computers) to build a machine that has
intentionality.
- Intentionality is the property of a mental state
that makes it about something in the world.
- Exactly what intentionality amounts to, however,
is not crucial to the argument.
- For our purposes, Searles aim is to argue that a
computer cant have some of the features we
associate with thought.
- In particular, it doesnt have states that are
meaningful that have semantic properties. In
the case well look at, the computer doesnt have
understanding.
24Searles Challenge
- Searles argument takes as an example a
story-understanding program from 1977 due to
Schank and Abelson.
- But his argument applies to any form of strong AI.
25The Chinese Room
- The Chinese Room is a thought experiment.
- Thought experiments are used not only in
philosophy what matters is not whether the
experiment is possible but whether it is
coherent. If it is coherent, what does it show?
26Elements
- The room contains an English speaker who does not
understand Chinese.
- There is a slot in the door through which symbols
in Chinese pass into and out of the room.
- The person in the room has a rule-book that tells
him what symbols to produce in response to
symbols introduced into the room when you get a
squiggle-squiggle, produce a squoggle-squoggle.
27The Turing Test Again
- The person in the room gets good enough to
convince a native Chinese speaker that he
understands Chinese.
- In other words, the Chinese Room passes the
Turing test (conducted in Chinese).
- But, of course, the man in the room doesnt
understand Chinese!
28First Argument
- With these ideas in hand, we can construct an
argument concerning the Turing test
- (P1) Passing the Turing test is supposed to be a
sufficient demonstration of intelligence.
- (P2) The person in the Chinese Room passes the
Turing test.
- (P3) However, the person in the room does not
have the relevant intelligence (the ability to
understand Chinese).
- ? (C) The Turing test is not an adequate test of
intelligence.
29Syntax and Semantics
- The more general and powerful argument made
possible by the Chinese Room case concerns the
very possibility of strong AI given current
computers. - That argument depends on the distinction between
syntax and semantics which weve already
encountered.
- Recall that the syntax of a string of symbols
concerns the rules for ordering symbols in the
right way.
- But these rules dont tell you enough to know
what the symbols mean. That is, they dont give
you semantics.
30Illustration
- Consider the sentence
- All members of the NDP drink Boréale.
- It is easy to tell that this is a good sentence.
The right sort of words (e.g. nouns, verbs, etc.)
are used unlike
- All members of the NDP Boréale.
- and they are used in the right order unlike
- Members Boréale all of drink NDP the.
- But you dont know what this sentence means.
- This shows that syntactical properties dont give
you semantic properties informally grammar
doesnt reveal meaning.
31Second Argument
- We can now formulate Searles main argument
- (P1) Computer programs are purely syntactical
things they operate on meaningless symbols.
- (P2) Minds have semantic properties they have
states that have meanings.
- (P3) The Chinese room shows that syntax is not
the same thing as, nor is it sufficient for,
semantics.
- ? (C1) Therefore, computer programs are not the
same thing as, nor sufficient for, minds.
- ? (C2) Therefore, AI is impossible (with
computers as we know them).
32Some Remarks
- It is important to see that the argument does not
(and was not intended) to show that machines
cant think, or that artificial intelligence of
any kind is impossible. - Searle accepts both of these claims.
- Rather, it is designed to show that a computer
cannot have genuine psychological properties
simply in virtue of its input and output
relations that is, simply in virtue of its
program. - On Searles view, thinking requires being a
physical thing with special causal properties.
- A machine that was as intelligent as the brain
would have to duplicate the causal properties of
the brain.
33Learning Machines
- There are computers that operate on rather
different principles from Turing machines.
- These machines, usually called connectionist or
neural networks, change their structure in
response to input and output. They learn.
- Whether these kinds of machines could be truly
intelligent or not depends on how you interpret
what they are doing.
- Some people think they are just special kinds of
Turing machines, in which case, Searles argument
applies to them too.
- Other people think they dont have programs so
they fall outside the scope of Searles argument.
34Functionalism and Computationalism
- Finally, if Searle is right, then both
functionalism and computationalism, as theories
of the mind, are false.
- The man in the Chinese room exhibits the right
functions and manipulates the right symbols to
behave as if he understood Chinese. But he
doesnt. - Merely exhibiting particular functional relations
(including manipulating symbols according to
particular rules) is not enough for thought.
- Computation alone is not sufficient for thought.
