Transcript print version - Homepages | The University of Aberdeen

Cognitive Science
Definition:
“the scientific study either of mind or of intelligence”
 Essential Questions
 What is intelligence?
 How is it possible to model it computationally?
 Takes ideas from






Psychology
Philosophy
Linguistics
Neuroscience
Artificial Intelligence / Computer Science
Maybe also minor contributions from:
 Anthropology, Sociology, Emotion studies,
Animal Cognition, Evolution
Origins of Cognitive Science
 Psychology of the early 20th century was dominated by “behaviourism”
 Everything should be treated as a behaviour
 “…purely objective experimental branch of natural science.”
- John B. Watson
 Goal: prediction and control of behaviour
 “Introspection forms no essential part of its methods” - John B. Watson
 Should not have to describe things in terms of “hypothetical” internals
 Such as the “mind”
 “Consciousness” not an appropriate question for scientific inquiry
 This changed around 1950s
 Partly as a result of investigations in Artificial Intelligence,
partly changing trends
 People started talking about
 Theories of mind
 Internal representations
 Computational procedures
 Term “Cognitive Science” born in 1973
 Came out of AI - Christopher Longuet-Higgins comment on “Lighthill report”
Cognitive Science – Information Processing
 Cognitive Science views the mind as an information processing system
 This is also called the computational view
 From this perspective: a human mind’s activity consists of





Receive information
Store information
Retrieve information
Transmit information
Transform information
 Example: a musician improvising





Listen to many tunes
Remember them
Find similarities
Come up with rules that say what sounds good together
Use those rules in real-time while playing
Understanding Information Processing Systems
1.
We attribute non-behavioural properties to the system




2.
Representation: information in the system can represent real things


o
o
3.
We say that it has a purpose, goals or desires
We say that it has internal beliefs and knowledge and competence
We attribute meaning to its external behaviour and internal information
We treat other humans like this all the time, call it folk psychology
For example: symbols could represent objects and relationships
This would allow a clear separation of what and how
Alternatively: it could be a messy representation
what and how tangled together
It has procedures for processing information





