My name is John Flynn and I am the Program Manager for the DARPA Agent Markup Language integration effort. The DARPA DAML Program Manager is Dr. Jim Hendler. This presentation is intended to provide an overview of some of the issues related to technologies that will support the future use of agents in the World Wide Web environment.

I would like to start by looking ahead a few years and examining what kinds of advances in technology we can realistically expect to see. A reasonable starting place is Joint Vision 2020. It provides some very well-thought-out ideas about the future requirements in the area of information technology and the need to achieve information dominance. However, it isn’t a recipe for how we are going to achieve information dominance and it doesn’t provide predictions of specific technologies that will get us there. We generally have a tendency to overestimate future technology progression. I remember when the movie 2001 came out in 1968 it didn’t seem totally unreasonable that within 30 years we might have spacecraft, and even computers like HAL, as depicted in the movie. Of course, now that we are in 2001 we can clearly see that technology has not advanced nearly so far.

One way to try to bound what we can expect from future technologies is to look back over a comparable period of time. For example, let’s take a person from 1980 and magically project that person to today. She would be able to get right in any automobile and drive right off. She could go to the airport and board a flight and not be particularly surprised at anything that she saw. She could turn on the TV and see new shows and characters but it would pretty much be the same old garbage she saw in 1980. The point is that over a span of twenty years a lot of things just don’t change in fundamentally significant ways. The improvements are definitely occurring but happen very gradually. One way to visualize this is as a series of technology rockets that don’t fly too high and don’t shine especially bright but they continue to lead us forward. However, once in a while a technology rocket is fired off that really is bigger and brighter. Even though the seeds of today’s Internet were solidly in place in 1980 through the ARPANet effort, our visitor from 1980 would probably be truly amazed if someone showed her how they use email on a daily basis. If you then opened your web browser and let this 1980 person surf the web for a while I think she would simply find it hard to believe the depth and breath of information from all over the world available at her fingertips. Although XML is still in the process of solidifying as a true force on the Internet, I am confident that it represents another of these technologies that will significantly change the way we work with information in the future. Finally, an emerging concept, called the Semantic Web, is poised to create a truly major leap forward that will forever change  how we interact with the web of the future.

Ray Kurzweil has a pretty solid reputation in the world of information technology. Here is what he thinks the world of the future will look like. As you read his projections, let’s jump even farther into the world of the future. Not so much as a prediction, but just to try to set the upper bounds on this concept of the Semantic Web. A concept popular in science fiction is that of the avatar. An avatar is basically your double in cyberspace. If you can imagine a far-future capability to electronically duplicate your mind and link it to the Internet. If this was a complete and true duplication, you would potentially be able to live forever in cyberspace and be able to interact with other avatars just as you might interact with others across the Internet today. If that may ever actually be possible is certainly a valid question, but it does capture the concept of a truly intelligent cyber agent.

So, what is an agent? It is just a software program that is designed to do some job for you. In this context we are talking about software programs that interact with the Web. They can be simple or very complex. They may be tightly controlled, having to report back to you before taking any action, such as purchasing something. Or, they may be quite autonomous, such as an agent you have authorized to search for the best price and buy you the latest Garth Brooks CD.

This concept of cyber agents is very important to the future of the Web. We have made great improvements in decision support systems that help us do our work easier, faster and better. But as long as we remain directly in the loop there are some fundamental limitations to how much more these system can help us. In order to make orders of magnitude improvements we need to be able to clone in software at least some of the functions we would normally perform ourselves.

There is another factor in the evolution of the World Wide Web that is driving the advancement of agent technology. Initially the main function of the Web was to provide access to information and that is still very important, but there is also an increasing capability for the Web to provide services. There are now many things you can search for and buy via Web services. A next step in this direction is the development of capabilities to provide more complex services. You might soon be able type in the basic parameters of your trip to this conference and have a software agent, which also knows your travel preferences and budget, put your entire travel arrangements together - including airplane, hotel, rental car reservations based on your organizational policies and personal preferences.

There are some fundamental problems that impact how effectively we can use the Web today. We can only search a relatively small percentage of the information that is actually available. Even then, we get back, and have to sift through, a lot of junk that isn’t what we wanted. And, these problems are only going to get worse as the total amount of information on the Web continues to grow.

