Artificial Intelligence (AI)

Download Report

Transcript Artificial Intelligence (AI)

Artificial Intelligence (AI)
Can Machines Think?
Advantage computer:
•
•
•
•
•
•
Calculate
Communicate
Process information
Storage and recall of facts
Make decisions using established rules of logic
Consistency
Advantage human:
• Perceive
• Reason
– Not all possibilities can be anticipated, and therefore
programmed
• Recognize patterns
– Unless a specific pattern has been anticipated and
‘programmed’, a computer can’t act on it
• Ambiguity
• Application of knowledge (child describing his
toys)
So, can they think??
• The “Turing Test”
– Developed by Alan Turing (1950)
– A person sits at a computer and types questions into a
terminal.
– If a computer were truly “intelligent”, the questioner
would not be able to determine whether the
responder was a human or a computer
– To date, no computer has even come close
– Some still consider the Turing Test to be the best
determinant of AI. Other researchers favor a more
lenient definition.
Defining AI
•
•
•
•
Hard to define
Many disagree
“…ability to perceive, reason, and act”
“…do things which, at the moment, people are
better”
• etc
Was Deep Blue “intelligent”?
• Deep Blue was a computer developed by IBM
that defeated Kasparov in chess.
– Rules were clearly defined
– Objectives were unmistakable
– Searching: Winning typically goes to the player who
can sift through the greatest number of possibilities
and outcomes
– Recall: Pattern recognition of board patterns and best
strategies to employ given a specific pattern
• Humans may have the edge here…
– $25 chess programs can defeat the greatest players in
the world
Language Translation
• Still work to be done…
• Shakespeare: “The spirit is willing, but the
flesh is rotten”
• Computer: “The wine is agreeable, but the
meat is rotten”
• “Out of sight, out of mind”
• Computer: “Invisible idiot”
Syntax vs Semantics
• Language rarely limits itself to a consistent set of rules and
structure
– There are always “exceptions”
• Sometimes, understanding the true, underlying meaning of
a single word can require a great deal of knowledge
• Syntax: the ‘rules’ of a language, definitions of words
• Semantics: the underlying meanings
–
–
–
–
–
–
Expressions
Idioms
Slang
Visual cues
Ambiguity: e.g. All that glitters is not gold.
Etc
Practical applications of AI
• Knowledge bases
• Expert systems
AI techniques
• Heuristics
• Pattern recognition
• Machine learning
Knowledge vs Facts
• Facts are details that are typically quantifiable
and reproducible
• Knowledge is the ability to form relationships
by using facts
– Humans are considerably better at inferring things
– Computer require tremendous input of data to
accomplish this same task, and even then, will
inevitably fall short at some point
Knowledge Base
• A computer KB will
1. Incorporate a database of facts
2. Incorporate a series of programmed rules
3. Attempt to derive new facts by applying steps
1 and 2
Expert Systems
• “A software program designed to replicate the
decision making process of a human expert”
• A collection of specialized knowledge where
facts and appropriate actions are obtained
from expert sources and programmed into a
database
• Usually involves a series of “IfThen”
question and answers.
Algorithms
• An expert system will frequently use a series
of algorithms to provide solutions to a given
question
• Here are a couple of examples of wellestablished medical algorithms:
Difficult Airway
Algorithm
ACLS Algorithm –
Cardiac Arrest
Pulmonary HTN Algorithm:
Fuzzy Logic
• Uncertainty is an inevitable part of the human
experience
• Computers do not handle ambiguity well
• Computers use likelihood (e.g. percentages) –
derived from as much factual data as possible
– to come up with the “best” solution
Expert Systems - examples
• Training
– Teaching “difficult airway” procedure to
anesthesiology residents
– “What do you do next?”
• Routine / repetitive task work
– Monitoring millions of credit card accounts for
unusual activity
• Expertise when human help is not available
– PDAs with medical databases
• Error reduction
– Checking for drug interactions in an EMR