Step back and look at the Science

Download Report

Transcript Step back and look at the Science

Course Overview
 What is AI?
 What are the Major Challenges?
 What are the Main Techniques?
Part I:
Introduce you to
what’s happening in
Artificial Intelligence
 Where are we failing, and why?

Done
 Step back and look at the Science
 Step back and look at the History of AI
 What are the Major Schools of Thought?
 What of the Future?
Part II:
Give you an
appreciation for
the big picture
 Why it is a
grand challenge
Course Overview
 What is AI?
 What are the Major Challenges?
 What are the Main Techniques?
Part I:
Introduce you to
what’s happening in
Artificial Intelligence
 Where are we failing, and why?

Done
Step back and look at the Science
 Step back and look at the History of AI
 What are the Major Schools of Thought?
 What of the Future?
Part II:
Give you an
appreciation for
the big picture
 Why it is a
grand challenge
Course Overview
 What is AI?
 What are the Major Challenges?
 What are the
Main Techniques?
Looking
at the Science
 Engineering vs. Science
 Introduction to Cognitive Science
 Cognitive
1
Step back
and lookPsychology
at the Science
 Cognitive Psychology 2
 Cognitive
Development
 Step back and
look at the
History of AI
 Linguistics
 What are the
Major Schools of Thought?
 Neuroscience
 Philosophy
 Where are we failing, and why?
 What of the Future?
Course Overview
 What is AI?
 What are the Major Challenges?
 What are the
Main Techniques?
Looking
at the Science
 Engineering vs. Science
 Introduction to Cognitive Science
 Cognitive
1
Step back
and lookPsychology
at the Science
 Cognitive Psychology 2
 Cognitive
Development
 Step back and
look at the
History of AI
 Linguistics
 What are the
Major Schools of Thought?
 Neuroscience
 Philosophy
 Where are we failing, and why?
 What of the Future?
Why Study Neuroscience?
Neuroscience: study of the nervous system
 physical structure and physical processes
1. Scientific interest – how to get a complete picture
 Thoughts  algorithms  neurons  molceules
2. Test psychology/linguistics theories
 Example: visual buffer idea, test what is active in brain
 Example: test for existence of language module
3. Find out something about cognitive architecture
 Capabilities of a proposed architecture depend on physical structure
 Knowing structure/processes should give an idea of architecture
4. Important to understand relationship:
physical structure  information processing capabilities
 Improve treatments for damage (accident or disease)
 Know human limits 


Better learning methods
Better computer systems to support humans (compensate for weakness)
Study at Different Scales
1m – CNS
10cm – systems (vision system)
1cm – maps
1mm – networks
100m – synapses
1Å – molecules (neurotransmitters)
Study at Different Scales
1m – CNS
10cm – systems (vision system)
1cm – maps
1mm – networks
100m – synapses (know a bit about this)
1Å – molecules (neurotransmitters)
Study at Different Scales
1m – CNS
10cm – systems (vision system)
1cm – maps (don’t know much about this)
1mm – networks (don’t know much about this)
100m – synapses (know a bit about this)
1Å – molecules (neurotransmitters)
Central Nervous System
Central Nervous System
Cerebrum
Cerebellum
Brain stem
Cerebrum
Cerebellum
Brain stem
Compare Brains of Other Animals
 Human similar to rat or monkey
 Early research tried to find special neural cells which
were unique to humans
 Didn’t find any
 …But human brain bigger
 Seems not special cells,
but more of them and more connections
Animal
Man
Bottlenose dolphin
Chimpanzee
Whales
Gorilla
Fox
African elephant
Dog
Squirrel
Cat
Horse
Sheep
Mouse
Rabbit
Encephalization quotient
7.4–7.8
5.3
2.2–2.5
1.8
1.5–1.8
1.6
1.3
1.2
1.1
1.0
0.9
0.8
0.5
0.4
Brain Development
 Macaque monkeys born with 60% of brain size
 Chimpanzee – 46%
 Human – 25%...
 Brain growth rate of foetus same… but…
 Human brain continues growth at rapid fetal rate for 2 yrs
 Child has higher density of connections
 Gradually eliminated
 Reach adult values by about 10yrs
 Windows for development




8 months can distinguish two foreign language sounds
12 months cannot
Strabismus can be corrected in early years
Infants born without callosum compensate (other pathways?)
Brain Development
 Plasticity of brain in children
 Language impairment produced by brain injury
can recover if before age 5
 Infants who had left half brain removed
many linguistic functions normal
Synapses
Release neurotransmitter chemicals
 Excitatory
 Inhibitory
Synapses
 Repeated stimulation can increase synaptic strengths
 For days or even weeks
 Learning?
Neurochemical Systems
 At least 40 different substances
 Serve a multitude of different functions
 2 types
 Neurotransmitter
 act in synapse
 Neuromodulator