What is keeping us from developing and using software agents today? HTML is wonderful. It has carried us a long way as the Web has grown and it will continue to serve us well. However, HTML markup is only designed to control how information appears in our Web browsers. It tells the Web browser if the information is to be displayed large or small, what color it will appear, and whether is will be in a frame or table. It doesn’t tell us anything at all about the meaning of the information. That is strictly up to us as humans to interpret. The fundamental problem is that the great bulk of the information on the Web is not machine readable and that must change before real progress can be made in the development of cyber agents.

XML is coming to the rescue. The Extensible Markup Language is a World Wide Web Consortium (W3C) recommendation, finalized in 1998, that has been widely adopted around the world. By now most of you have been exposed to XML either through actual projects your organizations are undertaking or through pressure you are probably receiving as to why your organization hasn’t started the move to XML. You can see on the slide the example of how an XML tag is used to specify a social security number. A software agent can read the XML tag and know unambiguously that it is dealing with a social security number and not a phone number, a part number, or any other kind of a number. There are a large number of communities of interest that are developing their own XML dialects. An XML dialect is basically just an agreement within a general area on the XML terms that will be shared across that community. XML is great and many applications are being developed to take advantage of this new lingua franca of the web. It does have a limited capability to express complex relationships between information objects. XML only supports “contained in” or “parent-child” relationships, such as an <address> may contain sub-elements of <street number>, <street name>, <city>, <state>, and <zip code>.

A relatively new initiative is the DARPA Agent Markup Language (DAML) effort. It leverages XML and another W3C recommendation called the Resource Description Framework or RDF. The current release of DAML focuses on the creation of ontologies. The word ontology seems to be intimidating to some people, possibly because of its long historical association with philosophy and a lot of pretty heady discourse on the concept of existence.  Ontologies, as used here, are the concepts and relationships that describe sets of information. DAML provides a formal, machine-readable language for describing these relationships in a way that agents can work with. We talked earlier about the concept of the Semantic Web where information will be represented in machine readable form and agents will be able to do many tasks to increase our productivity. The W3C is vigorously pursuing the definition of the language recommendation that will be used to implement the Semantic Web and DAML is an active participant in that process. Although useable in its current release, DAML is still under development and will be adding additional features such as logic and rules constructs to give agents additional power. A limitation in using DAML today is that the tools needed to make it easy to use, such as high-level markup editors and DAML-enhanced search engines, are still emerging.

This is a very simple example of using HTML, XML and DAML to represent information. The HTML markup provides the information in a form that you could read via your Web browser and you would understand perfectly well that the person is an Air Force Captain. The XML representation uses element tags to make each piece of information explicit so that a software application could recognize that the word “Captain” was a rank. In the DAML markup you can see that the rank is identified through reference to a URI at http://www.af.mil/personnel. That URI would contain DAML markup representing an ontology of Air Force ranks and how they relate to one another. If the DAML markup was describing someone in the Navy, it would reference a URI at http://www.navy.mil/personnel and the relative position of Captain in the Navy rank ontology would be quite different.

The DAML effort has been underway for less that a year. There is a broad base of world-class research teams working on various aspects of DAML. A total of 130 ontologies and over 2.2 million DAML statements have been created so far. Also, DAML is being developed in close collaboration with researchers in the European Union. DAML is being used as a key technology in the Horus project in support of the Intelink Management Office. The project is closely aligned with the W3C, which has several members actively working on the DAML effort. Please visit our web site at www.daml.org to see more details on the effort and join any of several mailing lists for different aspects of the DAML work. You can also subscribe to the HotDAML Newsletter that periodically provides information on new DAML developments. The May issue of Scientific American contains very informative article on the Semantic Web which is also available via the daml.org website.

Despite years of research in agent technologies, software agents are still not in widespread use. A major problem has been the lack of infrastructure and significant amounts of machine-readable information to allow agents to accomplish things meaningful to most of us. The advent of the World Wide Web, XML and now DAML provides the capability to markup very large amounts of information and the development of the agents to exploit that markup will follow close behind. It will become compelling to markup your information in a machine-readable format. If one company has their information available for use by agents and another company does not, the first company will have a significant competitive advantage. The same situation will exist for military Web sites that are trying to get important information out to key subscribers. Once the ball starts rolling and the tools become available to make it easy to use machine-readable markup, the transformation into the Semantic Web will begin to accelerate rapidly.

Agents, enabled by DAML and the Semantic Web, will fundamentally change the way we interact with the Web in the future.