Artificial Intelligence In the Real World

1
Artificial IntelligenceIn the Real World
Computing Science University of Aberdeen
2
Artificial IntelligenceIn the Real World
Artificial IntelligenceIn the Movies
3
Artificial IntelligenceIn the Real World
Artificial IntelligenceIn the Movies
4
Artificial IntelligenceIn the Real World
Artificial IntelligenceIn the Movies
?
5
Artificial Intelligence Began in 1956

• Great expectations

Machines will be capable, within twenty years,
of doing any work that a man can do. Herbert
Simon, 1965.
6
Machines will be capable, within twenty years,
of doing any work that a man can do. Herbert
Simon, 1965.
What Happened?
7
Machines will be capable, within twenty years,
of doing any work that a man can do. Herbert
Simon, 1965.
What Happened?

• Machines cant do everything a man can do
• People thought machines could replace
  humansinstead they are usually supporting humans

8
Machines will be capable, within twenty years,
of doing any work that a man can do. Herbert
Simon, 1965.
What Happened?

• Machines cant do everything a man can do
• People thought machines could replace
  humansinstead they are usually supporting
  humans
• Healthcare, Science, Government, Business,
  Military

9
Machines will be capable, within twenty years,
of doing any work that a man can do. Herbert
Simon, 1965.
What Happened?

• Machines cant do everything a man can do
• People thought machines could replace
  humansinstead they are usually supporting
  humans
• Healthcare, Science, Government, Business,
  Military
• Most difficult problems are solved by
  humanmachine
• astronomy, nuclear physics, genetics, maths, drug
  discovery

10
Neural Networks

• Neural Networks are a popular Artificial
  Intelligence technique
• Used in many applications which help humans
• The idea comes from trying to copy the human
  brain

11
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
• 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 2x106km of wiring (to the moon and back
  twice)
• 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
• What about copying neurons in Computers?

12

• Artificial Neural Network (ANN)
• loosely based on biological neuron
• Each unit is simple, but many connected in a
  complex network
• If enough inputs are received
• Neuron gets excited
• Passes on a signal, or fires
• ANN different to biological
• ANN outputs a single value
• Biological neuron sends out a complex series of
  spikes
• Biological neurons not fully understood

Image from Purves et al., Life The Science of
Biology, 4th Edition, by Sinauer Associates and
WH Freeman
13

