[
Lists Home |
Date Index |
Thread Index
]
From: Didier PH Martin [mailto:martind@netfolder.com]
Thank God it's Friday!
Len said:
I would say that there is no
basis for the web in implicit or tacit ontologies. There are people
who have habituated themselves to certain sign/signified relationships
and the web is a means of amplifying the speed of that habit forming
behavior.
Didier replies:
Hummm, if you look at this through the perspective of behaviorism yes you
are right. If you look at this through the lenses of cognition, there is a
mental model and a tacit ontology in their mind. Keywords people are typing
reflect their view of the world.
Len replies: Yes. They have acquired these keywords by habit formation.
This is behavioral. The results modify the behavior and they search again.
One can ask the question, is the band jamming or is the jam banding? Both
models are useful. I like the band model because it deals with practiced
signal generators who must process information faster than real time. They
do this by discerning setup patterns that can be modified in real time to
produce behaviors that are novel to the observers.
>Recently I came across this through a
>site's log. Some people used "green tea weight loss" others used herbal
>weight loss". You can notice here that some are using a more general term
>than the others. Statistically, we can notice that about these keywords are
>closer or farther than the concept they are searching for.
The issue is the degree of certainty one has apriori that the concept you
have identified is the attractor. They can disagree. That is why WordNet
has word senses. Yes, we can construct models based on communities of
discourse (more below) and if the user has opted into membership, require
them to ontologically commit.
>What we can do though is constructing a working model of this tacit
ontology in order to
>find what are the different keywords related to a particular concept and
how
>far from the concept they are. We may use possibility theory to formalize
>such distance. Then, to process this we can encode it in a language able to
>express this type of relation. They are not judgment call if their thinking
>is right or wrong, this is simply an a posteriori model.
Sure, and then deal with the drift created through the interface and
the interactions. Some signs have a statistically stable relationship
to the signified through the signifier, others change at some rate.
What one wants to find is a relationship that is changing nonlinearly.
That is interesting.
Len said:
Again, this is not a revolution. It is just more of the same kind
of thing the marketing companies have done to us and with us via other
media since cave painting. They infer the sign/signfier relationship,
then manipulate the presentation to manipulate the behaviors of the
wetware.
Didier replies:
OK if the term revolution is too strong for you, let's forget it. However,
if we have the possibility to process concepts emerging from facts created
by people _and_ automated agents like search engine, we have reached a new
level of computing.
Len replies: Computing has reached the level of a marketing droid. It is
mapping and the maps are used to influence opinions and behavior. This is
all good. It becomes problematic when it becomes authoritative. That is
the witless part of the web. Too easily gamed, too easily used to con,
too influential for its accuracy. In short, an amplifier with filters
and a lot of people with input and varying competencies but equal access.
>I agree with you, though, that in the last 20 years AI
>hasn't delivered on the initial promises. Not because these systems doesn't
>work (just look at the results of Oncocyn, mycin and other similar systems)
>but more because they are costly to build.
Len Replies: AI does work. Always has. It is not a generalized problem
solver because it became apparent in the 19th century that no such thing
is possible. The recognition of domain boundaries, even if semi-permeable
and drifting, is fundamental to AI, and particulary ontologies. A set and
the operations on the set. A topic that shows up frequently on this list.
It is the selection of the set, the abduction, that is most fascinating.
The rest is troll work. :-)
Len said:
These are made. You are making the Platonic error of assuming these
are real in the sense of 'from nothing'. They are an output from a
feedback mediated system in which the humans are the initiating source.
Didier replies:
Yes they are made and are expression of some internal process called
cognition. Who told you that I am making the platonic error of assuming
these are real? Certainly not me! I just said "there exist a corpus of
relations between >keyphrases/topics/themes/concepts". These things are
mental constructs nothing more nothing less.
Len replies: Good. They are constructed.
>There is noting in this
>statement that says that it existed before human beings. Simply that we can
>observe, deduct, infer that there exist a corpus or relationship.
How do we choose what we observe? What does what we observe choose
to present? What details are of interest to us? Which are not?
We induct the relationship; then we deduct results. One can't jump
over the induction aspect of ontology building.
>And I do
>not know if these are the result of a feedback loop.
Len Replies: There is and it is inferable from the semantic drift. For
that to occur, there must be a coupler. Again, the interesting quest is
for examples which drift in a nonlinear way.
Len said:
We can automate the heck out of it. How predictive is it?
Didier replies:
Sorry, I do not know yet how predictive it is, since I am only at the
beginning of studying this phenomenon. I can only hope this research can
lead to a predictive model. Note that I said model not platonic certainties
Len.
Len replies: But the predictability of semantic certainty is precisely
what one wants to measure. Otherwise, the reliability of the ontology
and any decisions made on it are suspect. It doesn't mean we can't use
the tools, it means we must take great care with them.
Len said:
A University of Pennsylvania study published a fascinating study.
When it comes to predicting conflict outcomes, role playing games
(RPGs) beat game theory and statistical analysis three to one. For
those who look at the semantic web and say, "so but so what", I look
at it and say, "what a rich mine for feeding RPGs!" and realize that
in combination with distributed online game systems (which, folks,
are a generation or two beyond distributed hypertext or hypermedia depending
on your particular semantic community), this is an incredible area
of innovation for commercial companies. Keep the wetware productive.
Blogs are just people thinking out loud. Online gaming is people
behaving together. It is community building, socialization,
and a very powerful means to predict and influence behavior. Use
the semantic web technology to feed metadata (situation semantics)
to a massive RPG and you have a very powerful and influential web
application.
Didier replies:
Probably in that context and that purpose. But I doubt this method could be
useful when two sources on the web state something differently about the
same concept. Role playing won't help. We need another way to resolve the
issue.
