Transcript What is Cognitive Science?

Artificial Intelligence
1. What is AI?
2. Issues in AI
An Overview
- AI is a science of making intelligent machines
- Intelligence is a type of computation:
What is a computation?
 Turing Machines
- How do we know if a machine is intelligent or
not?
 Turing Test
1. What is AI?
• Artificial intelligence is the science and
engineering of making computer programs that exhibit
characteristics of human intelligence.
• Scientific aim: To understand the requirements for and
mechanisms of human, animal, machine, robotic
intelligence
• Engineering aim: To apply such knowledge in building
useful artifacts (machines & robots) capable to do
things done by humans or animals
• What is intelligence?
- Intelligence is the computational part of the ability to solve
problems and achieve goals in the world in an efficient manner
(McCarthy)
- ‘Computational part .. to do … efficiently’  Algorithm
(e.g.)
Tower of Hanoi Problem:
Tic-tac-toe:
Chess:
No. of all possible board states: 10120!!
- Combinatorial explosion problem
- Blind search – intractable
Tower of Hanoi
Branches of AI
Knowledge representation
- Processing information about and representing facts about the
world in some abstract way
Pattern recognition
- Extracting knowledge from images (e.g., letters, face, X-ray
data, satellite photos)
Reasoning and inference
- Deriving conclusions from premises or incomplete
observations (e.g., logical deduction, math theorem proving,
medical diagnosis, stock market/weather forecasting)
Machine learning
- Improving performance from experience (e.g., rule induction &
adaptive modification)
Planning
- Planning a complex sequence of actions (e.g., playing chess)
Natural language processing
- Production and interpretation of spoken and written language
-
Pattern recognition
Knowledge representation
Reasoning & inference
Machine learning
Applications of AI
Computer vision
- IRIS (biometric identification device), detection of
forgeries, chip inspection
Expert systems
- MYCIN (medical diagnosis), HYPO (legal reasoning),
auto pilot, intelligent tutoring system
Game playing
- IBMS’ Deep Blue (search 2m positions per sec)
Speech recognition
- Dragon Naturally Speaking
Robotics
- robot moles in Mars exploration
Dartmouth Workshop
(1956)
- Summer workshop that officially launched the field
known as ‘Artificial Intelligence’ (named by McCarthy)
- Participants included: McCarthy (Stanford), Minsky
(MIT), Shannon (Lucent), Newell (CMU), Simon (CMU)
General Problem Solver (GPS)
(Newell & Simon, 1960’s)
- Landmark computer program that solves simple
problems/puzzles (e.g., Tower of Hanoi) and even comes
up with proofs for mathematical theorems
- Based on a general problem solving strategy called the
‘mean-ends analysis’ (work backward from the goal to
decide on what action(s) will help you achieve in which
goals are decomposed into subgoals in a recursive
fashion)
Weak AI vs Strong AI
in the Study of Mind
(Searl 1980)
Weak AI:
- “The principal value of the computer in the study of
mind is that it gives us a very powerful tool.”
Strong AI:
- “An appropriately programmed computer literally has
cognitive states and therefore explains how the
human mind works.”
2. Issues in AI
Issue #1:
What is a computation?
Turing Machines
(Turing, 1937)
- A Turing Machine, an idealized, mathematical abstraction of a
digital computer, consists of
(1) 1-dim tape of cells of unlimited length
(written on each cell is a symbol from finite alphabet)
(2) read/write head
(3) control (action) table or program
Control program:
“Condition” (IF)
Current
state
“Action” (THEN)
Symbol
read
Symbol
to write
Move the New state
head to
S1
‘0’
‘1’
Left
S3
S2
‘1’
‘1’
Right
S1
S3
‘1’
‘0’
Left
S2
- State of head: {S1, S2, S3}
- Binary alphabet on tape: {0,1}
- Movement of head: {Left, Right}
A Turing machine that computes:
“2 x 4 = 8”
(Turing’s) Definition of computation
- “A function is said to be computable if it can be implemented on
a Turing Machine.”
- Such functions are called Turing computable functions
(e.g., f(x) = 0; natural log e; +/x; if-then)
- Roughly speaking, a function or task is computable if its
solution can be found in “finite” time (or polynomial time).
- A problem in which the time required to solve grows
exponentially as the problem size grows said to be
uncomputable (i.e., unsolvable), thereby requiring “infinite”
time to solve  NP-hard problem
(e.g., Traveling Salesman Problem)
Traveling Salesman Problem
16-city problem
A candidate solution
Universal (Turing) Machine
- Turing also showed that it is possible to design a single
Turing machine that can simulate any Turing machine.
Such a machine is called a Universal Turing Machine
Church-Turing Thesis:
In essence, “A Universal Turing Machine can compute
any non-NP-hard problem.”
(e.g.)
- Programmable computers (PC, MaC)
von Neumann Machine
- program – control/action table unit
- CPU – read/write head unit
- RAM - tape
- DNA (biological computation device)
Issue #2:
How do we know if a machine is
intelligent or not?
Turing Test
(Turing, 1951)
- First attempt to define an operational definition of
intelligence
- Turing defined intelligent behavior as the ability to
exhibit human-like performance, sufficient to fool an
interrogator in an “imitation game”
Can the Turing test be a definition of
intelligence?!%
1. A computer may pass the test but without ‘real’
understanding of the conversation that took place (e.g.,
Searl’s Chinese Room)
2. Many ‘real’ human beings might fail the test.
3. A computer often exhibits intelligence without being a
conversational partner (e.g., autopilot)
Chinese Room
(Searle, 1980)
- Thought experiment developed as an attack on the
Turing Test (againt Strong AI)
- Showed that in theory, it is possible to create a system
that exhibits intelligent output without understanding
(i.e., in the absence of mind), thus passing the Turing
test
- Would it be practically possible to build such a system?
Why or why not?