This is a simple scenario building excercise for a Telecommunications & Information Technology which I thought of conducting online. This is just an experimnent and the results of this could lead us to opportunities and warn us about coming Threats. I invite one and all, related with it to participate in it. The book "The Art of Long View" by Peter Schwartz the master scenario-builder said that in such an excercise, it is important to learn the opinion of everyone related to the focus of the scenario...and so here we go....
This is a simple scenario building excercise for a Telecommunications & Information Technology which I thought of conducting online. This is just an experimnent and the results of this could lead us to opportunities and warn us about coming Threats. I invite one and all, related with it to participate in it. The book "The Art of Long View" by Peter Schwartz the master scenario-builder said that in such an excercise, it is important to learn the opinion of everyone related to the focus of the scenario...and so here we go....
Dr. Deborah McGuinness is associate director and senior research scientist of the Knowledge Systems Laboratory at Stanford University. Here she answers audience questions on her presentation of how the semantic web works.
Question: You haven't said anything about scope or scale. In your experience, what's a measure and what's the size of like a toy example or something that was minimally practical or would really have some commercial life to it?
McGinness: I think you can get commercial bite with a small footprint. I started at AT&T; at the personal online services division. We were doing very specific medical applications. So we were only doing lipid disorders, for example. And then we were also only doing pharmaceuticals. They were aimed at both laypeople and primary care docs.
-- ADVERTISEMENT --
So I guess what could be behind your question is do I have to build some enormous ontology that I have to get somebody like me to build that might be expensive or hard to find, that might take a long time to build before I can actually start to get payoff? And I think the answer is no there. I think you can stay in small domains and still have something that somebody would find value in, thus something that somebody would either pay to advertise on or pay for a service—as well as applications that could benefit by having much larger ontologies.
Question: It seems that you have an ontological framework, but it is not common across all fields. In other words, for each one you're really building ontology relative to that particular view for that particular set of applications. Your high school example highlighted this with the particular word choices where there might be other words that would also be common to this and would create maybe a foundation, from which all of these ontological frameworks could build up. Is that a difficulty, to have something in one domain that might be relevant to another domain but they're not going to be able to cross-connect each other because of the different views of those particular ontological frameworks?
McGinness: There's a lot of debate over this. Some people believe that there should be one foundational ontology, and everything should fit into that. You see proponents of SUMO, the Suggested Upper Merged Ontology—and arguably you could fit Cycorp in this—that gives out a large upper ontology benefit from connecting into. You see also people supporting that within the UNSPSC example. They claim that they're the top-level structure for all the business to business. "I don't care how many different words you use; just fit into this top integrating structure." There're a lot of benefits to that; it supports interoperability better. The knowledge representation community would support that view.
When I was a pure academic, I would have come down quite hard saying this is the way it should go, because of all the benefits. [Now I say] if you can operationalize that space, great. Then I'm going to exploit all your benefits up the wazoo, but I don't want to only be in that space because look at the web: There are tons of people putting information into Wikipedia and these kinds of naturally occurring, big information sources. I can't train all of them. No matter how much literature I put out and how many little examples I give to cut and paste from, they're going to do it the way they want to do it.
In fact, I want that, because even though I like building ontologies, I want the unwashed masses to do it. So I've got to find ways to be able to work with this little cluster of information over here and that over there, and there're a lot of different methods. One is tools for merging for interoperability, for recognizing when two terms might be the same. And I think we just have to work in hybrid space. And for the web to really take off, we can't be sitting in an academic ivory tower perspective saying, "Thou shalt use my language."
Question: I'm glad you brought up Cycorp. What are some of the things they found in terms of what it takes to go out and do this kind of unification or what the pluses and minuses are on that?
McGinness: Cycorp is a great example. They're I think the longest running and largest really big, complicated, deep ontology. Last time I was there I think they had 400 different inference engines. But there're a lot of inference engines and a lot of different contexts and views of the world.
They get tremendous value, and they get much deeper reasoning than we're seeing emerge in the distributed world of the web. They do a very good job, and they've got great training classes. They actually have a little army of people who built this stuff, and they have one or two gurus who do handle the top-level structure. And then they have smart, trained people but not rocket scientists doing integration, and they take feedback for updates.
That's not an unreasonable way of maintaining these big complicated structures, and you see social networks emerging where UNSPSC might be one level down from that. They're only doing taxonomies, but they're doing it on a much broader basis, and they've got this social process about the way you ask for updates.
You see a spectrum from the example of Cycorp to this middle ground of here's the way you ask for updates, to maybe the Wikipedia world where here's this big space.