Act more globally
… but extremely specific functions
Nanogram of angiotensin II : intense and prolonged drinking
Acts as trigger
Neurochemical Systems
 Psychoactive drugs
 Mimic/enhance/disrupt effects of body’s neurochemicals
 Molecular structures may resemble neurotransmitters
 LSD similar to seratonin
 Mescaline similar to dopamine
 Cocaine blocks reuptake of neurotransmitter
 More neurotransmitter remains in synapse to stimulate further
“To describe the global meteorological system
around the earth would be an extremely trivial
task compared with description of the complete
behaviour of the brain.”
Teuvo Kohonen
(Artificial Neural Network researcher)
Fascinating Brain Facts…
 100,000,000,000 = 1011 neurons
 100 000 are irretrievably lost each day
 Each neuron connects to 10,000 -150,000 others
 About 1014 or 1015 synapses
 Every person on planet make 200 000 phone calls
 same number of connections as in a single human brain in a day
 Grey part folded to fit - would cover surface of office desk
 The gray cells occupy only 5% of our brains
 95% is taken up by the communication network between them
 About 2x105km of wiring
 Pulses travel at more than 400 km/h (250 mph)
 2% of body weight… but consumes 20% of oxygen
 All the time! Even when sleeping
 Any loss of oxygen… cells die in minutes…
 Damage permanent in adults
That’s the brain hardware…
That’s the brain hardware…
How do we find how different parts
perform different functions?
Mapping Functions to Brain Areas
 See what disabilities result from specific physical
damage
 In humans wait for accident
 In animals do damage deliberately
 To see where axons terminate
 Inject dye, transported along axon
 Slice up brain and examine under microscope
 Insert microelectrodes
 Into a single neuron
 Monitor changes in electrical potential
 Brain imaging
 While patient is doing a particular task
Imaging
 Positron Emission Tomography (PET)




Inject radioactive glucose
Gets absorbed by active regions
Get the subject to do a task for about two minutes
Record image
 Functional magnetic resonance imaging (fMRI)
 Hemoglobin: metalloprotein red blood cells
 Carries oxygen
 Hemoglobin diamagnetic when oxygenated
paramagnetic when deoxygenated
More Brain Facts…
In right-handed individuals (91% of people)
Right side of the brain controls:
 musical talent,
 fantasy, imagination, dreams,
 drawing and painting.
Left side of the brain controls:




mathematical ability,
ability to solve logic problems,
language skills,
remembers names, dates, and facts
Mapping Functions to Brain Areas
Mapping Functions to Brain Areas
Maps change with experience
 Microelectrodes to map monkey’s hand and fingers area
 Trained monkey rotating disk for food reward
 After 20 weeks…
 Brain area for hand had expanded markedly
Innate or Learned? Nature or Nurture?
 Major neural pathways very similar in all mammals
 Suggests genetic hardwiring
 Study of eye in water flea
 Made cloned insects (genetically identical)
 Studied neurons
 Same number of sensory neurons,
and connect to same number of cells
 Different number of synapses,
and shape of axons
 Similar results for pigeons
 Final system could be quite different in cloned animals
Neuropsychology
 Tries to discover relationships between




Neuroscience and Psychology
…or between :
Cognitive models of information processing
And structures and processes in the brain
 Examples:
 Study functions of people who have suffered brain damage
 Build an artificial neural network to solve some problem
 (In a biologically plausible way)
 Then damage the network
 Study the results and compare with real patients
 Imaging techniques (as discussed before)
Neuropsychology studying Alexia
 Alexia: damage to brain causes loss of ability to read
 Ability to read and write can be affected, while speech is normal
 Theoretical model of reading: dual route
(parallel processing)
 Lexical route – retrieve words from a lexicon
 Phonological route – sound out words
Neuropsychology studying Alexia
 Phonological route







Group word into syllables (by vowels/consonants)
Convert syllables to proper phonemes
Evident in children sounding words
Some patients have difficulty pronouncing unfamiliar words
Example: non-words like “troat”
Otherwise no problem reading
“Phonological deficit hypothesis”
 Lexical route






Identify base root: antiabortion  anti-abort-ion
Access Lexical memory to recall proper pronunciation of parts
Some patients have difficulty accessing lexicon
Must pronounce out word
Difficulty with exceptions: yacht, come, have
“Dyseidetic” or “visual dyslexia”
Neuropsychology is Difficult
 Usually have a dodgy cognitive model of a process
 Must find a match between this and multiple possible brain
implementations
 Evidence from patients who have suffered damage difficult…






Damage can affect many areas
Patients often on medication with not entirely clear effects
Patients “grouped” together with “same” damage often different
Example: Trauma and tumour quite different
Imaging Scans have limited accuracy
Patient’s age very relevant
 One solution: case study with individuals
Relationship With Artificial Neural Networks
 ANN typically leave out many aspects of real networks










Real neurons generate sequences of action potentials
Frequency and phase significant
(ANN has a simple number output)
Real networks have multiple neurotransmitters
Many distinct types of neurons with different shapes
(ANN has one type)
Real networks have microcircuits
Compute complex nonlinear functions
(ANN usually a sum)
Maybe synapse should be unit rather than neuron
 Some success:
 trained a network to recognise object position relative to eye direction
 Used hidden layer in ANN
 Resulting values closely resembled measurements from macaque
monkey neurons
Course Overview
 What is AI?
 What are the Major Challenges?
 What are the
Main Techniques?
Looking
at the Science
 Engineering vs. Science
 Introduction to Cognitive Science
 Cognitive
1
Step back
and lookPsychology
at the Science
 Cognitive Psychology 2
 Cognitive
Development
 Step back and
look at the
History of AI
 Linguistics
 What are the
Major Schools of Thought?
 Neuroscience
 Philosophy
 Where are we failing, and why?
 What of the Future?