Network: Never mind where we want to go today, where will you take us tomorrow?

It could be argued that Microsoft's success is the result of brilliant marketing rather than innovative technology. But to maintain its dominant position in the industry, the company is investing $2bn a year in research. When Bill Gates looks to the future, he looks to Nathan Myhrvold. Paul Smith talks to him.

It is commonly thought that the man who holds the vision for the future of computing is William H Gates III. The Microsoft chairman controls the destiny of an industry that underpins so much of what is important now and will be into the next century. Not just our PCs, but our banking, our entertainment, our education - all of our information.

But if you really want to know what computers are going to be like - what they will and won't be able to do, how we are going to use them - then the person you need to talk to is Nathan Myhrvold. The chief technology officer and vice-president in charge of research knows more about what the future holds for computing than anyone else at Microsoft. The reason is simple: he's the one building that future.

On a first meeting, you certainly would not credit him with the status of uber-geek. If anything, he reminds one of an Ewok character from Revenge of the Jedi, a jolly, avuncular teddy bear. And he displays a conversational range of interests that stretches far beyond the single-minded obsession of the true anorak.

He boasts more than a casual knowledge of palaeontology, and an interest in philosophy, cosmology and photography. He collects fossils and books. He sky-dives, bungee-jumps and climbs mountains. He is also an accomplished chef, cooking - when he can - at Rover's, one of Seattle's best restaurants. He earned an advanced culinary degree from La Varenne in Burgundy, and competed twice at the world barbecue championship in Texas, winning first and second place awards. When he came to London to announce Microsoft's $50m investment in a research facility in Cambridge, he hosted a lunch at Chez Nico, insisting on arriving half-an-hour early to spend it with the chef.

Myhrvold finished high school at 14, and has a Master's in maths from UCLA and another from Princeton, where he also got his PhD. He was offered a post-doctoral fellowship at Cambridge by Stephen Hawking, researching cosmology, quantum field theory in curved space time and quantum theories of gravitation. A colleague is quoted as saying, "He is one of the few people who can work on a peer level with Bill [Gates]."

His role at Microsoft, which he joined in 1986 as director of special projects, has always reflected this gigantic, roving intellect. He is essentially hired to think: about the future of Microsoft, and the future of technology.

At first, he spent a lot of time considering the forces affecting Microsoft and how they should be handled. Usually, people think of its competitors as being other computer companies - Netscape, Intel, possibly the US Department of Justice. In reality, it is more likely to be the likes of Time Warner or Disney.

Of course, being bright does not always make you right. Myhrvold was the product manager of Windows 2.0, a product so dire that not even his formidable talents could save it. And he is widely credited as the man who steered Microsoft away from the Internet when the rest of industry saw it as a watershed. That decision nearly caused the company to founder, and it was only a turnaround of admirable nimbleness that helped Microsoft to regain its dominant position.

All has been forgiven. In October 1996, Myhrvold's role was crystallised when he was given responsibility for the $2bn annual investment in research and development at Microsoft. He and his colleagues have set up an academic- style research lab, complete with an open, and highly counter-cultural, approach to research. Any work that they are doing is available on the Net or presented at conferences. Everything is subject to peer review. And they are garnering kudos, both from other academics and also, more significantly, in the people whom Microsoft is hiring, some of the foremost computer scientists in the field.

We spoke at length about the research going on at Microsoft, both the specifics and the nature of technology research itself. The first question I wanted an answer to was this: what will computers look like in two, five and 10 years from now?

"Well, two years from now is pretty easy, because it's an evolution of what we have now, and a bunch of things get better: the cost gets lower, the machine gets faster.

"Once you get out to five and 10 years, you start to get more things out of research; more things are unexpected. Although we have a notion of what it will be like, it's not very exact. Let me give you an example. If you look at the whole history of Windows, version 1.0 shipped in 1985, version 3.0, the first successful version, shipped in 1990. So that is an example of something that was a product for five years before it even caught on.

"Although we like to think of technology as fast-moving, and in some senses it is, often it takes a lot longer than you'd expect for things to be mainstream.