Question: Within the various industries like financial services, manufacturing, they all have worked for the last 10 years on various markup languages for structural document standards. Do you see what I and some people see, that they're slower in moving up to semantic modeling? How do we get the industry standards that have been traditionally on the structural side to move up? How do you encourage semantic modeling?
And could you make some comments about the use of semantic web not within web but within the enterprise—for people who think that building composite applications would require or would benefit from semantic modeling?
McGinness: How do I get them to move up? First I compliment them, and say what you've done so far is a great starting point—because they actually have started. The fact that they've got some markup language at all is a starting point.
Then I say, What can you do: You can do some limited kind of retrieval, but you can't do some additional small kind of structured retrieval that you would like. So in whatever domain they happen to be, I try to find a couple of structured retrieval applications that they really want to do but their current markup is not allowing. But yet, it's really not hard to show them that with this one additional field and this small additional set of classes, they could get that information.
I typically stroke their backs, stroke their ego, find two compelling examples, and then show them in basically one viewgraph that it's not hard how to get benefits. If I'm consulting with them, I make sure that we've got a deliverable in a month or a short period of time where they can do this one trivial thing that adds value. Kind of get them on the slippery slope so that they're the owner and they want to do it themselves.
And then for enterprise versus the web, I think this is a slippery slope. You could see webs behind firewalls; the enterprise world is both more fun and more challenging, and sometimes easier and harder. One way it's easier is sometimes you can put handcuffs on people. You can say, Here are your three tools, and you can constrain their environment to get faster results and less complicated results. They don't have to integrate with the whole rest of the world; they only have to integrate with the team sites instead of those 10,000 sites.
There I typically find a couple of most important applications to integrate with dealing with firewall issues or security issues or whatever is paramount, and show—I guess I follow the same methodology—show some quick win in some particular case that they care about, ignoring the web but considering whatever they think is the most important about their enterprise, either their security or their firewall or integration with this particular system.
One of my mantras is always start small. Show some win in some small domain. Don't under any circumstance start with saying, I'll just build you this enormous ontology for the next two years and then your world will be better. Just say I'm going to build this tiny little ontology and enable this small application over here. I do this kind of iterative development, always making sure that my small ontology is enabling the small win.
Tell us what you think of this post using our On or Off rating system. Only your most recent vote will count.
Member Comments
Cycorp seems to be show questionable commercial success, esp. given the amount of funding spent after it. And there are lot more successful ontologies and ontology-driven applications than what some members seem to have been exposed to. From commercial perspective, it is possible to show much better commercial/industry examples at Enterprise level. Here are some commercial companies based on academic research (at least one of these has deployed ontology-driven semantic application at Fortune 500 customers; all have developed ontologies or have tools that are used to develop ontologies that seem to have been put to practical/commercial use):
# Univerisity (Researcher) -> company -> (Example presentation or publication)
* Univ. of Georgia/LSDIS lab (Prof Sheth) -> Semagix http://www.semagix.com -> (see http://lsdis.cs.uga.edu/lib/download/S+2002-SCORE-IC.pdf)
* U. Manchester (Prof Horrocks, others) -> Network Inference -> www.networkinference.com
* Karlsruhe (Prof Studer) -> Ontoprise http://www.ontoprise.com
You've made a very poorly construicted point: but let me be brief about it. You assert that there does not exist a search tool which incorporates all the features of Autonomy then negate the evidence that it could be applied commercially as an online search tool. I suppose your real contention is with the GEE-WHIZ kids over at Autonomy, or maybe it is in the context of how YOU use the tool; I've never had a problem finding good stuff.
"May be someday, Google or any search engine will come up with a personal engine that learns over time, what one usually searches for, its context and creates a predictive model ahead." It is already here; and has been here for some time: http://www.autonomy.com
For Google to get anywhere near smart, the system has to receive feedback as to the context of the search and another feedback from users who find what they are looking for, after 6th page. On top of that, the threshold of each context changes for the same user and could be wildly variable for the next user. As Google keeps tweaking, people are looking for alternate search engines. Some swear by Teoma and other are waiting for search engines that are topic based. May be someday, Google or any search engine will come up with a personal engine that learns over time, what one usually searches for, its context and creates a predictive model ahead. It will be interesting if anything really comes out of Stanford anytime soon.
The idea of constructing an ontology seems a bit flawed or challenging to say the least. the structure of a large functional ontology (the apparent goal) seems to be probably so incoherent (and N scale design challenging) in potential linkages and rules sets as to be impossible. An opne ended feedback architecture seems to make a lot more sense. Wikipedia is a nice human powered mechanism for limited self referential ontology although one could argue it is more like a robust form of taxonomy.
