The document discusses context and knowledge, comparing the human brain and computer intelligence. It examines how humans and computers acquire and store knowledge differently, with humans relying more on abstract concepts and emotional associations while computers excel at processing large amounts of structured data. Several examples of intelligent computer systems are provided, such as Deep Blue, MYCIN, and translation programs, alongside their limitations in areas like natural language understanding.
2. Summary
Context
Knowledge
Human brain
Human brain vs. computer
Can computers be considered intelligent?
Positive examples
DeepBlue
MYCIN
Negative examples
Expressing knowledge through language
3. Context
Definition
Several definitions
– Discourse that surrounds a language unit and
helps to determine its interpretation
– The set of facts or circumstances that
surround a situation or an event
4. Context
Some context related properties
Contexts increase inferential power
Learning (new information) occurs in specific
context
Knowledge can be generalised from specific
contexts to more general ones
Contexts themselves can be objects of inference
Different contexts can be selected depending on
previous contexts
Whether something acts as a context or not could
itself be context dependent
5. Knowledge
Definition
The act or state of knowing; clear
perception of fact, truth or duty; cognition
The psychological result of perception of
learning and reasoning
Knowledge is information that has been
pared, shaped, interpreted, selected and
transformed (Ray Kurzweil)
! Facts alone do not constitute knowledge
6. Knowledge
Human vs. Computer
Human intelligence
– Remarkable ability of creating links between
ideas
– Weak at storing information on which
knowledge is based
The natural strengths of computers are roughly
the opposite powerful allies of the human
intellect
7. Human Knowledge
Abstract concepts
When we come in contact with a new concept we add new
links
Knowledge structures are not affected by the failure of the
hardware (50000 neurons die each day in an adult brain,
but our concepts and ideas do not necessary deteriorate)
We are capable of storing apparently contradictory ideas
Unless a new idea is reinforced it will eventually die out
Strong links between our emotions and our knowledge
Our knowledge is closely tied to our pattern-recognition
capabilities
We are able to change our minds change our internal
networks of knowledge
8. Computer Knowledge
Propaedia
A section of the 15th edition of Encyclopaedia
Britannica (1980)
