Gordon Moore is most famous for coining Moore's Law, his 1965 prediction that the number of transistors that could be packed into an integrated circuit would double every year. A decade later, he revised that estimate to every two years-a prediction that has held remarkably true ever since and is often used as a baseline for evaluating performance in other spheres of computing. But the semiconductor pioneer and cofounder of Intel claims he has never really been good at predicting the future. In fact, he says he's pretty bad at it.

In a wide-ranging conversation from his vacation home on Hawaii's big island, this 72-year-old avid fisherman and pragmatic environmentalist spoke with Technology Review editor at large Robert Buderi about the future of Moore's Law, which he's revising again. He also discussed the newly created Gordon E. and Betty I. Moore Foundation, which he plans to endow with $5 billion worth of Intel stock (about half of his estimated wealth). The new foundation will be geared to supporting far-out university research and advancing programs that protect the environment.

Finally, Moore reveals why he and a few pals named Hewlett and Packard picked up the tab for the Search for Extraterrestrial Intelligence-and shares how it's possible to succeed wildly in business without having a crystal ball.

TR: Where do you think we are heading technologically? What excites you most?
Moore: I calibrate my ability to predict the future by saying that in 1980 if you'd asked me about the most important applications of the microprocessor, I probably would have missed the PC. In 1990 I would have missed the Internet. So here we are just past 2000, I'm probably missing something very important.
Certainly the things going on in molecular biology these days are exciting. We really are understanding increasingly how all the life processes work. In the information sciences, the increasing capability of networks is changing the way we do everything. It's going to result in all of us having a lot of bandwidth whenever we want it, to go with a lot of computing power. As far as what are we going to do with all the computing power, I'm sure I'll miss the most important things. But the one capability that to me will make a qualitative difference in how we do things is truly good speech recognition. That is, a machine that can recognize if you mean t-o, or t-w-o, or t-o-o by understanding in context what you're saying.