35Replies
- Searle considers a number of possible objections
to his argument in the original paper.
- In the last twenty-two years a huge literature
has grown up around the argument.
- (Searle says in an article from 2002 I have
already responded to more criticisms of the
Chinese Room Argument than to all of the
criticisms of all of the other controversial
philosophical theses that I have advanced in my
life.) - Well consider a few important lines of argument
here.
36The Systems Reply
- According to this line of thought, Searle commits
the fallacy of taking the part for the whole.
- The argument says that the man in the room
doesnt understand Chinese, and he concludes from
this that the Chinese room as a whole doesnt.
- But, so the objection runs, the man, the
rule-book, the symbols, and all the other
relevant elements of the function of the Chinese
room do understand Chinese.
37Searles Rejoinder
- Searle has two main replies.
- First he says that the man in the room could
internalize all of the relevant parts of the
procedure he could memorize the rule-book he
could work outdoors and so on. - Whatever was in the Chinese room initially is now
entirely in the mans mind. But he still doesnt
understand Chinese.
- Second, he claims that it is utterly implausible
to say that while the man doesnt understand
Chinese, the man plus some bits of paper and a
rule-book do.
38The Robot Reply
- There is no understanding in the Chinese room
because the relations between the symbols
manipulated by the man in the room and the things
they refer to are missing. - If we created a robot that had the same symbols
and rule-book but and could interact with the
environment, the robot would understand Chinese.
39Searles Rejoinder
- Searle notes, first, that this reply already
concedes that the basic premise of AI is false
syntax is not enough for semantics. You need
causal interaction too. - Second, Searle asks us to imagine a sort of
mobile Chinese room a robot it gets input from
the environment and produces behavior.
- So what? The man inside the room is still only
engaging in transactions with symbols (even if
they are symbols representing the outside world).
- The robot can pass a more complicated Turing
test, but the Turing test is always inadequate to
measure intelligence The problem remains.
40The Brain Simulator Reply
- Suppose that instead of manipulating symbols in
Chinese the man in the room simulates the neuron
firings of a brain that is in the process of
understanding some sentence in Chinese (or
producing a sentence in Chinese). - Surely then the Chinese room would understand
Chinese because it is doing just what the brain
of a Chinese speaker does.
41Searles Rejoinder
- Adapt the Chinese room so that it contains an
elaborate set of water pipes with valves.
- In response to input in Chinese, the man in the
room opens and shuts some of the valves in a
particular way where each valve opening or
closing simulates what some neuron does. As a
result, out pops the correct response in
Chinese. - The man still doesnt understand Chinese nor do
the pipes!
- Why? Because the Chinese room merely simulates
the brain. It captures the structure of the
brains activity just as a program captures the
structure of what a mind does. But in both cases,
the causal relations arent there, so there is no
mind.
42The Combination Reply
- Take all of the suggestions weve just looked at
a computer in a robot running a program that
simulates what the brain does and you would
have genuine understanding.
43Searles Rejoinder
- None of the features of this solution worked
separately, and there is no reason to think that
putting them together produces something new.
44The Other Minds Reply
- How do we know in ordinary life whether someone
understands Chinese? If they behave as if they
do if they can speak Chinese.
- Since the Chinese room behaves as if it speaks
Chinese, then one ought to say that it
understands Chinese for the same reason as one
says that a Chinese-speaker understands Chinese.
45Searles Rejoinder
- Notice first that this is just a defense of the
Turing test. Since Searles argument purports to
demonstrate that the Turing test is inadequate,
this begs the question against Searle. - In addition, Searle says that it misses the
point. The important issue is not how we decide
whether someone understands Chinese but what is
involved in understanding Chinese. - The argument shows that computation is not enough
for understanding. If one were to attribute
understanding to the Chinese room, one would get
the wrong answer even though the criterion used
gives the right answer in the case of Chinese
speakers.
46The Many Mansions Reply
- Even if Searle is right that programming alone is
insufficient for AI, there might be other ways of
producing intelligent machines.
47Searles Rejoinder
- This also misses the point because it defines AI
in a trivial way as whatever produces artificial
intelligence.
- The actual hypothesis of the strong AI research
program is that intelligence can be produced by
programming alone. The Chinese room argument
shows that this is false. - If the hypothesis is trivialized, then the
Chinese room argument doesnt apply, but only
because, in effect, AI has become a self-sealing
doctrine.