We call these procedures algorithms in computer speak
Describes how it does what it does
A clear set of steps that need to be followed
Like the recipe for making a cake
Like the instructions for long multiplication
Three Levels in Information Processing Systems
(Marr’s three levels)
What
What information is coming in?
What information is outputted?
What is the relationship?
(also explains why it’s important)
Representation ties together
How
Procedure/Algorithm
– clear set of instructions
(how to process the inputoutput)
Physical
Implementation
Must be physically carried out
–Man with paper and pen
–Mechanical computer
–Modern PC
–Human brain (neurons)
Three Levels in Information Processing Systems
Caveat: is coming in?
What information
What
information
is outputted?
This is
a particular
philosophical
Whatcalled
is the relationship?
position,
“Functionalism”.
What
Some philosophers do not accept it.
Interesting:
Representation ties together
Unlike other
sciences we can
study top two levels
independently from
the physical level
Functionalism: mental states
(beliefs, desires, being in pain, etc.)
are constituted solely by their
Procedure/Algorithm
functional
role; i.e. their causal
– clear
set ofmental
instructions
relations
to other
states,
sensory inputs, and behavioural
outputs.
How
Mustabe
physically
Consequence:
mind
can becarried out
–Man
withof
paper
and pen
implemented
in lots
different
–Mechanical computer
physical hardware,
so long as it
–Modern PC
performs the
right functions.
–Human brain (neurons)
Physical
Implementation
Three Levels in Information Processing Systems
What’s special about a mind then?
Caveat: is coming in?
What information
We know it can do things a computer
What
information
is outputted?
a particular
philosophical
can’t do… This is
Whatcalled
is the relationship?
position,
“Functionalism”.
Some
do not accept it.
A Functionalist claims that
the philosophers
special
Interesting:
thing about
the mind is the special
Representation ties together
Functionalism:
mental states
information processing
tasks,
Unlike
other
(beliefs,itdesires,
representations
and algorithms
uses being in pain, etc.)
are constituted solely by their
sciences we can
Procedure/Algorithm
functional
role; i.e. their causal
Onetwo
could
implement the same
functions
study top
levels
– clear
set ofmental
instructions
relations
to other
states,
in
a
computer
–
don’t
need
organic
independently from
neurons sensory inputs, and behavioural
the physical level
outputs.
What
How
Mustabe
physically
Consequence:
mind
can becarried out
–Man
withof
paper
and pen
implemented
in lots
different
–Mechanical computer
physical hardware,
so long as it
–Modern PC
performs the
right functions.
–Human brain (neurons)
Physical
Implementation
Important to Study All Three Levels
What
Could have elegant
mathematical theory
which no algorithm
can implement
How
What information is coming in?
What information is outputted?
What is the relationship?
Procedure/Algorithm
– clear set of instructions
Physical
Implementation
Must be physically carried out
–Man with paper and pen
–Mechanical computer
–Modern PC
–Human brain (neurons)
Important to Study All Three Levels
What
But without top
level…
Lose sight of what
your information
processing is trying
to achieve
How
What information is coming in?
What information is outputted?
What is the relationship?
Procedure/Algorithm
– clear set of instructions
Physical
Implementation
Must be physically carried out
–Man with paper and pen
–Mechanical computer
–Modern PC
–Human brain (neurons)
Important to Study All Three Levels
What
Could have a nice
algorithm, but
might take too
much physical
hardware to be
practical
How
What information is coming in?
What information is outputted?
What is the relationship?
Procedure/Algorithm
– clear set of instructions
Physical
Implementation
Must be physically carried out
–Man with paper and pen
–Mechanical computer
–Modern PC
–Human brain (neurons)
Important to Study All Three Levels
What
Focussing on the
physical
interactions here
gives you no idea
of what their
purpose is
How
What information is coming in?
What information is outputted?
What is the relationship?
Procedure/Algorithm
– clear set of instructions
Physical
Implementation
Must be physically carried out
–Man with paper and pen
–Mechanical computer
–Modern PC
–Human brain (neurons)
Important to Study All Three Levels
Insights from
studying the brain
could give clues
about the
algorithms and
representations
which are
(or are not)
being used
What
How
What information is coming in?
What information is outputted?
What is the relationship?
Procedure/Algorithm
– clear set of instructions
Physical
Implementation
Must be physically carried out
–Man with paper and pen
–Mechanical computer
–Modern PC
–Human brain (neurons)
Another Perspective on Cognitive Science
 Studying different information processing tasks at different levels
Vision Language Memory
What
(Info Proc Task)
How
(Algorithm)
Physical
Implementation
Problem
Solving
Learning
AI and Cognitive Science
 Two way interaction between AI and Cognitive Science
 AI informs Cognitive Science




Common to implement a cognitive theory in a computer
Run the program and see the ramifications of the theory
(Scientific hypothesis testing)
Running it may be necessary because theory is complicated
 Also, existing AI theories may shed light on the way humans do it
 Cognitive Science informs AI




