Atos Origin blog
Alison Booth>>The reality of a fully-fledged Semantic Web just seems beyond reach
I've been following the Semantic Web for a number of years now. The vision portrays an intelligent Web that understands meaning and can further infer new knowledge.
Recently we have seen the first step - Semantic Search. But have you been impressed by Twine or Wolfram Alpha? Probably not.
The Semantic Web aims to build on today's Web by giving applications the ability to find the meaning of a piece of information and how it related to other pieces of information. Using well-defined rules and logic, intelligent applications and agents will be able to infer further information from these linked pieces of information, thus mining knowledge. They will have to be able to prove their findings when required, using a trust framework to establish the validity of information.
It would be able to differentiate between Turkey and a turkey, to understand that two people are related if they are father and son and that Brighton is in East Sussex... or do you mean the 'Brighton' in Tasmania, Ontario, Illinois, Michigan, Iowa,...?
So why is it so important?
The vision portrays the Semantic Web as dramatically changing business's ability to turn their information into knowledge by enabling information from many diverse sources, including Web pages, databases and legacy systems, to be brought together and mined, producing further, deeper inferred knowledge. There would be applications for it in data integration, information searching, content management, customer relationship management, business activity monitoring and much more. Beyond information, the Semantic Web would also enable services to be advertised, found and made use of automatically.
Consider the following scenario as an example...
Like most companies and most individuals, both Jane and her insurance company use intelligent software agents to carry out mundane tasks. These agents are able to work together over the Semantic Web, negotiating between them in order to fulfil their individual objectives. They are autonomous software entities that are able to work asynchronously, each optimised for a specific purpose. The Semantic Web provides them with meaningful information, from which they can infer the knowledge they need in order to carry out their tasks.
One of the insurance company's intelligent software agents is allocated to manage the claim that Jane's personal virtual assistant agent has just submitted on her behalf. The insurance agent proactively ensures it is validated. It sends the relevant information to various fraud agencies, using the Semantic Web to ensure the consistent meaning of terms, and on receiving positive feedback, the agent approves her claim.
The agent next starts negotiations with the third party's insurance company's agent. The Semantic Web enables this machine-2-machine interaction to take place, ensuring that each term that is discussed in the negotiations is understood both agents to mean the same thing. For example driver, passenger, address, bumper, dent, ... are just some of the terms that would not only need to be understood. And not just understood - the agents need to understand how these terms relate to each other in order to infer further knowledge - the driver drives the car. Address is a location where someone lives...
Overnight, the insurance agent works with Jane's own personal agent, and an agent from the insurance company's approved repairers to book the car in for repairs. Using the Semantic Web, the diary and location information from Jane's personal agent, and a work schedule from the body shop, they are able to work together to negotiate several possible dates. The Semantic Web would enable the agents to communicate in the same language, ensuring that each understands the meaning of any term discussed and how it relates to other things. Additionally it would enable the agents to infer possible dates from the diary and work schedule through their understanding of the meaning of events in each and the implication of what each event entails.
In the morning, Jane's personal agent gives her three different options for repair dates, but Jane is not happy with any of the solutions - the body shop does not use original manufacturer's parts. Jane sets her agent on its way again. Using a Semantic Search, the agent is able to quickly find an alternative suitable, local body shop ...
.... and the story continues.
How does it work?
Without going deeply technical, the Semantic Web makes use of ontologies to provide this functionality. An ontology is a machine readable definition of an object and how it relates to other objects that can be used to infer further knowledge. For example, an ontology about fathers could define fathers as men with children. Another ontology about George could say that George is a man and that George has two children. From this, an intelligent application or agent could infer that George is a father.
Thus, in order for it to be successful, the Semantic Web would needs vast amounts of information telling it about objects and how they relate to each other.
Not enough information?
The problem today is that there just isn't enough of this information available for it to work well enough. A bit of a chicken and egg really - you won't see it's full potential until enough information is available, but less focus will go into generating the information if the rewards cannot be seen.
What we have at the moment is akin to inventing a new vehicle, but having no sensible way of creating the fuel needed to run it. The concept is great, but the means of generating and maintaining the information required is flawed.
How about generating it automatically
The only way forward, as far as I can see, is for the ontologies to be generated automatically. Could Wikipedia and online dictionaries be turned into ontologies?
Still this would not be sufficient. Not only is our language very complex and ever changing, but dictionaries and wikis describes objects well, but not the relationships between them.
We are simply not in a position to create all of the information that is needed - automatically or otherwise - and keep it up to date. And who is to stop duplication or who would say which version is right or wrong?
The reality of a fully fledged Semantic Web just seems beyond reach.
What for the future?
It is my belief that the Semantic Web needs to evolve into something very different if it is to become a reality. These changes may well be so radical that they take the Semantic Web to a place that is far from the original vision.
I do believe that something will come out of this - but exactly what remains to be seen.....