48Concluding Note
- There is wide consensus in cognitive science and
philosophy that the Chinese room argument is
wrong or worse badly flawed and not deserving
of its fame. - But the argument continues to attract serious
criticism.
- And it continues to elicit disagreement among its
critics about exactly where it goes wrong!
- Dead views typically do neither.
49Later Searle Syntax and Biology
50Syntax Again
- The Chinese Room argument is supposed to show
that syntax is not sufficient for semantics and
therefore not sufficient for minds.
- Searle assumes that the syntactical properties of
a computer are uncontroversial.
- In some later work, however, he argues that even
the notion of the syntax of computer
representations is problematic.
- He infers from this something stronger even than
the conclusion of the Chinese Room argument
artificial intelligence is not only not possible
(with current computers) it is incoherent!
51Observer-Dependence
- The basic idea on which Searles argument depends
is the distinction between something being
observer-independent and observer-dependent.
- Something is observer-independent if that thing
would exist even if there was no one to observer
them.
- Something is observer-dependent if its existence
requires the existence of human observers.
- Some observer-independent things are mass,
molecules, cells.
- Some observer-dependent things are political
parties, money, hockey.
52Syntax and Observers
- Is syntax an observer-independent or
observer-independent property?
- Consider again the sentence
- All members of the NDP drink Boréale.
- exhibits correct syntax only relative to the
rules of English.
- But of course if there were no human beings,
English wouldnt exist, nor would English
syntax.
- Clearly, then, syntax is an observer-relative
feature of symbols.
- As Searle says Syntax is not intrinsic to
physics.
53Computation Again
- The idea, roughly, is that the words in English
only have syntactic properties because people
treat them as symbols rather than as mere marks
on paper or sounds. - What follows from this?
- Computers manipulate symbols according to
syntactic rules.
- Therefore, the notion of computation is a
syntactic one.
- But if syntax is observer-relative, and
computation is defined syntactically, then
computations are also observer-dependent
entities.
54Semantics Again
- What about semantics? Is it observer-dependent or
observer-independent?
- The fact that I understand the sentence
- All members of the NDP drink Boréale.
- does not depend on there being an observer. I
would understand that sentence even if there was
no one in the world but me and I was not
observing myself understanding the sentence. - So the semantic properties of the mind seem to be
observer-independent.
55The Argument
- With these further ideas about syntax and
semantics, Searle can make a final assault on
AI
- (P1) Computation is observer-dependent.
- (P2) Minds have semantic properties.
- (P3) Semantic properties of minds are
observer- independent.
- ? (C) Therefore, computer programs are not
necessary for indeed, not even relevant to
minds.
- (Recall that the conclusion of the Chinese Room
argument was that computer programs are not
sufficient for minds.)
56Language Again
- Recall that at the beginning of the course we
distinguished different things that one can do
with language linguistic acts, speech acts, and
conversational acts. - Linguistic acts involve saying something
meaningful.
- If Searle is right, then computers cant perform
linguistic acts because they cant attribute
meanings to language.
- Indeed, even the basic feature of linguistic acts
putting words together in the right order is
dependent on the intentions of a programmer.
- So if Searle is right, then computers cant do
even the most basic things with language and
cannot therefore be anything like human minds.
57The Bottom Line
- Computer programs are not even in the running to
be the basis for minds.
- They are just the wrong kind of thing.
58Searles Alternative
- Well, if programs are not relevant, what is?
- Or, as it is sometimes put, what could cognition
be if not computation?
- In brief, Searle thinks that mentality is a
biological phenomenon.
- He calls this view biological naturalism.
59The Basic Idea
- Think of water molecules. Each molecule has a
number of physical properties molecular weight,
charge, and so on.
- But when youve got a lot of molecules together,
new properties emerge liquidity, transparency,
and so on.
- The brain is made up of lots of neurons each of
which has a number of physical properties.
- When taken together, however, new properties
emerge understanding, consciousness semantic
properties.
- Mental properties are properties of complex
biological systems.
60The Future of AI
- Where does that leave AI?
- According to Searle, we know that AI is possible
brains are machines and they are intelligent.
- So if we want to create AI, we have to learn (1)
how brains work and (2) how to create artificial
brains.
- There may be other kinds of systems that can
think too, but we have no idea what they might be
like.
- Our best bet, therefore is the brain. Thus, the
future of AI lies in neurobiology.