Issue

Article

Vol.30 No.2, April 1998

Article

Issue

An Overview of the ATG Intelligent Systems Program

References
Acknowledgments
About the Author
Author's Address

The potential of intelligent user interfaces have been obvious for many years one only has to look at the Knowledge Navigator video or any of a large number of science fiction novels or films to appreciate what it could be like to interact with computers in ways analogous to how people interact with other people. However, the computing industry, throughout this time, has remained incapable of building systems with these capabilities. The technology demands are great, as are the human interaction design problems that must ultimately be solved to yield a system that is truly useful to people.

Consequentially, the Intelligent Systems Program chose a different approach: that of partial understanding. Since the dream of truly intelligent interaction lies beyond our capabilities, we worked to approximate this dream with technologies that were available to us. This led to a different approach to the intelligent interface question and to our work: we thought less about problems that have resisted solution for decades, and more about the real needs that users have and how they can best be satisfied. The tools of the intelligent interfaces community -- parsers, inference engines, knowledge representation, language understanding -- remained central our work, lurking beneath the surface. However, they were applied in ways that work today, and that yield successful solutions to user needs. One of these solutions -- Apple Data Detectors (1) [5] -- is shipping as an Apple product today; we hope and expect that other solutions based on this approach will join it soon.

Over the lifetime of the program, we applied this general strategy of partial understanding to several problem areas. These addressed different questions about user tasks and interaction styles, and presented different opportunities for technologies to help solve those problems. Several of the papers in this volume describe our experiences with this approach; in particular:

• Apple Data Detectors and beyond: Our work on Apple Data Detectors was based, in part, on a desire to move towards a broader notion of what constitutes a document, and how documents can be transformed from today's simple streams of characters into highly interactive computational objects. Our work on LiveDoc [4] and Drop Zones [3] outline increasingly powerful systems that take us in this direction by finding meaningful components of documents and making them the locus of task-based interaction.
  
• Language-based knowledge mining. Perhaps, at some time in the future, computers will be able to derive the same kinds of understandings of written documents that humans can. But what do we do until then? One approach, which we have pursued in our work on natural language processing, tries to find a middle ground between these statistical techniques and AI-based understanding techniques. The idea here is to use a low-level linguistic analysis of the document to identify terms from the document that are probably central to that document's meaning. It is then possible to use these terms to reconstitute how the terms are interrelated and, from there, some of the document's high-level structure. The paper on SCOOP [2] describes this set of language technologies and some sample applications of them; the paper on dynamic document presentation [1] shows how this approach can be used as a central part of an innovative interactive system.
  
• Networked communities and distance education: Finding the content of documents does us little good if we can't communicate that content to the people who need it. Hence, we also studied how communities grow up around bodies of information, and what technologies can insure the successful dissemination of that information to the members of that community. One experience with such a community is described in [6], both in terms of the technologies we found to be useful for providing community support, and in how the participants in the community felt about these kinds of interaction and collaboration.
  

It's important to be clear about the perspective we chose for this work. This was not a rejection of artificial intelligence as a source of useful technologies. It was, however, a fairly explicit rejection of the belief that we could count on the imminent solution of "the AI problem" -- that, suddenly, systems with the near-human capabilities envisioned for so long would suddenly become available. But this was not so much pessimism as pragmatism and practicality. We were not working to find great new applications of AI technologies, but to identify problems in peoples' everyday lives and to propose solutions for them. If we chose to let go of the full-bore AI dream, we were perhaps choosing instead the freedom that comes from making use of the technologies and strategies that we have today, and that can be applied to real problems with immediate as well as long-term payoff. And, if the AI problem gets solved next week, there are plenty of ways to fit them into what you'll read about here....

References

1. Boguraev, B. & Bellamy, R. (1998). Dynamic document presentation. SIGCHI Bulletin, this volume.

2. Boguraev, B., Kennedy, C., & Brawer, S. (1998). An architecture for content analysis of documents and its use in information and knowledge management tasks. SIGCHI Bulletin, this volume.

3. Bonura, T., & Miller, J. R. (1998). Drop Zones: An extension to LiveDoc. SIGCHI Bulletin, this volume.

4. Miller, J. R., & Bonura, T (1998). From documents to objects: An overview of LiveDoc. SIGCHI Bulletin, this volume.

5. Nardi, B. A., Miller, J. R., & Wright, D. J. (1998). Collaborative, programmable intelligent agents. Communications of the ACM, Vol. 41, No. 3 March,1998.

6. Nardi, B. A., Reilly, B., & Steinbeck, R. (1998). An online digital photography course for high school teachers. SIGCHI Bulletin, this volume.

Acknowledgments

As ATG came to an end, I was joined in the Intelligent Systems Program by Bran Boguraev, Tom Bonura, Eric Hulteen, Bonnie Nardi, and Dave Wright. Thanks are also due to our colleagues from previous years, including Yu-Ying Chow, Laile Di Silvestro, Dan Russell, and Bob Strong, as well as a number of contractors and interns, including (and with apologies to those I may have forgotten) Ken Anderson, Elizabeth Bratt, Sascha Brawer, Jon Doyel, Bruce Horn, Chris Kennedy, Henry Lieberman, Pattie Maes, Dave McDonald, Max Metral, Justina Ohaeri, Jeremy Sandmel, Shilpa Shulka, Oliver Steele, Heinrich Schwarz, Jason Swartz, Marc Verhagen, Yin Yin Wong, and Dave Yost.

About the Author

Jim Miller, until recently, was the program manager for Intelligent Systems in Apple's Advanced Technology Group. He is currently exploring consumer applications of Internet technology as part of Miramontes Computing.

Author's Address

Jim Miller
Miramontes Computing
828 Sladky Avenue
Mountain View CA 94040, USA
email: jmiller@millerclan.com
Tel: +1-650-967-2102

Footnotes

(1)

Downloadable at http://applescript.apple.com/data_detectors

Issue

Article

Vol.30 No.2, April 1998

Article

Issue