Microsoft research offers peek into future

Watch out – your PC is cozying up to you. Today’s trends of faster processors, abundant memory, and massive storage are the framework to support systems that work more intimately with people, from sorting your e-mail to speaking in your own voice, to providing Star Trek holodeck-quality graphics sooner than you might think.

Those are just a few glimpses of the future emerging from the labs of Microsoft Corp. Research (MSR), a pure research facility with four offices around the world. Typical of pure research, many of the projects start as technology without a product. But not for long, says Rick Rashid, senior vic- president of Microsoft Research. He spoke Tuesday in Microsoft’s Silicon Valley Speaker Series, at the campus that in May became the latest site of Microsoft Research facilities.

“Our number one priority is to expand the state of the art in each of the areas in which we do research,” Rashid says. The close number two: to transfer those innovations to Microsoft products.

Consider, for example, software under development that monitors and models your behaviour. The initial implementation might be an intelligent e-mail sorter that assigns numeric values to all your messages, based on your past handling of similar e-mail. The boss’s missives may rise to the top as among the most promptly read; routine cc’s drop near the bottom, and junk mail hits the cellar. Microsoft uses such a filter internally.

The even smarter assistant will monitor your work habits and correlate them with your calendar. It may book extra time for an appointment that involves travel, based on commute data available from online traffic monitoring.

“Your notebook knows all that about you already,” Rashid says. “We think we could put this information to use,” helping you get more use from technology.

Some fruits of the research are fairly quickly evident. Specific products borne of MSR endeavors include ClearType, the more readable text used in the Pocket PCs, and the Tablet PC. Researchers have contributed to Microsoft’s streaming media, notably Microsoft Audio. They not only check for bugs in packaged software, they continue to develop new ways to test for bugs. Rashid estimates that as much as 20 per cent of MSR’s efforts are implemented in programming and development tools.

That Microsoft was able to simultaneously ship Windows 95 and Office 95 is due in large part to MSR’s development of optimization technology, so the programs required only 8M bytes of RAM. MSR was working on that technique several years before the Office and Windows product teams realized they could put it to good use.

“Today, it’s hard to find anything we do as a company that hasn’t been influenced by the research group” Rashid says.

Rashid came to Microsoft in 1991, the year MSR was founded. He had spent 12 years as a professor at Carnegie Mellon University, where he directed the CMU Mach Operating System Project. The Mach kernel is used by the Open Software Foundation as well as in corporate and University research laboratories.

Because of Rashid’s background and those of the many PhDs who work in MSR, the approach is heavily academic. The organization frequently partners with academic institutions on projects, and welcomes interns and academic visitors, he says.

Rashid won’t say how many patents MSR has produced in its decade of operations, but says it’s substantial. The staff also regularly publishes professional papers at programming conferences and academic events. MSR has sites in Redmond, Wash.; Mountain View, Calif.; China; and Cambridge, U.K.

In its first decade, MSR has explored aspects of technology that have grown even faster than Moore’s Law (the Intel adage that processor power doubles every two years). Among the areas that Rashid significant advances in the next five to 10 years:

– graphics, with PC rendering soon resembling 3D “reality level” images. Already, game consoles like Xbox and some PCs can produce the level of graphics in Pixar’s acclaimed Toy Story movie.

– plentiful, cheap, massive storage. “We’re on the verge of making a terabyte of data storage available even on laptops,” Rashid says. He describes a terabyte – which is a thousand gigabytes

– as enough storage to hold every conversation you’ve had your entire life, from birth to death. MSR is exploring the capability of huge databases with the TerraStorage Project, an enormous and ongoing geographic database. Already it fills 17 terabytes with files of photos and maps from the U.S. Geological Survey and Russian space satellite photographs.

– ever better speech technology. In 1962 AT&T’s Bell Labs made a computer the size of a room sing Daisy, reproduced by HAL in the movie 2001: A Space Odyssey. Rashid credits those forerunners but points proudly to a far better – if slightly flat – performance by synthesized voice nearly 40 years later. “This is not a technology that in any way threatens Simon & Garfunkel,” Rashid adds.

First the software studies your voice patterns. “You have to have a throat mike on for four hours, talking constantly, to extract the auditory qualities of someone’s voice,” Rashid says. Then the system can produce and “play” that same voice, not as recordings, but actually producing new speech that mimics the lilts and tones of the speaker.

– Mindnet “a huge natural language database automatically created from dictionaries,” in development at MSR. The dictionary information helps links words by their relationships. It is a much more sophisticated sibling of the grammar checking technology in Microsoft Word.

– Ask MSR, a natural language, automated question answering site that is used internally for fun more than work. It researches questions across the Internet.

“It’s a very early stage of the technology,” Rashid says. “We get about 50 per cent of the answers right.” For now, he says, that technology is best suited to help answer the very specific, expensive questions at the end of Who Wants to be a Millionaire, rather than the common-sense queries in the early rounds. Computers are apparently still learning common sense.