Cart
[Home] [By Thread] [By Date] [Recent Entries]
http://www.infotoday.com/searcher/jan00/feldman.htm Stereotyping can take place wherever you have a tool and the will to do it. Common stereotyping languages and topic maps probably have a lot in common. They are an opinion about a relationship one finds "usually true" so worthy of "the risk of maintaining a stereotype". Official stereotypes are a solution and a problem. No matter how you do this, it is risky. But the risk is similar to the risk of creating and using patterns in the real world. The SW can add a lot of weight to an official stereotype by reference, what we might call, "broadcast credibility" (said often and by enough people it has a high truth value). But that is dangerous. History has come up several times on this list today. How often has the history been wrong? What would be the consequence of taking action based on a false history or a true one? Some lose, some gain. A broadcast system can be gamed and behaviors based on using it can become superstitious. How many times have you seen the statement: "HTML is a subset of XML."? Dead wrong. Often repeated. Almost impossible to remove as an assertion in the system because it has been said too many times by too many "credible" authors in "credible" publications. The challenge is expungement of false assertions in very large replicated databases many of which you do not have update privileges over. We deal with this in the court systems all the time. The reasonable rule of thumb about a global semantic web is that if it is updateable and enough people/systems are using it, it will be corrected. That's the theory. That is also the risk. Right to correct is a contract. But CAN you correct it? Did the fact escape into the environment and spawn? Remember, this is a very large scale semantic web we are discussing. Understand Nabster is proving to be very hard to control now and it IS centralized by index. Suddenly those who have trumpeted the decentralized web are discovering that without a powerful navy, the oceans become pirate-infested. Caution: Mammals At Work. The semantic web must be considered a service for services, a layer of information over information which services can use to go about their business. If the quality of information and the concept mappings are good, then the service will be effective, but it cannot be the ultimate means to resolve disputes or to discover "truth". That is meaningless. If it gets you to the goal, it works, but you will be responsible for knowing when and whom your goals are in conflict. We are able to cope with a stereotyped world and we find that authority and legitimacy of source, how obtained and how maintained become crucial issues. These are not, as far as I know, automatable. We can train agents but insofar as they represent us, we are culpable for their actions. The only truth is immediate. The only justice individual. Unless you can negotiate and correct the Semantic Web, don't build it. It becomes a Golem. If you understand the legend, you understand the problem and the solution. Len http://www.mp3.com/LenBullard Ekam sat.h, Vipraah bahudhaa vadanti. Daamyata. Datta. Dayadhvam.h -----Original Message----- From: Danny Ayers [mailto:danny@p...] Sent: Thursday, May 10, 2001 11:06 AM To: xml-dev@l... Subject: RE: First Order Logic and Semantic Web RE: NPR, Godel, Semantic W eb I would question where the stereotyping is to take place, and who or what makes the assumptions - "when someone has a Ph.D., it is ok to assume that he is nearsighted". An assumption is made and then the reasoning is made using FOL - why so? If the data isn't clear cut, why use clear cut logic? It has oft been pointed out that a lot of the techniques for reasoning as discussed in the context of the semantic web have been around for decades (some for millennia). There have also been techniques around for inexact reasoning for a good while too. You can either (rather inefficiently) do number crunching directly on the data/documents or you can develop metrics and work with these. Let's not forget that in addition to the information about a document that might have been humanly-specified in Dublin Core or might be deduced from the location of the document, there is also the content of the document from which metadata can be mechanically extracted. There's a lot of data around. One of the common problems with e.g. neural networks is having large enough data sets with which to train them. The web is pretty big though. Another requirement is the computational power - the web's not really lacking there either. Let the machine make its own assumptions - it might even figure out what activity the good doctors indulged in to cause their near-sightedness.
|
