WHY I AM OPTIMISTIC

1
WHY I AM OPTIMISTIC

• Patrick H. Winston
• MIT Artificial Intelligence Laboratory

2
The salients

• Applications sideWe have already won
• Science sideWe are bound to win

3
The Applications
4
We have expanded our frontiers
Lots of good
Lots of people
5
A little good for a lot of people
6
A little good for a lot of people
7
A lot of good for a few people
8
A lot of good for a lot of people
9
We have exemplars of all kinds

• Large software companies
• Large entertainment companies
• Companies with huge IPOs
• Multidimensional multinationals
• A multitude of small companies

10
We were a one-horse field
Rule chaining
Inheritance
11
Now we ride many horses
Neural nets
Rule chaining
Inheritance
Generate and test
Constraint propagation
Search
Genetic algorithms
Tree building
Bayes nets
Learning
Agents
12
And not just reasoning horses

• Vision
• Language and speech
• Infrastructure

13
(No Transcript)
14
(No Transcript)
15
We had a Pyrrhic victory
Network
Tapes
IO
Power
Cables
Disk
Memory
16
We learned negative lessons

• Nobody cares about saving money
• Using cutting edge technology
• To replace expensive experts

17
We learned positive lessons

• Everybody cares about
• New revenues
• Saving a mountain of money
• Increasing competitiveness

18
We changed the business model
Replaces Expensive People
Saves Mountains Of Money
Creates New Revenue
Ferrets
Blunder stoppers
Novices
Experts
19
The critic and the billionaire
20
Whats next connections
People
Enhanced Reality
Human Computer Interaction
Information Access
Intelligent StructuresUseful robots
Computers
Physical World
Global Net
21
The click-in phenomenon

• The fax machine
• The world wide web

22
The Science
23
Shrobes point

• Applications drive science
• Unless they all look alike

24
Atkesons point

• We could move to the center
• But, we might be kidding ourselves

25
My point

• AI is applied computer science
• Much energy wonderfully used
• But consequently diverted

26
A 100 year enterprise
Molecular Biology
1950
2000
2050
1900
Artificial Intelligence
27
Why we are the way we are
Powerful Ideas
Models of Thinking
ReflectionBiologyPsychology
28
The standard paradigm
The IntelligentReasoner
Input/Output Channels
Language
Vision
29
The dawn age
30
What went wrong?

• We think with our eyes
• We think with our mouths
• We think with our hands
• Each faculty helps the others

31
What is the evidence?

• Armchair psychology
• Clues from the brain

32
Armchair psychology

• Hilliss observation on the value of talking to
  yourself
• Everyones observation on the value of drawing a
  sketch.

33
From brain scanning
34
Intelligence is in the I/O
The Explanation
MotorReasoner
LinguisticReasoner
VisualReasoner
35
Language is first among equals

• Language reasons
• Language tells vision how to see
• Language tells motor how to act

36
Is it time to start over?

• An I/O oriented paradigm
• Essentially free computation
• Important, inspiring allies

37
From brain rewiring
38
From watching infants
39
Is it time to start over?

• An I/O oriented paradigm
• Essentially free computation
• Important, inspiring allies
• Accumulation of powerful ideas

40
Six powerful ideas

• Recreated condition
• One-shot learning
• Memory is cheap
• Change matters
• Survival of the smallest
• Bi-directional search

41
Recreated condition Minsky
----------------------------------
K-line --------------gt ---------gt
P
/ \
P P /
\ P
P /
\ / \
P P P
P / \ / \ /
\ / \ P P P
P P P P P / \
/ \ / \ / \ / \ / \ /
\ / \ P P P P P P P P P
P P P P P P P
42
One-shot Yip and Sussman
Rule Memory
Word Memory
ae
p
l
Time
43
One-shot Yip and Sussman
Rule Memory
Word Memory
ae
p
l
z
Time
44
One-shot Yip and Sussman
100
Accuracy
Trials 500
45
Memory is cheap Atkeson
46
Atkesons practice tables
47
Atkesons practice results
Feedback only
One stored trajectory
Three stored trajectories
48
Change matters Borchardt
49
Borchardts ladder diagrams
D
A
Distance
D
A
Speed
A
A
A
Contact
50
Survival of the smallest Kirby
51
Kirbys phase transitions
Coverage
Time
52
Bi-directional search Ullman
Model
Image
53
Joyous inferences

• Powerful ideas
• Marvelous engineering
• Essential alliances

54
What about

• Bayes and Markov
• Neural nets and connectionism
• Logic

55
WHAT WE MUST NOT DO
56
Loose our faith

• It will take 300 years
• All the low hanging fruit is gone
• We shouldnt make predictions

57
Waste time arguing

• Is it possible?
• Is it successful?
• Is it really AI?

58
Squander our capital

• One thousand people
• 10 interested in the science side
• 10 actually working on it
• 10 of the time

59
WHAT WESHOULD DO
60
Human Intelligence Enterprise

• Vision, language, motor
• Free hardware
• Clues from the brain
• Powerful ideas
• Conceive and test models

61
Why we should do it

• It can only be done once
• Revolutionary applications

62
What we should ask

• Why do we have discrete words?
• What do our inner agents say?
• How do they learn what to say?
• Do we see what chimps see?
• How did our faculties evolve?
• Why cant we all play the piano?

63
So, here is why Im optimistic

• Nothing could possibly be morefun, exciting,
  rewarding, and glorious, than
• Applications that really matter
• Figuring out our own intelligence