Rethinking Artificial Intelligence


In September, 1997, the MIT Artificial Intelligence Laboratory and the MIT Industrial Liaison Program sponsored a briefing for senior technical management and corporate strategists on the future business impact of accumulating Artificial Intelligence technology. This document is a summary of the most salient points, as seen from the perspective of the briefing chair, Patrick H. Winston.

The Big Messages


Session 1: The AI Business: Past, Present, and Future

During the 80s, AI experts developed systems for solving problems ranging from chemical-plant optimization to oil-well log analysis. Some of these systems were spectacular successes, with payback measured in hours. But in spite of such early successes, Esther Dyson, editor of the influential trade publication Release 1.0, predicted that AI would not become truly important commercially until AI became embedded in main-stream, strategically important systems like raisins in a loaf of raisin bread.

Time has proven Dyson's prediction correct. In the 90s, AI, as a field, is becoming more important as emphasis shifts away from replacing expensive human experts with stand-alone expert systems toward main-stream computing systems that create strategic advantage. Accordingly, many of today's AI systems are connected to large data bases, they deal with legacy data, they talk to networks, they handle noise and data corruption with style and grace, they are implemented in popular languages, and they run on standard operating systems. Moreover, human users usually are important contributors to the total solution.

In this session, the speakers explained how the shifted emphasis has put AI to work in such industries as defense, transportation, manufacturing, and entertainment.

Patrick Winston
Former Director, AI Laboratory and Ascent Technology

What I learned about business after I thought I knew everything.

Philip Brou
Ascent Technology
Transportation Industry

Situation assessment, allocating resources, and the application that a former DARPA director said "justified much of the money spent on AI research."

Joseph Mundy
General Electric
Manufacturing Industry

How computer vision is developed and applied for a broad range of applications and what constitutes a viable application.

Eric Horvitz

Role of AI in the future of user-friendly software

Ted Dintersmith
Charles River Ventures
Venture Capital Industry

The AI Business as seen from today's venture capital community.

Session 2: Information Access and Presentation: Making People Smarter

Because better information means better decisions, decision makers naturally want access to large quantities of information expressed in diverse forms. But the world wide web, new sensor technology, and other information sources have combined to create quantity and diversity, such that it has become increasingly difficult to provide decision makers with the right information, at the right time, in the right quantity, in the right form.

Of course, practical difficulty means research opportunity, and many of today's AI research efforts focus on the development of systems that anticipate information needs, find needed information, distill needed information appropriately, and display distilled information in new ways. Some such systems help decision makers locate and query information sources---human or computer---via the world-wide web. Other systems distill tidal-waves of information into simple presentations that engage human problem-solving capabilities. Still other systems bring the computers into our world, so that we can dispense with the small screens, awkward keyboards, and distracting mice that the computers of today insist we use.

In this session, the speakers explained how AI is shaping the future via AI enabled information access, AI enabled human-computer interaction, and AI driven advances in interface infrastructure.

Eric Grimson
The Enhanced Reality Project: X-Ray Vision for Surgeons

How vision research led to a system that enables surgeons to do their work in 1/3 less time, with great benefit to the patient, and to undertake operations that would have been too risky otherwise.

Thomas Knight
MIT and various spinoffs
The Human-Computer Interaction Project

Creating the Infrastructure: Wall-sized displays.

Boris Katz
The START Project: Information access via natural language and the www

How hundreds of thousands of Web users access text, pictures, images, maps, tables, video, and everything else.

Ramanathan Guha

Role of AI in the future of information access.

Marc Raibert
Formerly MIT and now Boston Dynamics

Simulation systems that enable medical students and doctors to feel what it is like to suture vessels and learn new procedures.

Howard Shrobe
The Intelligent Room: a basis for intelligent collaborative problem solving

Session 3: Beyond Expert Systems: Making Computers Smart Enough

The rise of rule-based expert systems in the 1980s was predicated on the idea that computers could do what human experts could do, only less expensively. Those interested in trying out the then-new expert-system technology were told that the first problem tackled should be doable by a person in more than an hour and less than a week.

Today, the emphasis is not on doing what people do. Instead, the emphasis is on exploiting opportunities to do tasks that people cannot do alone.

In this session, the speakers provide examples of how AI can, in fact, do what people cannot do alone, visiting a variety of applications, with goals that include digging regularity out of data in the search for new pharmaceuticals, capturing design rationale and exploiting captured rationale to improve product design, and working through tediously complex calculations to better guess what a computer user needs to know.

Rodney Brooks
Director, AI Laboratory and IS Robotics

Building robots: From theories of intelligence to cleaning up land mines and exploring the surface of Mars.

David Waltz
NEC Research Institute
Database Mining: Experiences and Potential

Why database mining is important, and experiences in introducing datamining technologies.

Randall Davis, MIT and various spinoffs

Rationale capture: preserving the thought as well as the conclusion.

David Kirsh
University of California, San Diego

Knowledge Management: Maximizing Intellectual Capital

  • Best practices transfer well only when properly supported by help, training and experienced people.
  • Technologies of collaboration are emerging -- telepresence, interactive learning environments, and virtual collaboratories.
  • Reward examples of information sharing, else information hoarding is rational.
  • To design workable knowledge management system's, make sure HCI design is sensitive to work practice behavior.
  • David Barrett
    Director, Walt Disney Imagineering

    The role of entertainment as an information-technology driver.