An ambitious attempt to organize all human
knowledge in a single hierarchy
Allows multiple classifications
Takes time to understand but it is successful in view of
the vast scope of the material it covers
Such data structures provide a formal
methodology for representing a broad class of
knowledge easily stored and manipulated by
the computer
9. Human brain and
knowledge
Human brain
Highly parallel early vision circuits
Visual cortex neuron clusters
Auditory cortex circuits
The hippocampus
The amygdala
10. Human Brain
Human brain on the order of 100 billion
neurons
One neuron thousands of synaptic connections
There is a speculation that certain long-term
memories are chemically coded in neuron cell
bodies
The capacity of each neuron 1000 bits the
brain has the capacity of 1014
bits
If we assume an average redundancy factor of
104
, that gives us 1010
bits per concept 10 6
concepts per human brain
11. Human Brain
It has been estimated that a “master” of
a particular domain of knowledge has
mastered about 50000 concepts, which
is about 5 percent of the total capacity,
according to the above estimate
12. Human Brain vs.
Computer
The human brain uses a type of circuitry
that is very slow
For tasks as vision, language or motor
control, the brain is more powerful than
1000 super computers
For certain tasks simple tasks such as
multiplying digital numbers it is less
powerful that the 4-bit microprocessor
found in a ten dollar calculator
13. Computer Learning vs.
Biological Learning
The brain is wired to learn in interaction with the world,
re-programming themselves over time
Computers don’t learn easy by experience
A human child
– Starts out listening to and understanding spoken
language
– Learns to speak
– Learns written language
Computer
– Starts with the ability to generate written languge
– Learning to understand it
– Speak with synthetic voices
– Understand continuous human speech (recently)
14. Deep Blue
Its predecessor Deep Thought appeared at
Carnegie Mellon University. In 1989 it was
beaten by Kasparov in 41 moves
Project continued at IBM’s T.J. Watson
Research centre
Improvements every year: now it has 30 Power
Two Super Chip Processors
Is capable of 200 million positions / second
(Kasparov of 3 positions / second)
Almost no use of psychology
15. Deep Blue
Its strenghts are the strenghts of a machine: it
has a database of opening games played by
grandmasters over the last 100 years
It does not think, it reacts
Only one specific job
It considers before deciding on a move 4
parameters: material, position (control of the
centre), King safety and tempo (losing tempo=
wasting time by indecision, and the opponent
making productive moves)
16. MYCIN
Created in mid 1970’s by E.H. Shortliffe at
Standford University
Medical diagnosis tool (attempts to identify the
cause of infection)
Suggests a course of medication
It uses 500 rules
Each rule has assigned a number its users
can assess the validity of it’s conclusion
(WHY)
Can recognise approximate 100 causes of
bacterial infection
18. MYCIN
Fragment of a dialog between Mycin and a doctor
>> What is the patient’s name?
John Doe
>>Male or female?
Male
>>Age?
52
>>Let’s call the most recent positive culture C1
From what site was C1 taken?
……
>>My recommendation is as follows: give gentamycin using a
dose of 119 mg…
19. Other intelligent programs
in medicine:
PUFF: a system for interpreting pulmonary
tests
ONCOCIN: a system for the design of
oncology chemotherapy protocols
CADUCEUS (former Internist): a system for
diagnosis within a broad domain of internal
medicine; it contains over 100,000
associations between symptoms (70% of the
relevant knowledge in the field)
20. Other domains
Teknowledge is creating a system for General
Motors that will assist garage mechanics
ISA (Intelligent Scheduling Assistant):
schedules manufacturing and shop floor activity
DENDRAL: embodied extensive knowledge of
molecular structure analysis ( Meta-DENDRAL)
SCI (Strategic Computing Initiative): several
prototypes, among which is Vision System (will
provide real-time analysis of imaging data from
intelligent weapons and reconnaissance
aircraft))
21. Expressing Knowledge
through Language
Language is the principal means by which
we share knowledge
Language in both its auditory and written
forms is hierarchical with multiple levels
To respond intelligently to human speech,
one need to know, among other things:
– The structure of the speech sounds
– The way speech is produced
– The patterns of sound
– The rules of word usage
22. Expressing Knowledge
through Language
Computers sentence-parsing systems
can do good jobs at analysing sentences
that confuses humans:
“This is the cheese that the rat that the cat
that the dog chased bit ate”
23. Expressing Knowledge
through Language
But with other types of sentences it
has difficulties:
“Time flies like an arrow”
or
“Squad Helps Dog Bite Victim”
The difficulties appear when a word has
several meanings or are used idiomatic
expressions
24. Expressing Knowledge
through Language
Explanation to the first sentence:
For the computer this sentence it might mean:
The time passes as quickly as an arrow passes,
Or maybe it is a command telling us to time flies
the same way that an arrow flies - Time flies like an
arrow would
Or it could be a command telling us to time only
those flies that are similar to arrows - Time flies that
are like an arrow
Or perhaps it means that the type of flies known
as time flies have a fondness for arrows - Time flies
like (that is cherish) an arrow.
25. Expressing Knowledge
through Language
The ambiguity of language is far grater
than may appear.
At MIT Speech Lab, a researcher found
a sentence published in a technical
journal with over 1,000,000 syntactically
correct interpretations!!!!!!!!
26. Expressing Knowledge
through Language
TRANSLATION:
one of the challenges in developing
computerized translation system
Each pair of languages represents a
different translation problem
Best solution known was given by a
Dutch firm named DLT
27. Expressing Knowledge
through Language
Solution found by DLT:
– Developed translators for six languages to and
from a standard root language (ESPERANTO)
– A translation from English to German would be
accomplished in 2 steps: from English to
Esperanto and from Esperanto to German
– Esperanto was selected because it is
particularly good at representing concepts in an
unambiguous way
– Translating among 6 different languages would
ordinarily require 30 different translators, but
with the DLT approach only 12 are required
28. R2D2
Robot in Star Wars
Designed to operate in deep space, interfacing
with fighter craft and computer systems to
augment the capabilities of ships and their pilots
Monitors flight performance, well-versed in star
ship repair, a.s.o.
Converses in a dense electronic language
(beeps, chirps, whistles)
Can understand most forms of human speech,
but must have his own communication
interpreted by other computers