Roger Needham, Director of Microsoft Research Europe
Microsoft Research - What and Why
When Microsoft was not so large as it is now, it could get its technical agenda in part by looking at things that people were doing with different sorts of computers (not PCs), see what was worth doing, and say: yes, let’s do that with PCs. A dozen years ago it was realized that this was not good enough. What companies like Microsoft do research for is to set their technological agenda, in order to develop the technology which people will turn into products in some years time. For this you get to have skill and experience which may be of use also in the short term. Microsoft Research has a sort of dual role: to have its eye on the far horizon, and to be enough in touch with the product activities today, to say: we can help them out. Therefore in Microsoft Research you always find some people doing short-term work with the product folk, although their overall goal remains the distant future.
Microsoft set up its research operation in Redmond about ten years ago, and hired Rick Rashid from Carnegie Mellon University in Pittsburgh to head it. After five years the leadership was very pleased since it turned out that, although the focus was long-term, practically every major Microsoft product had benefited from the effects of the research programme. It was decided to triple the research efforts, which amounted to hiring about three hundred people. The Microsoft Research lab in Cambridge started in July 1997 as part of this effort; my duty was “to hire the best people and have them do what they are good at”. Luca Cardelli came over from the West Coast to work on programming language theory, resulting in a very strong group in that area. There is a logic-based tradition in computer science theory in Europe, which you find in the US hardly at all. Probably this is the main area of computer science where the European tradition is thoroughly distinctive.
Our business is completely people-based: the worst mistake is to take on someone who is not absolutely excellent because the area is important. We are free to choose our own areas of work, but are not uninfluenced by what is done at Redmond. For example, we don’t work in graphics or in natural language understanding. We do work, however, in adaptive systems and statistical learning theory, although they have a strong crew at Redmond there. This is because we were able to get individuals who are of extreme value to the company. It is largely for this reason that research is done in the same area in several places - for example the sun never sets on Microsoft vision research.
A list of what we’re presently working on includes: programming languages, learning, vision, adaptive systems, networks, congestion avoidance, and performance management. Statistics runs as an intellectual thread through much of our work. In information retrieval we work on extension from text to sounds and images. We’re starting work in operating systems and hardware, and putting together a group in security. We’re now 45 researchers, increasing 10-15 a year.
University people considering joining us hear that we want to regard ourselves, and be regarded, as part of the intellectual scene in computer science research. We realize that most research is done by other people. It is therefore in our own interest to be open and forthcoming with others about what we do. Our university relations are very important: presently we have five people to look after our university research relations in Europe (Microsoft’s Europe includes Russia, Israel, and South-Africa). We intend to extend our contacts to 40-50 of the best academic research departments in Europe, in addition to bodies like INRIA and CWI, which are not like anything in North America.
Our contacts cover a wide spectrum: from helping a university research group move their research to work on Microsoft platforms because their industrial sponsors ask for it, to paying for research and sponsoring students.
Our informal mixture between the long-term and watching out for short-term benefit would be a good recipe for universities and other research outfits. Across the world university computer science departments have come under pressure to show their relevance, and to maintain useful contacts with industry. This makes it difficult for them to undertake work which is doubtful of outcome, long-term, speculative, chancy. But that is what research is meant to be about. It ought to be possible to do both, and that is what we try to do. It is a sad state of affairs when companies such as Microsoft find themselves supporting work in universities that is longer-term than the regular funding bodies will pay for. Something here is not quite as it should be.