“We’re still doing a lot of basic research,” it’s not just development. That’s what Jeff Welser, lab director of IBM Research-Almaden told me on a recent visit.
Given that IBM’s Watson technology was initially a system designed to play the TV game show Jeopardy, but is now a general-purpose machine learning system thought to be the fastest-growing part of IBM’s business, it’s not surprising that the company is hoping another wild seed will bear profitable fruit. But where do these seeds come from?
Welser told me that research projects often originate from open mike sessions, held once a year. These are serious events and, at the same time, entertainment along the lines of Shark Tank meets American Idol.
“One or two dozen people stand on stage and pitch in the first round,” Welser explains. With feedback from the audience, usually in a free-flowing Q&A, the researchers usually winnow themselves down to three or four, based on the community’s interest. Then we do another round in the auditorium with those researchers fleshing out what they want to do, including how many people they will need from where in the organization. Then the judges—myself and key technical leaders from the lab—winnow that down to one or two projects.”
Welser says that, ideally, these selections are what the company calls “Grand Challenges,” that is, they are trying to answer a challenging scientific or technology question as well as to build a usable technology or proof of concept.
“One year,” he says, “we also used online voting. People could ‘like’ something and could also indicate ‘I would work on it.’ This process focused on finding new, smaller projects that could combine different threads of work going on in the lab, rather than defining new grand challenges.
IBM’s neuromorphic chip, for example, came out of an open mike session in 2006.
IBM also sets its research agenda in a more formal process, involving all of its labs around the world. Welser explained: “We annually run our global technology outlook. We ask people to submit a page or two outlining an important trend or technology disruption. We’ll get things describing, say, the ‘aging demographic,’ or ‘blockchain.’ Typically we’ll have 200 or so submissions. Then we try to group the submissions that are around the same topic areas, without doing any winnowing away of ideas we don’t like. We’ll send those back to the people who submitted them, and ask them to work together and come back with the kernel of the idea that they were all trying to get at.”
Welser adds that those ideas are presented to people at IBM’s business units, professors at universities, and others to try to figure out which ideas have legs. “By the end of the year, we want to be down to six or seven topics. Then we spend a full day with [IBM CEO] Ginni Rometty to determine [what it will take to pursue a given idea], including what it will take in research funding, what it will mean for the business units,” and how best to move forward.
IBM’s involvement in the open-source Hyperledger project, which involves using blockchain for internal business transactions, came out of this process, Welser says.
Finally, a lot of research ideas come from summer interns; the company has an extensive summer intern program.
Recently, one intern was interested in looking at what he could tell about the work environment by using Bluetooth beacons. That led to the RouteMe2 project, a collaboration with the University of California at Santa Cruz that developed assistive navigation technology for use by special needs passengers taking mass transit. The project, funded by a $1 million grant from the National Science Foundation, had offshoots focusing on monitoring patients, staff, and resources in senior care centers, as well as providing location context information for people navigating telepresence robots.
Another intern, interested in food spoilage, used AI to track subtle changes in the shape and behavior of small single cell organisms when they are exposed to the harmful bacteria associated with milk spoilage. This has evolved into a joint research program with Cornell University.
Correction made 9 Oct 2017