How would this work? You feel that the computers of the next 10, 20 years, in addition to having the rich interfaces we have, will have personal agents as well?

One of the ways I think about looking ahead into the future is to try and find analogies that might actually make some sense, and also to look for driving forces. One of the driving forces for the PARC-type user interface came just from there being inexpensive integrated circuits around. You start getting a proliferation of computers that are inexpensive enough for people to buy, and all of a sudden the kinds of people who might want to use computers changes completely. So all of a sudden you need a much easier-to-use user interface. There is a driving force now to do something, because it isn't just graduate students any more.

To me, the driving force for agents is pervasive networking, because the techniques used on the Macintosh don't work well when you're connected up to a trillion objects scattered all over the world. You need something looking for potential objects that will further your own goals. And you need that something to be looking 24 hours a day.

We think that what we'll have is 10, 15, 20 or more little agents, many of them not particularly intelligent, but able to flexibly take on a goal that we have. An example of one is an agent that goes out and finds you the newspaper you'd most like to read at breakfast every morning. All night long it works. It can touch dozens of different news sources, the Associated Press, New York Times and so forth, looking for things that are relevant to you. It can go to other sources for getting photographs and so forth. It can do the news gathering with a particular interest in the kinds of things that you have been involved in. A headline could say, "New fighting in Afghanistan," or it might say, "Your 3 o'clock meeting was cancelled today," because news now could involve your own electronic mail. The sidebar might say, "Your children slept well last night."

This is an interesting example of an agent, because it's one that was built about ten years ago. It did not require a large amount of intelligence in order to work. Its major strength was its ability to work 24 hours a day while you weren't there. With a limited ability of doing matching against what you said you wanted and what it thought you wanted, it could do a great deal of useful work for you.

There have been a number of revolutions in the history of computers so far, but most people think of the computer as a standalone desktop object, right? Tying this in with your Dynabook concept, what you've been saying about networking and agents, where do you see the next thing taking us?

The way I think about that is these three very different ways of relating the human to the computer. One is the institutional way of the time-sharing mainframe. One is the desktop way where you control all the stuff. The third way is the intimate way, which is the Dynabook way, which is continuously connected into the worldwide informational network.