Transcript The Promise and Perils of Artificial Intelligence

Faculty of Computer Science
Technology and the Future of Medicine
Promise and Perils of AI
Part I
Osmar R. Zaïane
Professor and Scientific Director
Alberta Innovates Centre for
Machine Learning
Continuous Professional Learning Course
© 2011
Promise and Perils of AI
The future is already here —
it's just not very evenly
distributed
William Ford Gibson
(American-Canadian writer born 1948)
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
Where do I stand vis-à-vis the Singularity?
 Professor in Computing Science Specializing in Data Mining
and Machine Learning  can’t predict
 Will the Technological Singularity happen?
– hypothetical future emergence of greater-than human
intelligence through technological means
 Yes, but not in the very near future
 Is it a promise of AI? Yes (AI will play a huge role, but AI is a moving target)
 Are there Perils? Yes (Will we be ready when it happens?)
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
110
105
100
95
107
96
90
2050 2100
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
Can technology change this
trend so that we can live long
and healthy lives? Possibly.
Office for National Statistics, UK
Currently we are extending the
life expectancy but not a healthy
life.
Source: http://www.publications.parliament.uk/pa/ld200506/ldselect/ldsctech/20/2004.htm
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
We will have a population of cyborgs
(cybernetic organisms Biological and artificial being - term coined in 1960 by Manfred Clynes)
What are the concequences of a world of cyborgs?
=
+
=
biological cells
prosthesis
human
+
machine
Will we all become
cyborgs?
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
The Science Fiction View
AI - Spilberg
Data – Star Trek
R2D2 – Star wars
I Robot
Terminator
- UofA
Colossus, The Forbin Project - 1969
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
What is Artificial Intelligence?
 Tools that exhibit human intelligence and behaviour including selflearning robots, expert systems, voice recognition, natural and
automated translation. Unesco/education
 The branch of computer science dealing with the reproduction or
mimicking of human-level thought in computers; The essential quality
of a machine which thinks in a manner similar to or on the same
general level as a human being. Wikipedia
 The branch of computer science that deals with writing computer
programs that can solve problems creatively. WordNetWeb
Ability to - reason and plan,
- solve problems,
- think abstractly,
- comprehend complex ideas,
- learn quickly.
- UofA
Artificial Intelligence is the science of
making machines do things that
would require intelligence if done by
men. -- Marvin Minsky
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
Road Map
 Promise and Perils of AI
Part I (September 28)
• Artificial Intelligence and Expert Systems
 Promise and Perils of AI
Part II (September 29)
• Machine Learning and Data Mining
 Promise and Perils of AI
Part III (October 13)
• Applications: Fiction or Reality; Risks and Potential
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
Artificial Intelligence
John McCarthy was responsible for the coining of
the term "Artificial Intelligence" in his 1955 proposal
for the 1956 Dartmouth Conference
"The study is to proceed on the basis of the conjecture that every aspect of
learning or any other feature of intelligence can in principle be so precisely
described that a machine can be made to simulate it."
“The study of intelligent behaviour and the attempt
to find ways in which such behaviour could be
engineered in any type of artefact”
Putting human intelligence into a machine
Branch of computer science that deals with intelligent behaviour, learning
and adaptation in Machines. Wikipedia
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
What is Intelligence
 The capacity of observing and interpreting the
environment then adapting to it; the ability of learning
from past experiences and profiting from these
experiences to solve problems, including planning.
 Interpret; learn; adapt.  understand to solve.
What is Computational Intelligence
 Programs that analyse and interpret data to learn
from observations and adapt to changing
situations. Hence, Machine Learning (in Part II).
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
Problem
 Human intelligence is not completely understood
 We cannot comprehend an intelligence superior to our own
 Weak AI: refers to AI that only simulates human thoughts and
actions : actions and ideas are programmed into the machine
– All current forms of AI are weak AI
 Strong AI (true AI): refers to AI that matches or exceeds
human intelligence – not just simulation
 Acting Humanly: Since we do not have a good definition of
"intelligence“ and we know humans are intelligent. If an entity
can masquerade as human so well that we can not detect the
difference, we would say this entity is intelligent.
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
Acting humanly: Turing Test
 Turing (1950) "Computing machinery and intelligence":
 "Can machines think?"  "Can machines behave intelligently?"
