Toward a society of interacting computers.

The talking head on the computer screen can smile, frown, or look perlexed. When you ask it a question, it barely pauses before responding with an appropriate facial expression and a suitable comment. And it's all done with software and two synchronized workstations.

Developed at the Sony Computer Science Laboratory in Tokyo, this system combines speech recognition, language understanding, and voice synthesis with a graphics capability that models the way muscles contort a face to create certain expressions. It also represents a step toward what Mario Tokoro, director of the Sony laboratory, calls the "intimate" computer.

Tokoro envisions the development of computers that can chat with their owners, offer advice, joke around, provide moral support, and so on. Although such versatile, friendly digital assistants and companions are not yet available, he contends that some of technologies required for such machines are already being investigated and developed.

"I would like to have a computer buddy," Tokoro says. "This would give us a new, different world."

Tokoro, who is also a professor at Keio University in Yokohama, offered these and other speculations about the future of competing in a presentation at the Association for Computing Machinery's conference on object-oriented programming systems, languages, and applications, held last week in Washington, D.C.

The last few years have seen tremendous growth in object-oriented computer programming. Roughly speaking, this technique involves writing computer programs in special kinds of chunks, known as "objects." Like physical objects in the real world, these software objects in the other objects according to well-defined rules, or laws. In other words, these capsules of digitally encoded information--including data, files, and directories--can "communicate" with each other in only certain ways.

Thus, with the object-oriented approach, programmers no longer have to worry about the details fo what's inside a particular object once it's created. Instead, they can focus on how one object interacts with another to accomplish a specific goal. This contrasts with the more traditional type of programming, in which the software author produces a list of instructions--a receipe--specifying in precise detail exactly how to proceed from beginning to end to achieve the desired result.

Tokoro has investigated a class of objects that also includes processors capable of executing commands. He likens these "concurrent" objects to the cells that make up a biological organism. Although the cells work together, each one has its own biochemical machinery and individual life. Similarly, a collection of concurrent objects could work together as an "autonomous agent", Tokoro says.

Each agent would respond to stimuli, have the ability to learn about its situation, be able to make decisions in a timely manner, and develop its own goals. All such agents acting together would constitute a "society," in which collective behaviors that cannot be ascribed to individual objects could emerge.

Tokoro's "intimate" computer represents an example of an autonomous agent--one that has a human face, understands language, and inspires respect and camaraderie. The "talking head" system is a first step in this direction.

At present, the system handles questions and comments (in Japanese) concerning Sony's line of computer products. Researchers are now looking into enlarging this domain, incorporating models of how people interact socially and adding such capabilities as reading a system user's facial expression.

With further development, such a system may in the short term evolve into an extraordinarily friendly instruction manual. Anyone need advice on how to program a VCR?