Title: Basic Questions
1Basic Questions
- What is intelligence?
- Is it natural category? Is it always
characterized by a common set of properties? - Can it be described by a set of abstract laws?
- Can intelligence exist outside the human mind?
IntroductionBasic Questions
2What is Intelligence?
- How is intelligence measured?
- Can it be defined without relating it to human
intelligence? - Is intelligence a process or knowledge?
- Is intentionality a necessary component?
IntroductionWhat is Intelligence?
3Measuring Intelligence
- Duplicate or simulate human intelligence
- Is all measurement subjective, depending on the
assumption of goals? - Problem of intentionality
- If it behaves as if it understands can we assume
it understands? What is understanding? - Turing test strictly behavioral
IntroductionMeasuring Intelligence
4Study of the Mind
From Franklin,Artificial Minds
Top Down
Psychology
Artificial Intelligence
Analytic
Synthetic
MIND
Neuroscience
Mechanism of mind
Bottom Up
5What is AI?
From RussellNovig(2003),AI, A Modern Approach,
Prentice Hall
IntroductionWhat is AI?
6Theory of Computability
- Church Thesis Any procedure that can be
described explicitly can be computed with
recursive functions. - Turings view framed in terms of a class of
automata, Turing machines. - Thesis claims computable process can be
programmed, i.e. is equivalent or less powerful
than a Turing machine.
IntroductionTheory of Computability
7Recursive Functions
- There is some method of computing three things
- 1. The zero function which will yield a value of
zero given any input of natural
numbers(0,1,2,3..) - 2. The successor function that yields the number
that follows any natural number - 3. The set of identity functions that given the
input of string of input arguments returns the
identical value of one of them as specified by
its position in the string
IntroductionRecursive Functions
8Recursive Functions (continued)
- There is a rote procedure for determining in a
finite time, the output of a function. - There is a finite set of operations that can be
applied to a given input and applied again to the
successive result to yield the functions output
in a finite time - Describes what is meant by an algorithmic
solution, equivalent to a Turing machine
IntroductionRecursive Functions(continued)
9Doctrine of Functionalism
- Presented by Cambridge psychologist Kenneth Craik
in The Nature of Explanation - States that the human mind translates aspects of
the external world into internal representations,
processes these representations and translates
these representation back into behavior - Mental processes are essentially computations
that manipulate and interpret representations of
the world
IntroductionDoctrine of Functionalism
10Can the Mind be Programmed?
Is the mind explainable?
no
yes
There IS a ghost in the machine!
Is the Church/Turing Thesis correct?
no
yes
The mind may depend on something noncomputable
Is functionalism correct?
no
yes
The mind can be modeled
The mind is computable
IntroductionCan the Mind be Programmed?
11The Nay Sayers
- John Searle says the idea of a non-biological
machine being intelligent is incoherent. - Hubert Dreyfus says that AI is impossible because
world of the subjective is more important than
world of the objective - Joseph Weizenbaum says the idea is obscene,
anti-human and immoral.
IntroductionThe Nay Sayers
12Alan Turing (1912-1954)
- Born in London to family employed in the Indian
Civil Service - Found school an almost incomprehensible
experience - Devastated by death of best friend, Christopher
Morcom
Alan Turing Early Life
13Alan Turing Decidability
- As an undergraduate at Kings College Cambridge,
working on probability theory, became interested
in question of decidability - Given a mathematical proposition can an algorithm
be found which will decide if it is true or false
Alan Turing Decidability
14Alan Turing Turing Machine
- Expressed his results in a theoretical machine,
the Turing machine - Alonzo Churchs paper published first in 1936
- His contribution was link between logical
instructions, action of the mind and a machine
Alan Turing Turing Machine
15Alan Turing Code Breaker
- Turing took up full-time work at the wartime
cryptanalytic headquarters, Bletchley Park - Enigma methods used in German Naval
communications were generally regarded as
unbreakable, but Turing cracked the system at the
end of 1939 - He planned the embodiment of the Universal Turing
Machine in electronic form, or in effect,
invented the digital computer
Alan Turing Code Breaker
16Alan TuringFounder of Computer Science
- Turing was almost uniquely in possession of three
key ideas - his own 1936 concept of the universal machine
- the potential speed and reliability of electronic
technology - the inefficiency in designing different machines
for different logical processes.
Alan Turing Founder of Computer Science
17Alan Turing Turing Test
- Computing Machinery and Intelligence appeared in
Mind 1950 - Turing's reasoning was that, assuming
intelligence could only be determined
behaviorally, then any agent that behaved like an
intelligent agent was intelligent - Conversation was the key to judging intelligence.
Alan Turing Turing Test
18Turing Test Requirements
- Natural language processing
- Knowledge representation to store what it knows
- Automated reasoning to use the stored information
to answer questions - Machine learning to adapt to new circumstances
Alan Turing Turing Test
19Turing Test
Interrogator The first line of the sonnet reads
"How shall I compare thee to as summer's day".