 Operational test for intelligent behavior: the Imitation Game
 Predicted that by 2000, a machine might have a 30% chance of fooling a lay
person for 5 minutes
 Anticipated all major arguments against AI in following 50 years
 Suggested major components of AI: knowledge, reasoning, language
understanding, learning
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
Was Turing Right?
A conversation with Jabberwacky, chatterbot
3rd place Loebner Prize 2003. http://www.jabberwacky.com
Jabberwacky: What is your name?
User: Fred?
Jabberwacky: No it isn’t
User: Do you mean I am lying?
Jabberwacky: Yes, I’m a very good liar. One of the best, in fact.
User: That doesn’t mean I am a liar if you are a good one
Jabberwacky: Yes it does.
User: Why is that?
Jabberwacky: I am god.
User: Do you have a temple for you?
Jabberwacky: No, that’s alright.
User: So you are god with a small g?
Jabberwacky: I am small g
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
Was Turing Right?
A conversation with ALICE, another chatterbot
http://alice.pandorabots.com/ .
ALICE: I am A.L.I.C.E
User: Where are you?
ALICE: I am in Oakland, California. Where are you?
User: in Canada
ALICE: Tell me a little about your life in canada.
User: it is cold
ALICE: I am warm here.
User: where is the best Chinese restaurant? ?
ALICE: I think it is a tie.
User: why is it a tie?
ALICE: Why do I exist?
User: to test artificial intelligence?
ALICE: What is the sound of one hand clapping?
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
Achievements
 Chess: Deep Blue – Gary Kasparov
– 1997: Deep Blue won a 6-game match
(2 wins to 1 + 3 draws)
 Checkers: Chinook – Dr. Marion. Tinsley
– 1994: won World Man-Machine Championship
– 2007: Checker solved (5X1020 positions)
http://www.cs.ualberta.ca/~chinook/
 Quiz Show Jeopardy: Watson – Ken
Jennings & Brad Rutter
– 2011: Wins Q&A Jeopardy by wide margin
 No-Hands across America (driverless vehicles)
– 1995: 3000 Miles Pittsburg to San Diego
– 2004-2007 DARPA Grand and Urban Challenge
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
Abridged modern history of AI
 1950
Turing's "Computing Machinery and Intelligence"
 1956
Dartmouth meeting: "Artificial Intelligence" adopted and field founded
 1957-1974 AI research heavily funded world wide. Funders optimistic about future
 1974
following criticism from researchers and politicians and pressure from
US congress to fund other productive projects funding was cut off (1st AI
winter)
 1970s
Development of Expert Systems
 1980s
AI revived by commercial success of Expert Systems
 1980s
Japan 5th generation computer project inspired research in US and
Europe  new funding
 1987
Collapse of the Lisp machine Market. AI back in disrepute (2nd AI winter)
 1990s
New success for AI thanks to (1) emphasis on solving specific problems;
Specific subproblems
(2) increase of computational power (Moore’s
Law)
1st AI winter
50
60
70
80
- UofA
2nd AI winter
90
00
10
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
Moore’s Law
The number of transistors that can be placed
inexpensively on an integrated circuit doubles
approximately every two years.
This is also verified with disk capacity;
digital camera pixels per price, etc.
with other hardware
An Osborne Executive portable computer, from 1982, and an
iPhone, released 2007. The Executive weighs 100 times as
much, is nearly 500 times as large by volume, costs 10 times
as much, and has 1/100th the clock frequency of the iPhone.
Source: Wikipedia
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
Why is Computational Intelligence Important?
 AI has become important in a number of fields in
helping to make better use of information,
increasing the efficiency and effectiveness of
applications, and enhancing productivity,
particularly when adaptability is relevant
 Research in AI is also important in understanding
and appreciating the complexity of human
intelligence and the human body itself.
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
From General Intelligence to Specific Sub-Problems
• Knowledge representation
• Reasoning and problem solving
• Planning
• Natural language processing
• Perception
• Learning
• Motion and manipulation
• Emotion
• Creativity
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
Knowledge Representation
• Most AI tasks require significant knowledge: context
knowledge, application knowledge, common sense knowledge
and general knowledge
• Knowledge representation is capital to AI
• How to model knowledge; how to represented concisely; how
to interpret knowledge; and how to provide efficient access
and retrieval when needed.
