Download Artificial Intelligence What is an expert system?

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Embodied cognitive science wikipedia , lookup

Computer Go wikipedia , lookup

Artificial intelligence in video games wikipedia , lookup

Technological singularity wikipedia , lookup

Knowledge representation and reasoning wikipedia , lookup

AI winter wikipedia , lookup

Philosophy of artificial intelligence wikipedia , lookup

Ethics of artificial intelligence wikipedia , lookup

Intelligence explosion wikipedia , lookup

History of artificial intelligence wikipedia , lookup

Existential risk from artificial general intelligence wikipedia , lookup

Transcript
What is an expert system?
●
A computer system whose performance is guided
by specific, expert knowledge in solving
problems.
●
A computer system that simulates the decisionmaking process of a human expert in a specific
domain.
●
One of the early (large-scale) successes of
artificial intelligence.
Artificial Intelligence
Expert Systems
CS 4633/6633 Artificial Intelligence
CS 4633/6633 Artificial Intelligence
Early examples
●
Early examples cont.
MACSYMA - a large, interactive mathematics
expert system developed in 1968 by Engleman,
Martin, and Moses at MIT to manipulate
mathematical expressions symbolically. One of the
first expert systems. Written in LISP. Evolved into a
widely used commercial product.
HEARSAY I and II - speech understanding
systems developed at CMU that accept a speech
wave as input and produces a list of hypotheses. In
1975, it reached 75% accuracy in interpreting a
1,000-word vocabulary.
CS 4633/6633 Artificial Intelligence
●
MYCIN - the best-known expert system in use today.
Developed in the mid 1970’s by Ed Feigenbaum at
Stanford to diagnose infectious blood diseases and
prescribe an antibiotic treatment. Reached human level
precision.
●
●
DENDRAL - one of the first expert systems was
developed by Ed Feigenbaum at Stanford University.
Heuristic Dendral establishes the structure of a molecule
given: (a) its atomic formula (e.g. H2O), and (b) its mass
spectrogram (the result of bombarding a sample with electrons,
CS 4633/6633 Artificial Intelligence
Modern examples
●
●
●
PROSPECTOR - a commercial expert system
●
developed in the late 1970’s at SRI by Duda, Hart, and
Barnett. The LISP-based system locates valuable ore
deposits and produces maps of geological site
evaluations.
●
CS 4633/6633 Artificial Intelligence
INTERNIST - an internal medicine expert system that
is now called CADUCEUS. Developed at the Univ of
Pittsburgh in the early 1970’s to analyze hundreds of
clinical problems. Written in LISP.
breaking its molecules into pieces, putting them through a magnetic field
and measuring their mass.)
Early examples cont.
●
●
●
control (air traffic)
debugging
design (computer
configuration)
medical diagnosis
instruction/training
interpretation (speech)
●
●
●
●
●
monitoring (nuclear
plant)
planning (mission
planning)
Factory scheduling
prediction (weather)
repair (telephone)
CS 4633/6633 Artificial Intelligence
Advantages and disadvantages
Human Expert
Expert System
Pro Expert perishable
System
unpredictable
slow reproduction
expensive
slow processing
permanent
consistent
quick replication
affordable
fast processing
Pro Human creative
adaptive
broad focus
common sense
CS 4633/6633 Artificial Intelligence
lacks inspiration
needs instruction
narrow focus
machine knowledge
Classes of expert systems
●
Type of problem (classification, configuration,
diagnosis, planning, etc.)
●
Type of KR and reasoning (rule-based, first-order
logic, probabilistic, fuzzy-logic, etc.)
●
Type of domain (mostly deterministic, mostly
stochastic).
CS 4633/6633 Artificial Intelligence
Components of a
Rule-based Expert System
Condition-action rules:
if <condition> then <action>
Facts
Results
How/Why
Queries
Facts
Inference
Engine
Rules
Rule-based systems
User
Interface
CS 4633/6633 Artificial Intelligence
1. Working memory: data structures representing the current
state of the system. (Facts = positive literals with no
variables)
2. Rule memory: set of condition action rules
<WM pattern> −−−> <WM changes>
(Rules can add and delete facts from working memory.)
3. Rule interpreter: applies production rules to the working
memory.
CS 4633/6633 Artificial Intelligence
Production rules in DENDRAL
Pattern Matching
Example:
IF [the spectrum of the molecule has two peaks
at masses x1 and x2 such that
a) x1 + x2 = M + 28; AND
b) x1 - 28 is a high peak; AND
c) x2 - 28 is a high peak; AND
d) at least one of x1 or x2 is high]
THEN [the molecule contains a ketone group]
Find all rules that are eligible to be “fired” by
matching left-hand side of rules to facts in
working memory
● If rules have variables, matching requires
unification (the RETE algorithm can do this
efficiently)
CS 4633/6633 Artificial Intelligence
●
CS 4633/6633 Artificial Intelligence
Forward Chaining
Conflicts and conflict resolution
Match phase:
Find the Conflict Set
When several rules issue contradictory
recommendations:
●
●
●
●
●
Is there one
Fire rule with the highest priority
Fire rule with the most specific condition
Fire the most recently used rule
Fire rule with the most recently used variable
Fire the most recently added rule
CS 4633/6633 Artificial Intelligence
Good for complex problems where humans
have expert knowledge
● Expressiveness and intuitiveness
– Rules can be understood by a non-programmer
Simplicity (uniform KR)
● Modularity and modifiability
●
– Individual rules can be changed and added
– But it is hard to analyze the effect of each new
rule on performance.
CS 4633/6633 Artificial Intelligence
STOP
Yes
Conflict resolution phase:
Choose an applicable rule
Act phase: Fire the Rule
CS 4633/6633 Artificial Intelligence
Advantages of rule-based
expert systems
●
No
Difficulties
●
“Knowledge acquisition bottleneck”
– Difficulty in acquiring expert knowledge and
putting it in system
Not good at common-sense reasoning
● Traditional systems not good at handling
uncertainty
●
– Recent techniques have overcome this
CS 4633/6633 Artificial Intelligence