Collecting that type of information means collecting millions upon millions of data points and then analyzing them using very high-powered computers and machine-learning techniques. (Machine learning is the study of computer algorithms that improve automatically through experience.) One of the papers presented in Seattle focuses on the ranking of search results. Search engines typically display results by date, which may or may not be useful. So, Dumais is looking for ways to teach a search engine to take into account previous Web or desktop searches. Even better might be a search engine that examines content already indexed on a user's PC and returns results with that in "mind." A major drawback of existing searches is the inability of the engines to incorporate much information in a query. Dumais says "users know a lot about what they are looking for. Searches should take a much richer context into account." That will change in the future, she says. Unlike pure research labs at old-style tech companies, Microsoft's research arm works closely with product groups. In January, the company formed LiveLabs, a partnership of Microsoft Research and MSN, designed to move ideas about search from the labs to the marketplace. The head of LiveLabs now reports directly to Chief Software Architect Ray Ozzie, a sign of the company's heavy emphasis on search.