• Rule-based, graph-based, logic-based, ontologies, semantic
networks, frame representations, concept maps, etc.
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
Reasoning and Problem Solving
•
Step by step reasoning to solve a problem such as solving a puzzle or making
logical deduction.
•
Combinatorial problems with large search spaces.  Heuristics for pruning
Planning and scheduling
• Intelligent agents, in a given context and new environment need to
choose actions to make in order to reach a goal.
• Action choice is made based on a utility to maximize (maximizing a
reward or minimizing a cost)
• Target a global optimum without falling in a local optimum
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
Natural Language Processing
• Ability to interpret and understand human languages
• Written language and spoken language
• Ability to generate sentences and express knowledge in human language
• Ability to acquire knowledge from natural language (written or spoken)
• Ability to summarize, paraphrase and translate natural languages
• Ability to make jokes, pans and interpret idiomatic expressions
• Ability to “read between the line”
Perception and Pattern Recognition
• Ability to use inputs from sensors such as cameras, microphones, etc., to
deduce aspects of the world
• Computer vision, speech/voice recognition, face recognition, object
recognition, etc.
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
Learning
• Learning is central to AI. For an AI program to adapt to its
environment, it has to learn
• Machine Learning provides means to learn from large data,
interpret the trends in the data and adapt to the data as
opposed to static programs
• There is supervised learning, unsupervised learning, active
learning, reinforcement learning inductive learning, etc.
• Machine learning will be covered in Part II
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
Motion and Manipulation
• Robotics and AI are cousins.
• There is some intelligence required to recognize and manipulate objects;
• There is intelligence required to move in a new environment after
identifying its own location a target place and planning the movement
Emotion
• Intelligent agents interacting with other agents or humans need social
skills (interpreting emotions and exhibiting emotions)
• Modeling human emotions to better interact with humans
• Game theory
Creativity
• AI addresses creativity theoretically and philosophically
• Artificial imagination
• Creations that generate feelings and emotions
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
Expert Systems
•
Expertise is required in many locations but experts are rare
•
Expert may retire and expert knowledge is lost
•
Can we preserve and duplicate this expert knowledge?
•
Conventional computer programs follow the exact procedure a developer
programmed in them. Expert systems don’t.
•
An expert system is a computer system that emulates the decision-making ability of
a human expert by reasoning about knowledge to solve complex problems given
some contextual facts.
•
There are two types of knowledge: expert knowledge representing the expertise and
typically coded in rules, called knowledge base or rule base; and contextual
knowledge representing the facts of the current case to solve.
•
IF condition THEN conclusion
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
General Architecture
Interview
Knowledge base
facts
Domain expert
IF the identity of the germ is not known with certainty
AND the germ is gram-positive AND the morphology of
the organism is "rod" AND the germ is aerobic THEN
there is a strong probability (0.8) that the germ is of type
enterobacteriacae
Inference engine
Problem
Expert System
The inference engine is a computer program based
on logic that is designed to produce a reasoning on
rules and facts to deduce more facts.
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
The Rise of Expert Systems
1967
Dendral – a rule-based system that infered molecular
structure from mass spectral and NMR data
1975
Mycin – a rule-based system to recommend antibiotic
therapy
1975
Meta-Dendral learned new rules of mass spectrometry,
the first discoveries by a computer to appear in a
refereed scientific journal
1979
EMycin – the first expert system shell
1980’s
The Age of Expert Systems coinciding with the
Japanese Fifth Generation project
1985
Revenue peaks at $1 billion before the 2nd AI winter
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
Expert Systems – Today: Medicine
One example domain, medicine, has expert systems whose
tasks include:
•arrhythmia recognition from electrocardiograms
•coronary heart disease risk group detection
•monitoring the prescription of restricted use antibiotics
•early melanoma diagnosis
•gene expression data analysis of human lymphoma
•breast cancer diagnosis
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
Major problem with Expert Systems
 knowledge engineering, knowledge collection and
interpretation into rules, is very difficult and tedious
 We do not know what we know
 Identifying contradictory rules
 Missing rules
 Inconsistencies between experts
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011
Promise and Perils of AI
AI made and is making big strides. There are promises and there are
perils. We do not know what to expect around the corner.
- UofA
Edmonton – September 2011
O.R. Zaïane © 2011