Testing our Assumptions about AI: Whose
knowledge is represented in expert / knowledge systems?
CYC and its critics- March 24, 2003
Overview of lecture:
1. What is CYC?
CYC is "a special kind of expert system" (page 3 kit) because
it will involve depth as well as breadth.
2. Problems faced by Lenat's team
developing CYC?
--> how program in common sense?
--> how program in capabilities for inconsistencies?
3. Adam's Critique of CYC
*******
1. What is CYC?
What knowledge should be represented?
" Although we expected encyclopedias to
play an important role, we soon realized
that what they contain is almost the complement of common sense. Assuming that
readers already
have common sense, can read, and so ,
they provide the next level of detail for reference purposes.. . ."
--> heir first example from an encyclopedia was
" Napoleon died on St. Helena. Wellington
was greatly saddened."
Moved from a bottom-up analysis of
background knowledge you need when moving from one
sentence to another to a top-down approach, "treating entire topics
one at a time and in moderate detail.
By 1996, we had told CYC about hundreds of topics."
This allows for inconsistent facts
and rules in the database; -- example of
Dracula...
As he says "we know that there are no vampires and that Dracula
was a vampire."
Applications of CYC:
- checking databases for inconsistent
facts
- sophisticated matching across databases;
example of matching "a strong and adventurous person" to
a caption of "a man climbing a rock face." To do that, it
used a few rules of the sort:
IF people do something for recreation
that puts them at risk of bodily harm,
THEN they are adventurous."
3. So what problems does Adams have
with CYC and what approach does she take to
critique his "common sense" expert
system?
3.1. Why is Adam's feminist epistemological critique of AI "more radical than, and qualitatively different
from traditional philosophies of AI"?
because 1) traditionalists are, by and
large, concerned with whether AI can
even be achieved, about the possibilities
of AI
- that is pass the Turing Test (or some
other criteria), and,
2) they don't take into account gender
issues...
3.2. What is epistemology?
--> theories of what knowledge is, and who "knowers" are,
and a consideration of how we know what
we know.
--> If you're going to build an AI "knowledge" system
you are in the realm of
epistemology ....
3. 3. What is feminist epistemology?
-> an examination of knowledge that considers whether women's ways
of knowing are different from men's ways
of knowing and if so, whether women's
ways are given due consideration.
3.4. Why is Adam's concerned with
AI knowledge systems-specifically CYC?
-> because we need to ask how
the knowledge inside the system is used and
what knowledge it uses. (If it's "male-centered" knowledge
than women are the
losers.
3.5. Is she unilaterally against
the use of knowledge systems?
-> no, she gives the example of the
expert system that would help women find
their way through sex discrimination
legislation to determine if they have
a case. In this instance, the application
is for women's use and incorporates a "view
from feminist legal
theory which challenges the views of
traditional jurisprudence in excluding
women's experiences from the development
of legal knowledge." (kit page 2))
-> see last line of article: she wants
to "become involved in the newer
areas of AI which are consciously addressing
the need to involve the body in the representation
of knowledge."
6. How does Adam make use of feminist
epistemology to make her points?
1. problems with the idea of the
(male) knowing subject in 'S-knows-that-p'
2. problems with the emphasis on
propositional knowledge in knowledge systems (for example
CYC) --the "knowing that" over
the knowing how" (skills) which
privileges male knowledge over female
knowledge, and
3. acknowledgement that the role
of the body is crucial to the creation of knowledge.
1. `S-knows-that-p' means that S
is the knowing subject who is taken to be
universal, and p is a piece of propositional
knowledge that the S knows.
-> the Subject is not a specific individual,
rather the subject is the ideal subject
not situated anywhere...
-> start with observational claims
about everyday objects -- S knows that
when
you drop a ball it falls to the ground,
but once you start building up the p's
(the facts) they may not be quite as "neutral" and
may indeed represent a gendered belief
rather than "knowledge."
->some people may think they know
something, but it may just be a belief
rather than real knowledge.
real issue: who is doing the distinguishing
between knowledge and belief?" e.g.,
If doctors deliver babies,
THEN we will have healthy babies.
if this is coded as knowledge by
TheWorldAsTheBuildersofCycBelieveItToBe
and another S claims that
If midwifes deliver babies,
THEN we will have healthy babies,
And this is coded as a "belief"
-> it's not part of the building up
of CYC's database of knowledge.
In short, what if the knowing subject
in CYC has a male bias?
2. Two types of knowledge: knowing
that ('p) and knowing how
Knowledge in CYC is represented in logical
rules that you can state as IF....THEN
rules
yet there are a vast array of things
that we know because we have done them--
skills develop that may be tacit
knowledge (as I described Kasparov's chess
skills)...much of it can't be written
down.so because it isn't in the database,
much of women's work knowledge will be
ignored and not valued.
3. Role of the body in making knowledge
earlier philosophers (such as Descartes)
saw a split between the mind and the
body-elevating the mind over the body
women's work has to do with the body
(reproduction itself, caring for the
body (housewife), nurse, etc. are stereotypic
women's work.
So if you want to break this sexist
dichotomy, wouldn't it be o.k. to have AI expert
systems that just focus on the "mind"?
(this argument is made by other feminists)
NO, because for Adam the body is
so important to concepts of intelligence :
For her, "cognitive skills can [NOT]
be treated as disembodied from perceptual-motor
skills. [This is] becoming one of the
most important issues for debate in AI."
(refer to first lecture on AI in which
motor skills is seen as auxiliary skills)