Would not a spring's day" do just as
well? Witness It wouldn't scan Interrogator
How about "a winter's day". That would scan all
right. Witness Yes, but nobody would like to be
compared to a winter's day. Interrogator Would
you say Mr..Pickwick reminded you of
Christmas? Witness In a way. Interrogator Yet
Christmas is a winter's day and I doubt that Mr..
Pickwick would mind the comparison. Witness I
don't think you are being serious. By a winter's
day one means a typical winter's day, not a
special one like Christmas.
IntroductionTuring test
20Turing TestObjection 1
- "The Theological Objection" God created every
man and woman and gave them a soul - but did not
do the same for machines or animals. - Turing dismisses this idea as out of hand, and
refutes the objection by showing that different
religions have different beliefs. Moslems for
example believe that women have no souls.
Turing TestObjection 1
21Turing TestObjection 2
- 2 "The 'Heads in the Sand' Objection" The
thought of machines that have intelligence is too
dreadful to even think about. - Turing asks why we think that Man is superior to
the rest of creation, but does not spend any time
in his paper refuting the argument, instead
dismissing it out of hand.
Turing TestObjection 2
22Turing TestObjection 3
- 3 "The Mathematical Objection" This rather
abstract argument involves a machine of infinite
capacity. It is claimed by some mathematicians
(Godel, Church, Rosser and Turing himself) that
there will always be some questions that the
computer simply cannot answer. - Turing's refutes this by asking why humans cannot
have certain questions that are unanswerable.
Turing TestObjection 3
23Turing TestObjection 4
- 4 "The Argument from Consciousness" 'Not until
a machine can write a sonnet or compose a
concerto because of thoughts and emotions felt,
could we agree that machine equals brain. - This argument renders the Turing test somewhat
ineffective because testing for creativity via a
terminal is restrictive in many ways. He also
argues there is no way to test for consciousness.
It is only possible to be certain of one's own
consciousness.
Turing TestObjection 4
24Turing TestObjection 5
- 5 "Arguments from Various Disabilities"
Although machines can undoubtedly do many things
that may appear to be intelligent, they can
never 'Be kind, beautiful, friendly, have
initiative, have a sense of humor, tell right
from wrong, make mistakes, fall in love, enjoy
strawberries and cream...', etc. - Turing claims that one day machines may be able
to complete such tasks, but at the moment
sufficient research into why man has such
capabilities has yet to be undertaken.
Turing TestObjection 5
25Turing TestObjection 6
- 6 "Lady Lovelace's Objection" Computers can do
only what they are told. - Turing argues that one day it may be possible to
construct electronic equipment that will 'think
for itself'. He then argues that perhaps no-one
can think of 'new' things, in other words,
everything we think is generated from what we
already know - we are simply adapting it.
Turing TestObjection 6
26Turing TestObjection 7
- 7 "Argument from Continuity in the Nervous
System" The nervous system is clearly different
from a computer in the way it works at neuron
level. The will never be modeled by a computer. - The way in which the machine/human operates is
not visible to the interrogator, and - according
to Turing - the same results should, one day, be
obtainable.
Turing TestObjection 7
27Turing TestObjection 8
- 8 "The Argument from Informality of Behavior"
It is impossible to model the behavior of humans
because there is so much variation from man to
man. - Turing argues that it is perfectly possible for a
computer to produce output from a certain input
that would be impossible to predict without many
1000s of years of observation. Who is to say
that, with an equal amount of observation, the
behavior of humans does not become predictable.
Turing TestObjection 8
28Turing TestObjection 9
- 9 The Argument from Extra-Sensory Perception It
involves a human using ESP to do tasks such as
guessing the suit of a card that the interrogator
is holding. By using ESP, they will be able
(according to Turing) to get statistically more
guesses correct than the computer. Turing claims
that the evidence for telepathy is
'overwhelming'. - Turing suggests that the competitors should be
placed in a 'telepathy-proof room'.
Turing Test Objection 9
29Turing Test Loebner Prize
- 1991 tournament billed as the first actual
administration of the Turing test - Added two rules
- Limiting the topic.
- Limiting the tenor (no trickery or guile)
- 10 contestants 6 computers and 4 humans
- Won by Eliza-type system
30Turing Test Criticism
- Provides weak equivalence where two systems are
computing the same function (or generating the
same external behavior), but that they are using
different procedures to do so - No restrictions on the judge
- Only tests for complete success
- Not a sufficient condition for intelligence
Alan Turing Turing Test
31Turing Test Criticism
From Turing Test Considered Harmful
- Doesnt take into account vision and robotics,
places all emphasis on natural language - Hard to completely define the experimental
conditions and has no stable end point - Test for complete success only and can be fooled
by an artificial con artists - There is a species detection aspect that is silly
like to try to paste feathers and a beak on
airplanes - Continues to give AI a bad name
Turing test Hayes and Ford
32Turing Test Criticism
From Turing Test Considered Harmful
- An adequate test should focus on more general
principles of cognition - All of computer science is AI
- Perhaps Turing deliberately picked a test of
sexual identity because of his own problems with
English society in the Fifties
Turing Test Hayes and Ford (cot)