Seeking inspiration to solve an AI problem
Study the way humans do it
Copy in computer
…or at least constrain the possible options under consideration
“AI can have two purposes. One is to use the power of
computers to augment human thinking,
just as we use motors to augment human or horse power.
Robotics and expert systems are major branches of that.
The other is to use a computer's artificial intelligence to
understand how humans think. In a humanoid way.
If you test your programs not merely by what they can
accomplish, but how they accomplish it, then you're really
doing cognitive science;
you're using AI to understand the human mind.”
Herbert Simon
Applications of Cognitive Science
 Education and Learning
 From Cognitive Psychology:
 Diagnose and treat children’s reading difficulties
 Stroke Therapy
 From Linguistics:
 Understanding of speech impairments when stroke in left
hemisphere of brain
 … better therapy
 Legal process
 From Cognitive Psychology:
 Understanding of reliability of memory
 Question reliability of legal witnesses
 Computing Technology
 From AI:
 You know loads of examples by now
Cognitive Science – Different Methods
 Psychology
 Controlled laboratory experiments
 Detailed observations of behaviour
 Philosophy
 Thought experiments
 Investigate consequences, and coherence of theories
 Linguistics
 Test speakers’ intuitions about “grammatical” sentences
 Analyse children’s acquisition and errors
 Neuroscience
 Study active brain regions when doing something
 Study neurons
 Artificial Intelligence / Computer Science
 Write programs, see where they succeed and fail
Cognitive Psychology
 What are the mental processes in between stimulus and response?
Central
systems
Sensory
systems
sensory
input
Motor
systems
Sight
Categorisation
Voice
Hearing
Attention
Limbs
Taste
Memory
Fingers
Smell
Knowledge representation
Head
Touch
Numerical cognition
Balance
Thinking
Heat/cold
Learning
…
Language
…
Motor
output
(rough model - Boundaries are not clear in reality)
Cognitive Science – Different Methods
Focus on central unit…
 Thinking
 Draw conclusions from facts, solve problems, plan actions…
 In many diverse domains
 Attention
 Helps us focus on some task
 Has limited capacity
 Memory (includes Knowledge Representation)
 Seems to be huge
 Seems to be no limit on how well it retrieves relevant information
 Learning
 Acquire new knowledge and sensorimotor skills
 How does this central unit work?
Physical Symbol System Hypothesis
“A physical symbol system has the necessary and
sufficient means of general intelligent action.”
Newell & Simon,
1963.
therefore…
human thinking = symbol manipulation
Physical Symbol System Hypothesis
“A physical symbol system has the necessary and
sufficient means of general intelligent action.”
Their symbols are taken to mean high level symbols
 Directly correspond to objects in the world,
 such as “monkey” and “table”.
…but the weights and connections in a neural network
could also be represented as symbols
 Use this to make a “scruffy” representation of “monkey”
 but that’s not considered to be what they meant
Physical Symbol System Hypothesis
“A physical symbol system has the necessary and
sufficient means of general intelligent action.”
 Most AI people nowadays would not accept the idea of
high level symbols being sufficient
 Seems to work well for
 playing chess, problem solving (if problem well defined)
 but doesn’t work so well for some “easy” problems
 Vision, moving around in the world
 But most AI people would accept the computational
theory of mind (i.e. Functionalism)
Universal Computing Machine
 Turing machine:
 Actions:
 Head can move left and right over the tape
 Can read and write symbols on the tape
– Can overwrite symbols on tape




 Machine has an internal state
 Takes Action depending on state
Turing’s thesis: “If an algorithm exists then there is an equivalent Turing
Machine”
Turing machine is the simplest possible description of a computer that
can do anything
All modern computers can be simulated by a Turing machine
Only real difference:
Turing machine has infinite tape, real computers have finite memory
Universal Computing Machine
 How many symbols and
states do you need?
States
Symbols
24
2
10
3
7
4
5
5
2
5
4
6
3
10
2
 Interesting…
 If you make some really fancy machine…





18
Loads of states
Loads of possible symbols
Multiple tapes
Multiple stacks for storing things
Many heads working in parallel
 You end up with something equivalent to the Turing machine
Universal Computing Machine
 The Turing machine has a set of rules
 These determine how it acts
 Can make a Universal Turing machine




Encode the rules you want it to use on the tape
The first thing it does is to read the rules
Then follow them…
Could also reprogram its rules as it goes along
 Important ability for learning
 Behaviour must change given experience
Universal Computing Machine
 We said
 “If an algorithm exists then there is an equivalent Turing Machine”
 i.e. a (different) Turing machine is available to do any job we want to do
 Now we can say
 “If an algorithm exists then it can be simulated on a Universal Turing Machine”
 i.e. all we need is a single Universal Turing Machine
 This can do anything
 This is the idea behind modern computers




Program instructions stored in memory just like any other data
Download a program off the web, and start running it
You don’t need a different computer for different jobs
One computer can do everything
 Games, spreadsheet, database, music, movies, photo editor, word processor…
Is the Brain a Universal Computing Machine?
 Warren McCullogh and Walter Pitts showed
 Small collections of neurons can act as “logic gates”
(building blocks of computers)
 Brain could be viewed as a computing device, just like Turing machine
 i.e. a brain can do what a computer can do
 Other direction is a stronger claim
 Can a computer do what a brain can do?
 Can’t be proved
 But universality of Turing machine suggests… maybe