|
Freethought
Association of West Michigan
Meeting Minutes for May 8, 2002, #115
As a reminder,
there is no regularly scheduled meeting at the Yankee Clipper
Library for May 22, due to a conflict with the library schedule.
In lieu of this, those interested are welcome to gather for "Happy
Hour at Kurly's Korner" at the usual time (7PM) at 740 Michigan,
NE near downtown Grand Rapids. This is where we adjourn to after
our regular meetings for informal socializing.
Plan on attending
our Annual Freethought Picnic on July 13. It will start @ 12 noon
at Hagar Park on 28th Avenue in Jenison; the Maplewood Pavilion.
Bring a dish to pass, table service and beverage. Hope to see
you there!
Our next Adopt-A-Highway
clean up has been re- scheduled for July 20. It is still at 10AM
and meeting at the Citgo station on Plainfield Ave., between 5
Mile and the East Beltline.
A newly added
topic to our schedule for this year is "The Science of Chiropractic"
to be presented by GVSU professor and chiropractor, Dr. Bryan
Mikula. This will be on October 9.
Our topic
for this meeting was to be "Meditation for Heretics"
presented by Steve Anderson. However, he is having surgery so
his presentation is now scheduled for August 28. We were fortunate
in being able to fill this suddenly vacated slot by getting Dr.
Guenter Tusch to speak to us on the topic of "Artificial
Intelligence."
Dr. Tusch
is a professor at Grand Valley State University in the Computer
Science Dept. where he teaches a course in A.I. He came to the
U.S. from Germany where he taught at a large medical school. It
was there that he had his first contact with artificial intelligence,
working, as he was, with medical information systems. There were
complaints about the inadequacy of these systems and he began
work in correcting these deficiencies, doing research at Stanford
and working on "expert systems" that can interface with
people more fluidly, resulting in more "user-friendly"
programs.
Our presenter
provided the four main categories of views on A.I. The 1st one
uses humans as a model and tries to have a system that thinks
in a humanly fashion by automating activities such as decision
making, problem solving and learning. The second example studies
how to make computers act more humanly; doing tasks, which at
the moment at least, humans do better. The third view deals the
study of mental faculties through the use of computational models
and looks toward thinking rationally. The last approach relates
to acting rationally. This is the branch of computer science that
is concerned with the automation of intelligent behavior and is
the approach that Dr. Tusch allies himself with. Later in his
presentation, he mentioned that rational behavior regards doing
the right thing, which in turn equates to that which is expected
to maximize goal achievement given the available information.
Next, we examined
the "Turing Test", named after the English mathematician
and logician, Alan Mathison Turing (1912-1954). In 1936 he described
a "universal computing machine" that could theoretically
be programmed to solve any problem capable of solution by a specifically
designed machine. The concept, which was to be called the "Turing
machine", foreshadowed the digital computer. He is considered
the first to suggest the possibility of machine learning and artificial
intelligence. The test that he proposed (in 1950) was to see if
a human test subject could distinguish between computer and human
responses to questions posed to them. The subject would be placed
in a room where he would interrogate a human in one room and a
computer in another. When the questioner would be unable to tell
from the responses given which one was the human, the machine
would be regarded as attaining a state of humanlike intelligence.
A flaw in this test is that it is not mathematically testable.
We then turned
to laws of thought including normative as opposed to descriptive
thought. We looked at the how the Greek scientist, philosopher
and logician, Aristotle created constructions that would test,
find flaws in, and check the validity of postulations in logic.
The problem, Dr. Tusch mentioned, was that not all intelligent
behavior can be seen on a formal, logical basis.
Since Dr.
Tusch's expertise involves the application of intelligent computational
systems to the medical arena, this was discussed a good deal.
The professor believes that these systems should always be regarded
as a tool to be used in the decision making process but not as
the ultimate arbiter. In medicine, in particular, there are so
many value-laden, ethical decisions involving relative risks,
costs, benefits, trade- offs, and what goal outcomes will be sought
by individual patients. The computer system can calculate data
related to various cost-benefit scenarios but may miss a piece
of critical information and cannot take into account the specific
beliefs, hopes, and long-term goals of the individual.
We talked
about how knowledge-based expert systems can tabulate, through
an extensive "if- then" checklist, the core medical
concern of the individual, going from broad analysis and funneling
down to a more precise fitting together of the pieces of information.
Dr. Tusch mentioned that this sort of approach is most effective
in engineered endeavors but loses some of its efficacy in the
more "fuzzy" domain of medical treatment. It becomes
more probabilistic and the exercise becomes one of sifting through
the different levels of probability outcomes to arrive at the
most logical answer.
Dr. Tusch
called our attention to the non-directive psychotherapist program
that was developed in 1966. This program took key words from the
patient's statements and responded mechanistically to them, sometimes
echoing the words, sometimes relating a term like "mother"
and inquiring about "family" relationships. It was deemed
to be unethical because no human was involved as a guide in how
the therapy progressed, but some patients claimed: "the program
understands me completely." This example brought with it
potential problems that could ensue when a human believes that
a program has a true understanding of the feelings generated by
him or her and sees a relationship between them that does not
accurately reflect the mechanistic program routines being utilized
by the dispassionate system. The Spielberg film "A.I."
was also mentioned in this regard. The goal is to make computer
systems that "know" preferences, respond to the individual
idiosyncrasies, interests and pursuits with minimal effort in
teaching by the user, but it would be incorrect to confuse what
the operating system "knows" about you, with it "caring"
about you.