• Now play with the flash animation to see how
  synapses work

http//www.mind.ilstu.edu/curriculum/neurons_intro
/flash_summary.php?modGUI232compGUI1828itemGUI
3160
14
The Perceptron
input1
weight1
input2
weight2
add
output
weight3
(threshold)
weight4
input3
input4
15
The Perceptron
input1
weight1
input2
weight2
add
output
weight3
(threshold)
weight4
input3
Save Graph and Data
input4
16
The Perceptron
Save Graph and Data
Note example from Alison Cawsey
17
The Perceptron
First last year _
0.2
add
_ output
Male _
0.2
Threshold 0.5
0.2
0.2
_ hardworking
_ Lives in halls
Note example from Alison Cawsey
18
The Perceptron
First last year _
0.15
add
_ output
0.15
Male _
Threshold 0.5
0.2
0.15
_ hardworking
_ Lives in halls
Note example from Alison Cawsey
19
The Perceptron
First last year _
0.15
add
_ output
0.15
Male _
Threshold 0.5
0.2
0.15
_ hardworking
_ Lives in halls
Note example from Alison Cawsey
20
The Perceptron
First last year _
0.2
add
_ output
Male _
0.2
0.25
Threshold 0.5
0.15
_ hardworking
_ Lives in halls
Note example from Alison Cawsey
21
The Perceptron
First last year _
0.2
add
_ output
Male _
0.2
0.25
Threshold 0.5
0.15
_ hardworking
_ Lives in halls
Note example from Alison Cawsey
22
The Perceptron
First last year _
0.2
add
_ output
Male _
0.2
0.25
Threshold 0.5
0.15
_ hardworking
_ Lives in halls
Note example from Alison Cawsey
23
The Perceptron
First last year _
0.2
add
_ output
Male _
0.2
0.25
Threshold 0.5
0.15
_ hardworking
_ Lives in halls
Note example from Alison Cawsey
24
The Perceptron
First last year _
0.2
add
_ output
Male _
0.2
0.25
Threshold 0.5
0.15
_ hardworking
_ Lives in halls
Note example from Alison Cawsey
25
The Perceptron
First last year _
0.2
add
_ output
0.15
Male _
0.20
Threshold 0.5
0.10
_ hardworking
_ Lives in halls
Note example from Alison Cawsey
26
The Perceptron
First last year _
0.2
add
_ output
0.15
Male _
0.20
Threshold 0.5
0.10
_ hardworking
_ Lives in halls
Note example from Alison Cawsey
27
The Perceptron
First last year _
0.2
add
_ output
0.15
Male _
0.20
Threshold 0.5
0.10
_ hardworking
_ Lives in halls
Note example from Alison Cawsey
28
The Perceptron
First last year _
0.2
add
_ output
0.15
Male _
0.20
Threshold 0.5
0.10
_ hardworking
_ Lives in halls
Note example from Alison Cawsey
29
The Perceptron
First last year _
0.2
add
_ output
0.15
Male _
0.20
Threshold 0.5
0.10
_ hardworking
_ Lives in halls
Note example from Alison Cawsey
30
The Perceptron
First last year _
0.2
add
_ output
0.15
Male _
0.20
Threshold 0.5
0.10
_ hardworking
_ Lives in halls
Note example from Alison Cawsey
31
The Perceptron
First last year _
0.25
add
_ output
0.15
Male _
0.25
Threshold 0.5
0.15
_ hardworking
_ Lives in halls
Note example from Alison Cawsey
32
The Perceptron
First last year _
0.25
add
_ output
0.15
Male _
0.25
Threshold 0.5
0.15
_ hardworking
_ Lives in halls
Note example from Alison Cawsey
33
The Perceptron
First last year _
0.25
add
_ output
0.10
Male _
0.20
Threshold 0.5
0.10
_ hardworking
_ Lives in halls
Note example from Alison Cawsey
34
The Perceptron
First last year _
0.25
add
_ output
0.10
Male _
0.20
Threshold 0.5
0.10
_ hardworking
_ Lives in halls

• Finished

Note example from Alison Cawsey
35
The Perceptron
First last year _
0.25
add
_ output
0.10
Male _
0.20
Threshold 0.5
0.10
_ hardworking
_ Lives in halls

• Finished
• Ready to try unseen examples

Note example from Alison Cawsey
36
The Perceptron
First last year _
0.25
add
_ output
0.10
Male _
0.20
Threshold 0.5
0.10
_ hardworking
_ Lives in halls
Note example from Alison Cawsey
37
The Perceptron
0.25
_ output
0.10
add
0.20
Threshold 0.5
0.10

• Simple perceptron works ok for this example but
  sometimes will never find weights that fit
  everything
• In our example
• Important Getting a first last year, Being
  hardworking
• Not so important Male, Living in halls
• Suppose there was an exclusive or -
• Important (male) OR (live in halls), but not
  both
• Cant capture this relationship

38
Stock Exchange Example
39
Multilayer Networks

• We saw perceptron cant capture relationships
  among inputs
• Multilayer networks can capture complicated
  relationships

40
Stock Exchange Example
Hidden Layer
41
Neural Net example ALVINN

• Autonomous vehicle controlled by Artificial
  Neural Network
• Drives up to 70mph on public highways

Note most images are from the online slides for
Tom Mitchells book Machine Learning
42
Neural Net example ALVINN

• Autonomous vehicle controlled by Artificial
  Neural Network
• Drives up to 70mph on public highways

• Note most images are from the online slides for
  Tom Mitchells book Machine Learning

