Copy Link
Add to Bookmark
Report
AIList Digest Volume 5 Issue 285
AIList Digest Friday, 18 Dec 1987 Volume 5 : Issue 285
Today's Topics:
Query & Puzzle - Probability Bounds,
Announcements - Simulation List & Issue of DAEDALUS on AI,
Philosophy - The Role of Biological Models in AI
----------------------------------------------------------------------
Date: 11 Dec 87 14:10:21 GMT
From: mcvax!ukc!its63b!hwcs!tom@uunet.uu.net (Tom Kane)
Subject: Probability Bounds from Bayes Theory: (A Problem).
I am sending this letter out to the network to ask for solutions to a
particular problem of Bayesian Inference. Below is the text of the
problem, and at the end is the mathematical statement of the information
given. Simply, I am asking the questions:
1) Can you find bounds on the final result. If so, how?
2) If not, why is it not possible to do so?
What is missing in the specification of the problem?
3) If you get nowhere with this problem, would you be able to solve it
if you were given the information: p(pv|t or l)=0.9?
I am interested in the problem of providing probability bounds for events
specified in a Bayesian setting when not all the necessary conditional
probabilities are provided in setting up the problem.
PROBLEM
~~~~~~~
(A problem relevant to the handling of Uncertainty in Expert Systems.)
We want to know the probability of a patient having both lung cancer and
tuberculosis based on the fact that this person has had a positive reading
in a chest X-ray. We are given the following pieces of information:
1. The probability that a person with lung cancer will have a positive
chest X-ray is 0.9.
2. The probability that a person with tuberculosis will have a positive
chest X-ray is 0.95.
3. The probability that a person with neither lung cancer nor tuberculosis
will have a positive chest X-ray is 0.07.
4. In the town of interest, 4 percent of the population have lung cancer,
and three percent have tuberculosis.
EVENTS
~~~~~~
l = lung cancer; t = tuberculosis; pv = positive chest X-ray
SETUP
~~~~~
In the statement of the problem below:-
~l means 'not l'.
~l, ~t means 'not l and not t'.
t or l means 't or l'
where 'not', 'and' , and 'or' are logical operators.
so that: p(~l, ~t) means probability( not l and not t).
Also,
p(pv|l) means the conditional probability of event pv, given event l.
PRIORS
~~~~~~
p(l) = 0.04; p(t) = 0.03; p(~l, ~t) = 0.95
CONDITIONALS
~~~~~~~~~~~~
p(pv|l) = 0.9; p(pv|t) = 0.95; p(pv| ~t,~l) = 0.07
(You are not given p(pv| t or l) )
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Please mail all solutions or comments to me, and I will let interested parties
know what the results are.
(I will specially treasure attempts which don't use independence assumptions.)
Thanks in advance to anyone who will spend time on this problem...
Regards,
Tom Kane.
------------------------------
Date: Thu, 17 Dec 87 09:30:06 EST
From: Paul Fishwick <fishwick@fish.cis.ufl.edu>
Subject: New Simulation List
------------------------------------------
****** NOTICE: NEW MAILING LIST **********
------------------------------------------
on
S I M U L A T I O N
GENERAL:
There has not been a news group on the topic of simulation, so I have
decided to start one. Actually, it is a mailing list and if it grows
into a popular forum then we can formally apply to have it made a
news "group" (which apparently requires votes,etc.).
TOPICS:
All topics connected with simulation are welcome (no flaming please!).
Some sample topics are listed:
real time simulation methods
flight simulation
parallel architectures for simulation analysis and modeling
simulation and training
distributed simulation
artificial intelligence and simulation
automatic generation and analysis of models
analog vs. digital methods, hybrids
continuous, discrete, and combined methods
qualitative modeling
application specific questions
theory of simulation and systems
queries and comments about available simulation software
announcements of simulation-related talks and seminars
graphics and image processing in simulation
HOW TO JOIN:
To participate in the mailing list you need to know two net addresses:
simulation@fish.cis.ufl.edu - for sending topical mail messages
simulation-request@fish.cis.ufl.edu - for subscribing/unsubscribing
to the list (administration)
METHOD:
At first, we will operate on an automatic mode (unedited list): All
mail sent to 'simulation' will be forwarded automatically to everyone
else on the list. My SUN is strictly acting as a mail handler. As
interesting topics come up and more people chip in, then I will try
my hand at moderating the list to form a digest which will be shipped
periodically. I'm sure that most net readers subscribe to both kinds
of these mailing lists already. So let's go!