We were next
presented with reproduced images of a few paintings and were asked
what we were able to discern from them. The first was highly abstracted.
Then there was a scene depicting more readily recognizable elements.
This made it easier to relate the abstracted forms to something
echoed in a more realistic manner. Other elements were magnified,
or focused on in subsequent pictures. This exercise was a learning
process where we became more competent in recognizing the abstracted
aspects of the pictures for what they harkened to. Professor Tusch
said that a computer needs to be able to relate things to other
known elements in the environment (or its data base) without them
being explicitly programmed in, to have an effective A.I. learning
capacity.
Dr. Tusch
next showed us how computer models could be made that used symbols
for objects. The computer would first recognize what constituted
a correct representation of the object and what fell outside the
established parameters. Then it could relate such objects to each
other correctly and then establish the functional relationships
and values added to them in different contexts; inter-relating
the assemblages of symbols. However, these models do not yet reflect
the complexity of our awareness.
Expert systems
are all limited to what is programmed into them (the quality and
quantity of the data) and Dr. Tusch showed how difficult it is
to approach many problems in a precise, rule-based way. Driving
was one example he gave of this. It is one thing for a human to
perform this task, and a very different intellectual operation,
to figure out how to program all the contingencies and subconsciously-
performed aspects of this task into a computer system.
Dr. Tusch
briefly touched on the work being done in neural networks, or
attempts to mimic the structure of the human brain in creating
an artificial intelligence. We talked about how initially people
had to adapt to the computer language and how programs ran in
a manner unnatural for humans, but that systems are becoming ever
more the reverse of this; the program adapting to the user who
interacts more normally and comfortably with it. The computer
becomes more of a partner, assisting the human in his/her pursuits
and goals.
Again, in
relating A.I. to medicine, we discussed how these systems could
find connections based on predictions from the body of knowledge
and analysis but that it was limited by not being able to take
everything into account and that not all pertinent information
could be readily measured and put into the system. The "gut
feeling" by the human is still an important element and no
matter how good the diagnostics become, diagnostics, in the end,
is not the main problem of medicine.
One of the
group members said that it seemed that a system that could initially
ask the right questions, guiding one to the heart of the matter
quickly, could perhaps bypass a battery of expensive tests involving
high-tech. scans, etc. Dr. Tusch maintained that even though there
is some overlap and redundancy involved when a series of different
tests are run, this actually makes for a built- in safety check.
Something that would be missed by one line of investigation is
more likely to be picked up on in another. One could also be tempted
in following a single line of plausible evidence down a blind
alley while the correct path remained hidden.
There was
discussion about the basic terminology-perhaps, instead of considering
"artificial intelligence" we should think in terms of
"organic vs. inorganic intelligence." The question was
raised whether inorganic intelligence exists and what exactly
constitutes intelligence. We no longer think of human intelligence
as a single thing but see the various intelligences, how they
interrelate, how one can compensate for another and how humans
can be gifted in different ways, whereas in the past, intellect
was considered mostly by achievement in areas of math and language.
We explored the question of whether computational devices could
make use of different "intelligences" should one aspect
no longer function or function correctly.
Professor
Tusch made a distinction between A.I and the data base itself,
with the analogy of the data base being like a warehouse. A.I
is what is used to access what is needed from the warehouse. A
more full warehouse has more facts that can be processed creating
a better knowledge representation.
One member
said that A.I. should recognize patterns that it was not told
(programmed explicitly) to recognize. This brought about discussion
of how an evolutionary approach has been used to arrive at creative
solutions in such fields as aerodynamics and even detergents (the
former regarding wing designs and the latter dealing with better
enzymes to clean fabrics). There were no set program parameters
or human intervention in the outcomes arrived at by the blind
design mutations. Indeed, such examples have been shown to demonstrate
that in biological evolution there is no need for an Intelligent
Designer to guide the course of life's forms.
One member
saw the "artificial" in this discussion as being about
an artificial distinction between our human mental functions and
the way that a computer works. He laid out how what we term "will"
is decision making based on scanning our data base and intuition
is merely unconscious scanning of the same and talked about classical
conditioning, reinforcement and motivational goal states.
It was mentioned
that a truly autonomous machine would have to gain its "experiences"
directly from the external environment and learn by its interactions,
trial and error, and expanding data base of connections arrived
at by these interactions---much as a growing human child does-rather
than having it all programmed into it by human operators.
Dr Tusch mentioned
the research being done into applications for A.I. in military
and in space exploration. The former need to have autonomous action
for effective operation in the field, in place of a human agent
being put in danger. With operations in space, explorational devices
must function without constant control by humans, since real-time
connection cannot be established and maintained as the devices
perform their tasks and they must be able to respond to a wide
variety of novel circumstances.
Secretary:
Charles LaRue
|