43
ALVINN
Sharp right
Straight ahead
Sharp left
30 output units
4 hidden units
1 input pixel
Input is 30x32 pixels 960 values
44
ALVINN
Sharp right
Straight ahead
Sharp left
30 output units
4 hidden units
Learning means adjusting weight values
1 input pixel
Input is 30x32 pixels 960 values
45
ALVINN
Sharp right
Straight ahead
Sharp left
30 output units
4 hidden units
1 input pixel
Input is 30x32 pixels 960 values
46
ALVINN
47
ALVINN

• This shows one hidden node
• Input is 30x32 array of pixel values
• 960 values
• Note no special visual processing
• Size/colour corresponds to weight on link

48
ALVINN

• Output is array of 30 values
• This corresponds to steering instructions
• E.g. hard left, hard right

• This shows one hidden node
• Input is 30x32 array of pixel values
• 960 values
• Note no special visual processing
• Size/colour corresponds to weight on link

49

• Lets try a more complicated example with the
  program
• In this example well get the program to help us
  to build the neural network

50
Neural Network Applications

• Particularly good for pattern recognition

51
Neural Network Applications

• Particularly good for pattern recognition
• Sound recognition voice, or medical

52
Neural Network Applications

• Particularly good for pattern recognition
• Sound recognition voice, or medical
• Character recognition (typed or handwritten)

53
Neural Network Applications

• Particularly good for pattern recognition
• Sound recognition voice, or medical
• Character recognition (typed or handwritten)
• Image recognition (e.g. human faces)

54
Neural Network Applications

• Particularly good for pattern recognition
• Sound recognition voice, or medical
• Character recognition (typed or handwritten)
• Image recognition (e.g. human faces)
• Robot control - hand-arm-block.mpg

55
Neural Network Applications

• Particularly good for pattern recognition
• Sound recognition voice, or medical
• Character recognition (typed or handwritten)
• Image recognition (e.g. human faces)
• Robot control
• ECG pattern had a heart attack?

56
Neural Network Applications

• Particularly good for pattern recognition
• Sound recognition voice, or medical
• Character recognition (typed or handwritten)
• Image recognition (e.g. human faces)
• Robot control
• ECG pattern had a heart attack?
• Application for credit card or mortgage

57
Neural Network Applications

• Particularly good for pattern recognition
• Sound recognition voice, or medical
• Character recognition (typed or handwritten)
• Image recognition (e.g. human faces)
• Robot control
• ECG pattern had a heart attack?
• Application for credit card or mortgage
• Data Mining on Customers

58
Neural Network Applications

• Particularly good for pattern recognition
• Sound recognition voice, or medical
• Character recognition (typed or handwritten)
• Image recognition (e.g. human faces)
• Robot control
• ECG pattern had a heart attack?
• Application for credit card or mortgage
• Other types of Data Mining - Science

59
Neural Network Applications

• Particularly good for pattern recognition
• Sound recognition voice, or medical
• Character recognition (typed or handwritten)
• Image recognition (e.g. human faces)
• Robot control
• ECG pattern had a heart attack?
• Application for credit card or mortgage
• Spam filtering

60
Neural Network Applications

• Particularly good for pattern recognition
• Sound recognition voice, or medical
• Character recognition (typed or handwritten)
• Image recognition (e.g. human faces)
• Robot control
• ECG pattern had a heart attack?
• Application for credit card or mortgage
• Shape in go

61
Neural Network Applications

• Particularly good for pattern recognition
• Sound recognition voice, or medical
• Character recognition (typed or handwritten)
• Image recognition (e.g. human faces)
• Robot control
• ECG pattern had a heart attack?
• Application for credit card or mortgage
• Data Mining on Customers
• Other types of Data Mining
• Spam filtering
• Shape in Go and many more!

62
What are Neural Networks Good For?

• When training data is noisy, or inaccurate
• E.g. camera or microphone inputs
• Very fast performance once network is trained
• Can accept input numbers from sensors directly
• Human doesnt need to interpret them first

63
Disadvantages?

• Need a lot of data training examples
• Training time could be very long
• This is the big problem for large networks
• Network is like a black box
• A human cant look inside and understand what has
  been learnt
• Logical rules would be easier to understand