An open architecutre allowing for a loosely connected probabalistic associative network seems more functional than a fixed or even "human designed" semi-variable ontology. Even a fixed ontology would seems challenged from a contextual or reference perspective (no pun intended). The ontology should in theory be non-hierarchical such that the value or importance of a structure within it varies as to the context at hand. This is implicit in the concept of anthropically creating AI. Conciousness percieved a single entity vs. perpetual branching etc. My bet is a successful ontology won't be created.
A successful open ended feedback architecture producing simple ontologies with variable references will be created. These multiple small variable ontologies could then coalesce to form a more anthropically recognized intelligence. My guess is like most robust functioing systems the constructs of such a thing would respect a power law giving the perception of a "center" of intelligence.
Similar to the Turing test, the value of most computation is actually anthropically measured. A protein folding is performing a myriad of "calculations" from one computational anthropic perspective but merely following out some fairly simple rules from another.
google is an example of simple openended dynamic ontology. By not having a feedback mechanism other than end users etc. the page rank algorithm is tweaked "albeit by engineers" to provide answers. And now for something completely outher, maybe a nice Tool set connecting a probablistic Wiki and the Google API will produce what is being looked for. AFter all an ontology from a funtional (anthropic judgement) is only a recursive intellgent taxonomic simulation.
Apple managed IP flow between the Kaleida/Taligent ventures like a one-way valve ... we're were always kept secure in the storage tank well beyond arm's length for fear of inadvertently transferring Apple ATG IP into the "arrangement" as it was referred to in certain circles ... Absurd, but true. Where is Larry Tesler when you need him? :-)
Do you remember the MacIvory project with Symbolics? How about the "Logic Manager" MacToolbox addition (ALS Inc I believe)? Ever hear of "Roger"? E.T.? Marc Porat's 'Crystal' (All past ATG code names)...
BTW, who was that "On Technology, Inc." guy at Kaleida? Forgot his name ... Peter something(?)
I was the assistant CTO at Kaleida, where I dealt regularly with the folks at Apple and ATG (so it's funny we never crossed paths, perhaps you had already moved on), because I was the functionary through whom all IP was officially transferred between Kaleida and it's neglectful, absentee landlord parents (IBM and Apple).
And I did my time in the AI/non-procedural language cult of the late 80s. So, when I hear that some top researcher at Stanford has the goods -- she has to have the goods or she'd/they'd be laughed off the planet. Right? There's gotta be some there there.
Thus, I understand your scepticism, but suspect they have "something."
No specific APPLICATION examples cited for CYC aka Cycorp ... Geez whatta name, Cycorp! Isn't that what they call the villianous "mega-corp" in every C-rated schlock sci-fi epic?
BTW, at one point, it was nearly common knowledge that Doug was reading sci-fi almost exclusively at MCC ... and who would blame him? Stuck in humid Austin with all that "City Limits" type music ... hey, wait a munite I like that stuff ... some of it anyway ...
Back then, I was the only guy in Apple ATG (Advanced Technology Group) who ever actually USED Cyc on the Symbolics workstation that the department purchased at my request, and I found many dubious "demo-itis" pieces of hardwired code strewn about the distribution. So back at the Labs at Apple ATG, we were always a bit suspect of Cyc's claimed vs. actual mileage... after looking under the hood & driving it of course.
Sorry to say, but the long-standing, "inside-joke" tag line for Cyc was originally from Curly Howard,
Back in the mid 80s until the early 90s, I was the member representative on the MCC (Microelectronics and Computer Technology Corporation) corporate consortium for both at Eastman Kodak and Apple Research. Member companies paid a steep fee to partake in the "5th Generation Madness" that was rampant at the time. Not much was accomplished IMO but hey, Austin has great music and food ... albeit a bit muggy sometimes ... Ask Dr. Alan Kay for MANY more details ... please!
Since the 80's when CYC was a "ten year vision", the world has eagerly awaited the applied results of the CYC wonders promised ... like CYC being as smart as a 4 year old girl (Doug Lenat's daughter), or being able to bootstrap itself by reading the enCYClopedia (get the naming cute-ness?)...
Nuthing concrete yet has materialized as far as we can tell ... so how about details on some specific *active* projects, showing some/any promise, using that wonderful & hideously complex CYC ontology? :-)
Like Tang from the Space Program, CYC did indirectly spawn some useful progeny, but only through the work of highly dedicated individuals like R.V. Guha (and his RDF contribution, among others). Underpinnings, of course, for the grand vision/tower of Babel that born-again AI researchers are paddling toward these daze.
IT Facts
The Honorary Award for Peace and Prosperity goes to Symantec
Dr. Deborah McGuinness is associate director and senior research scientist of the Knowledge Systems Laboratory at Stanford University. Here she answers audience questions on her presentation of how the semantic web works.
E-NEWSLETTER SERIES