Agent Amplified Communication
Henry Kautz, Al Milewski, and Bart Selman
AT&T Bell Laboratories
Murray Hill, NJ 07974
fkautz, aem, email@example.com
We propose an agent-based framework for assisting and simplifying person-to-person communication for information gathering tasks. As an example, we focus on expertise location within a large organization. In our approach, the informal person-to-person networks that exist within an organization are used to ?referral chain? requests for expertise. User-agents are employed to automate the referral chaining process. We also present simulation results demonstrating the effectiveness of our approach.
There are basically two ways of finding something out by using a computer: ?ask a program? and ?ask a person?.
The first covers all ways of accessing information stored online, including the use of traditional database programs; file indexing and retrieval programs such as glimpse (Manber and Wu 1994) or Apple's AppleSearch; news filtering programs such as Hoover (SandPoint Corp.); and even more simply, the use of tools such as ftp, awk, and text editors to retrieve and view files.
The second, ?ask a person?, covers ways that a computer can be used as a communication medium between people. Currently the prime examples are electronic mail, including both personal e-mail and mailing lists, and bulletin boards and newsgroups. The growing integration of computers and telephones allows us to also view telephony as a computerbased communication medium. Simple examples of such integration are telephone address book programs that run on a personal or pocket computer and dial numbers for you; more sophisticated is the explosion in the use of computerbased FAX. Today it is hard to even buy a modem that does not have FAX capability, and by far the heaviest use of FAX is for person-to-person communication.
There are inherent problems with both general approaches to obtaining information. It has often been noted that as the world of online information sources expands, the ?ask a program? approach suffers from the problem of knowing where to look. For example, the Mosaic system overcomes many of the technical problems in accessing a wide variety of information on the Internet, by automatically handling
the low-level details of different communication protocols. It is easy and entertaining to browse through an enormous hypermedia space. However, finding an answer to a specific question using Mosaic tends to be slow and frustrating, and often results in failure. One response to this problem has been the attempt to design systems that incorporate knowledge about the location of information (Etzioni and Weld 1994; Kirk et al. 1995; Knoblock et al. 1994; Maes 1993). However, a deeper problem remains, that no solution based solely on building a better search-engine can address. This is the fact that much valuable information is simply not online, but only exists in people's heads. Furthermore, there are economic, social, and political reasons that much valuable information will never be made publicly accessible on the Internet or any other network. Indeed, part of the value of a piece of information resides in the degree to which it is not easily accessible.
This is perhaps most obvious in relationship to proprietary corporate information. For instance, if I am involved in trading General Motors stock, I may be vitally interested in knowing the specifications of the cars to be introduced next year. That such information exists is certain ? indeed, factories are already being set up to produce the vehicles ? but I am certainly not going to be able to find this information in any database to which I have access.
For a more mundane example (at least one of less concern to the SEC), suppose I need to have my house painted, and want to know if Acme Painters Inc. does good work. It is highly unlikely that I am going to be able to access a database of ?reliable housepainters?. Conceivably the Better Business Bureau might offer a service that could tell me whether many people have actually taken Acme to court, but that would hardly be all that I would want to know. Any recommendations offered by such a public service would have to be either advertisements or sufficiently innocuous to avoid legal entanglements. An even more telling case would be where I am trying to decide whether or not to hire a certain professor John Smith to do some consulting for me, and I want to know whether or not Smith knows what he is talking about. It is certainly not the case that I will be able to access a database of academics, with entries such as ?solid, careful thinker? or ?full of hot air?.