"I would say speech recognition is at least five years away as a mainstream thing. There are speech recognition products today that will work under certain circumstances. I know people who have carpal tunnel syndrome and can't type who use them. But, as a mainstream thing, probably five years away - plus or minus three!

"Some degree of vision is probably only a few years [away]. But it depends what you want. If you mean computer vision, where we will stop driving our own cars and that they'll drive for us, that is, I'd say, 10 or 15 years out, just for the wealth of social factors. Using it for things like sensing who's in the room, figuring out what's going on, if your attention is directed to the machine or not directed to the machine, those things are all getting very possible ... I think within two to three years, you'll start seeing things like that. Within five to eight years, it will be quite common."

These areas - speech recognition and vision control - are two of the main areas of research at Microsoft. They are obvious areas and, indeed, there are many solutions available today. In fact, a senior researcher working on vision control admits that Microsoft is not leading the field: "You could have seen a much better demo than ours at MIT three years ago."

One of the problems with any technology research is choosing which fields to enter in the first place: it is hard to imagine how a new technology will be used. After all, we have only the experience of existing technologies to base it on. How do you look forward when all you can do with accuracy is look back?

"First of all, it's very hard; there are no great answers. But the way you try to make progress is the following: the thing that determines how we use technology is not the technology itself, but us. It is aspects of human nature, and the geometry of our bodies.

"PCs are built the way they are because of the length of our arms, fundamentally. If you are going to do detailed work, write something, our eyes focus best at arm's length. Our eyes focusing best on our toes makes no sense. So if you have a device that is going to be used for reading or writing in very detailed fashion, it needs to be that far away from your face; if you use it for long periods of time, you are going to sit down. And, if you work it out, you've pretty much described the personal computer from a human factors standpoint.

"Culturally, we have patterns in how we adopt technology and what we tend to use technology for. And so really this study of how we use technology in the future is to understand the things that count, which is humans and human nature. It's not exact, but you can get a surprising way by reasoning on the basis of what people are likely to do."

A combination of thinking about human factors, and a formidable intellectual curiosity, has led Myhrvold down some strange paths. "I did a calculation once to try to determine what the bandwidth of taste receptors was. When you taste something, how much data is coming out of there? Is it more or less than video? ... It's about the same as audio.

"People have done studies where they dilute sugar samples to see how many gradations you can get, which will tell you how many bits you have. Now it turns out you have about eight bits of resolution. If I dilute something by one 256th, a trained taster can just taste it, but one 500th and you can't tell. Then you've got a bunch of combinations of flavours; it's not just one or even three mixtures.

"It's a little more complex than colour. With colour, you have three samples. So it's roughly eight bits per sample. The key thing is that taste is slow: it takes time for taste to develop in your mouth. A CD takes a 16-bit sample 44,000 times per second. If, in fact, you take half a dozen eight-bit samples 10 or 20 times a second, let's say 100 times a second, you'd probably have enough. And so it is about the same as CD audio."

It is this sort of left-field thinking that drives much of Microsoft research.

"We try to do a few things here that are not obvious. For example, we've just started an ambitious project in theoretical computer science. Now, most industrial research labs don't do a lot of theory. Even a lot of universities don't do a lot of theory.

"We hired a bunch of the world's best mathematicians to use some of the sophisticated techniques from mathematics to rethink the fundamentals of computer science. This is highly non-obvious, in that no computer science head of department would have done that. In the long run, this could lead to a different understanding of computer science from a theoretical perspective, which would be very interesting. The other interesting thing is that it might, just might, lead to new models of computation.

"There are people already experimenting with radically different models. Quantum computing is an example, and so is DNA computing. Our brain clearly is another one. So we know of at least one other model of computation, which is in our heads. And we know that we can do things very easily that today's computers cannot, and vice versa, so one of the most interesting things would be if we could invent new models of computation. So [theoretical research] has a low probability of reaching its goal but, boy, it sure would be valuable if it did."

Microsoft research, it turns out, is not about the future of computers so much as the future of human beings. And the linchpin is Microsoft's Renaissance Man, seeking to know what we are all about. "Understanding technology itself isn't a key thing. The fact is, we are insatiable in our desires. So it does help to understand other things."