Paul Fishwick
University of Florida
------------------------------
Date: Mon, 14 Dec 87 15:22:40 EST
From: amcad!billb@husc6.harvard.edu
Subject: New Issue of DAEDALUS on AI
In response to numerous queries re. forthcoming issue of DAEDALUS on AI,
we would like to provide Table of Contents for this 320-page volume and
information on how to get a copy.
Contents include essays by the following:
Seymour Papert - "One AI or Many?"
Hubert L. Dreyfus & Stuart E. Dreyfus - "Making a Mind Versus Modeling
a Brain: AI Back at a Branchpoint"
Robert Sokolowski - "Natural and Artificial Intelligence"
Pamela McCorduck - "Artificial Intelligence: An Apercu"
Jack D. Cowan & David H. Sharp - "Neural Nets and AI"
Jacob T. Schwartz - "The New Connectionism: Developing Relationships
Between Neuroscience and AI"
George N. Reeke Jr. & Gerald M. Edelman - "Real Brains and AI"
W. Daniel Hillis - "Intelligence as an Emergent Behavior; or,
The Songs of Eden"
David L. Waltz - "The Prospects for Building Truly Intelligent
Machines"
Anya Hurlbert & Tomasio Poggio - "Making Machines (and AI) See"
Sherry Turkle - "AI and Psychoanalysis: A New Alliance"
Hilary Putnam - "Much Ado About Not Very Much"
Daniel C. Dennett - "When Philosophers Encounter AI"
John McCarthy - "Mathematical Logic and AI"
Copies of this volume of DAEDALUS are available @ $5 each ($1 additional for
surface mail delivery outside the U.S.) by writing to:
DAEDALUS Business Office
P.O. Box 515
Canton, MA. 02021 U.S.A.
Email orders can be sent, along with credit card billing number to:
daedalus%amcad.uucp@husc6.harvard.edu
or
harvard!husc6!amcad!daedalus
Holiday greetings,
Guild Nichols
DAEDALUS
------------------------------
Date: 15 Dec 87 03:06:02 GMT
From: marque!gryphon!sarima@csd1.milw.wisc.edu (Stan Friesen)
Subject: Re: the role of biological models in ai
In article <23@gollum.Columbia.NCR.COM> rolandi@gollum.UUCP () writes:
>
> According to some AI theorists, (see Schank,
>R.C., (1984) The Cognitive Computer. Reading, Mass.: Addison-Wesley)
>AI is "an investigation into human understanding through which we learn
>...about the complexities of our own intelligence." Thus, at least for
>some AI researchers, the automation of intelligent behavior is secondary
>to the expansion and formalization of our self-understanding. This is
>assumed to be the result of creating computational "accounts" of (typically
>intellectual) behavior. Researchers write programs which display the
>performance characteristics of humans within some given domain. The
>efficacy of a program is a function of the similarity of its performance
>to the human performance after which it was modeled. Thus AI programs are
>(often) created in order to "explain" the processes that they model.
>
My problem with this class of AI research is that I question it
validity/usefulness. Why should there be only *one* algorithm for a
particular 'behavior'? What evidence do we have that the algorithms that
we are writing into our programs are in fact related in any way th the
ones used by the human brain? Mere parallel behavior is NOT sufficient
evidence to claim increased understanding of a human behavior, some
evidence from neurology and psychology is necessary to at least
demonstrate applicibility. In particular, I find most current AI
algorithms to be far too analytical to be realistic models of human,
or even animal, cognition.
------------------------------
End of AIList Digest
********************