Len replies: What is to be resolved? Which definition or use is correct?
Can you do that without the role in which it is used? Conflict prediction
is precisely what role playing is applied to; to show possible sitations
and paths by which the commitment to the ontology occurs. In other words,
the outcome of the conflict is the commitment.
>Science is using a certain process to resolve this kind of issues.
>Can we automated that? Several proposals are on the table. OWL and RDF and
>the official semantic web technologies coming from W3 do not help. Is there
>a way to do it efficiently and economically? The question is open.
>This said, I agree with you that role playing games are very useful to
>uncover the other side motives. Being in the shoes of the other parties has
>always been the basis of any negotiation process.
Their motives ARE their meaning. The ontology either reflects that or
is in conflict. Again, is the ontology the map of their meaning or the
means to control their meaningfulness? Does the band jam or does the jam
band?
>Does it help to discover the truth?
Who's truth? I'm not trying to be relativistic. I want to point out that
the toolmaker shapes the tool and the tool shapes the toolmaker and that
for any ontologically based means of determining truth, there is the
potentially
non-linear coupling. One can look for non-linearity in the web, but
all it is is the tool, the map. It is only one side.
>Not necessarily. Does it help to convince or influence? Probably.
>Does it help to find what statement between the following two is closer to
>reality? I doubt.
>a) The earth is revolving around the sun
>b) The sun is revolving around the earth.
It helps a lot if you want to put anything in orbit around the moon.
BTW, the three body problem is a classic nonlinear problem.
>This is precisely what an active agent would have to resolve in the real
>world semantic web: conflicting statement about the same reality. Is it
>possible to imagine a mechanism that would help provide a solution to this
>issue? I do not know, I am just exploring....
It isn't the same reality for any useful meaning of that. It is the
ontological commitment. Pick a groove and go for it. If the band
is practiced enough to see the setup, they fall into the groove. If
not, it sounds like crap and the perception of the third party
observer is that the band sucks. In this case, the jam bands.
One can't resolve a conflicting statement without entering the
system, and then it can drift unless all parties agree that the
meaning is the same so no conflict exists. If the conflict is
unresolvable, the meaning is not the same and the term has a
weakened in the entropic sense (in web terms, you need a new
URI because the old one just bifurcated).
Len said:
Yes, and that is a feedback loop. The ontology is emerging from the
'what others are saying' because an ontology without a community of
discourse is just a blog without context. The machine can extract
a faux ontology, but it requires agreement to become anything of interest.
Didier replies:
Usually a feedback loop modifies the behavior or state of an entity. To be
less fuzzy and more precise, what is modified? Is this the ranking you are
talking about? Do you mean that the resource ---> concept/class relationship
is changed because of the number of links? Is this really a feedback loop?
Len replies:
The number of links is being used to derive the ranking of the meaning. The
question is not that; it is how the decision to link was made initially.
The web is a fairly unhealthy source of scholarship because it enables a
culture of meaning to grow up without attribution of sources simply by
the act of linking. The feedback loop, (John says. Mary says John says.
George accepts what Mary says. John respects George's opinions. John
says He Said.), is used to confer authority when actually it only amplifies
the initial signal. It is a good means to qualify a search because in
the majority of cases, people are reliable evaluators given good initial
information. But the question makes the case. Is John the Walrus? Is
the Volkswagen license plate with 28IF on it in the Abbey Road photo really
a sign that Paul is Dead. Without care, one can extract any pattern
and feed it back to such a system and after a number of iterations, the
system will take the shape of the pattern, particularly if any where
in the system, the signal loops back on itself and scales. That is
the model of superstitious acquisition and the danger of acquiring
patterns by abduction.
Len said:
The nonlinear systems effects that Roger is looking for are not in the
web itself although it maps them much the way one maps nonlinear equations
in n-dimensional graphics. The web is not fractal; it maps fractal
behaviors. They are a result of the active interface between the
humans who create information and the humans who consume information
and then create more information. The result is a drift in semantics
which do have attractors, but which are mathematically, a lattice of
theories. The semantic web can be a means to equalize, tune or
otherwise filter the signal being amplified, but it isn't the source,
and if presented as that, that is a power grab by those who wish
to control that signal.
Didier replies:
>I do not know if the semantic web will be the source. I can only, with a
>certain probability, state that if more and more data/information is
>published, we'll need to know what is behind it. What is the meaning of,
>what is the mental model behind this data? For instance, how can I know,
>without having certain knowledge of its country of origin what could help
me
>understand without errors the following statements?
>a) Joe is using rubbers (If this is slang I have a complete different
>meaning :-)
>b) Joe is a chemist (completely a different meaning if the source of this
>statement is from England or from the USA)
>As there is more than one English (or more than one French or more than one
>Portuguese) there is more than one language. If English and American do not
>agree on the same meaning for the same words(Ex: chemist) or French from
>France and French from Quebec not agreeing on the same meaning for a
>particular word (Ex: poudrerie), imagine the potential confusion between
>totally different language and cultures. Can ontologies help? This is
>precisely what I am exploring....
Yes. Welcome to HumanML. To make that evaluation work, you need an
ontology
of the communicators to create the context of the communication. The
ontology
for that is simple; the amount of data it requires to work is substantial,
and
again, one is back to limiting the domain of the abduction. In a semiotic
system, the communicators are recursive semiotes, and they are in an
environment
which is also semiotic. Behavioral cybernetics.
But I'm not interested in classifying every living being past and present. I
am interested in instances of the models that can be used in simulations to
make rough predictions. So I would feed HumanML data to RPG models and
watch them. Beware the local minima problem of annealing systems. HTML
is a good example of that.
len
|
