Upheaval at MERL: Mitsubishi Electric Breaks Up Famous Computer Science Lab
Mitsubishi Electric Company of Japan has quietly disbanded the long-term research wing of its most famous international outpost, the Mitsubishi Electric Research Laboratories (MERL) in Cambridge, MA, Xconomy has learned.
The research lab, long home to some of the world’s leading investigators in computer graphics, artificial intelligence, user interfaces, and speech recognition, lost its administrative and spiritual leader last October when research director Joseph Marks was fired in an internecine dispute over the need for basic research and external collaboration at the company, according to current and former MERL staff members. A subsequent wave of layoffs and resignations among senior scientists was capped in June by a formal reorganization that eliminated the research lab as an independent entity, according to Marks, who spoke with Xconomy at length Monday night.
MERL CEO Richard Waters confirmed the change, but says it was part of a much-needed overhaul of MERL’s scientific and engineering staff. This restructuring takes the lab from two divisions—the research lab, which historically focused on studies with longer-term payoffs, and an advanced technology development lab, which had a shorter-term thrust—into six product-oriented groups organized around technology areas such as computer graphics and digital communications. The dozen or so research lab members who remain at the company have been folded into these groups.
But while the restructured organization will keep its name and its offices at 201 Broadway in Kendall Square, “the original MERL is gone,” Marks said in a telephone interview from Los Angeles, where he is now vice president of software research and development at Disney Animation Studios.
At least eight researchers are confirmed to have departed the lab since Marks’s firing, including:
• Candace Sidner, an expert in computational linguistics and collaborative interfaces who has joined British aerospace firm BAE Systems in Burlington, MA
• Charles Rich, Sidner’s husband, who is known for his work on multi-user interactive multimedia environments and is now a professor of interactive media and game development at Worcester Polytechnic Institute in Worcester, MA
• Hanspeter Pfister, a 3D visualization expert and former associate research director of MERL, who has joined the faculty at Harvard University’s School of Engineering and Applied Sciences and now directs visual computing research for the school’s Initiative in Innovative Computing
• Baback Moghaddam, a computer vision researcher now employed by NASA’s Jet Propulsion Laboratory in Pasadena, CA
At least three more researchers still employed by MERL are known to be on the verge of leaving, according to the four independent sources (including Marks) whom we contacted for this story. And thanks to a steady flow of CVs and job-talk visits from MERL researchers at other industry research labs, MERL’s restructuring has become common knowledge in the computer science community. Several companies, including Adobe Systems, are actively recruiting among MERL’s remaining research staff; “a panty raid” is how one of the sources describes the situation.
Mitsubishi Electric has kept silent until now about the changes. MERL’s Waters told Xconomy today that the reorganization was intended to better align MERL’s work with the needs of its parent, Mitsubishi Electric. “There is no question that there has been some rearranging around here due to the need to focus on what the business units are doing,” Waters says. “That is one of the realities of corporate research labs. There were a couple of people last fall who were let go on those grounds, and since then there have been some more people who self-selected themselves to leave the lab. But it is not a question of either lab dying. It is just a reorganizing based on a different top-level organizational principle.”
Reorganizations are, indeed, part of corporate life, and the reputations of industrial research labs have always waxed and waned in sync with the fortunes of their parent companies. What is unusual about the housecleaning at MERL, in the view of Marks and other former MERL researchers, is that it comes at the height of the lab’s prominence and productivity.
MERL was founded in 1991 by computer scientist Laszlo “Les” Belady. With the blessing of executives in Japan, Belady gave researchers broad freedom to pursue basic problems in AI and other areas, bringing American-style innovation into a corporate culture where the Japanese tradition of lifelong employment discouraged risk-taking, according to former staffers familiar with MERL’s history. And the lab quickly established itself as a powerhouse, generating advances in areas such as collaborative computer interfaces, digital audio, digital typography, and image recognition. The research group also contributed several papers each year to the Association for Computing Machinery’s prestigious annual conference on computer graphics, SIGGRAPH, often running neck-and-neck with Microsoft Research for the highest number of papers accepted.
From 1996 to 2000, the lab’s research director was James D. Foley, legendary co-author of several of the most widely used university textbook on computer graphics. Marks says that when he took over the post in 2000, he faced a challenge: Mitsubishi Electric didn’t always have product lines or markets where the lab’s innovations could be directly applied. To extract more value from the lab’s research and help to pay its own way inside Mitsubishi Electric, Marks says, he tried broaden the lab’s research funding base by engaging in external projects with government agencies and other technology companies.
“The problem I inherited when I became research director was how do you structure a world-class research lab at a company that doesn’t have products or services that are in the top five in the world—in contrast with research at a Microsoft or an Intel or a Disney, where the brands are virtual monopolies and it’s clear that the research labs should feed the monopoly,” says Marks (pictured at right). “My answer was to try to sell the research two or three times—once to the company, once to the government through research grants, and once more by licensing our technology to non-competitors. This is essentially what Xerox PARC and other labs have been doing for a number of years, and I think that’s the new wave of industrial research.”
Marks scored some successes, notably licensing a digital typography system called Saffron to Adobe Systems and famous type foundry Monotype, and winning funding from U.S. intelligence agencies for the development of DiamondTouch, a multi-user, multi-touch display system that was one of the precursors to the Microsoft Surface tabletop computing system. “I thought we were creating some good traction, and initially management in Japan seemed to be quite happy with that,” says Marks.
“We were really the darlings, the crown jewel” of Mitsubishi Electric’s research operations, at least until the early 2000s, confirms one former researcher who spoke with Xconomy on condition of anonymity. But “over time, this perception got eroded down,” the source says.
In recent years, according to Marks and other former and current MERL researchers, executives at Mitsubishi Electric’s headquarters in Japan began to question Marks’s strategy and MERL’s practical contribution to the company. Management in Japan pressured MERL researchers to participate more directly in product development, these sources say.
For a time, research continued as usual at the lab. “One of the great things about MERL was that people like Joe Marks and to some extent our CEO Dick Waters really shielded us from this Japanese-style management,” says one anonymous source. “Boy, did that change.”
According to Marks and other sources, the final crisis began with the appointment of Kazuo Kyuma, former director of Mitsubishi Electric’s Central Research Laboratory in Amagasaki, Hyogo, Japan, as head of the company’s Corporate Research and Development division. Sources say that Kyuma, a long-time product manager with the company, had been friendly toward MERL in the 1990s, but that as he rose through the company’s ranks, his skepticism toward the lab’s basic-research focus seemed to increase. Current and former MERL researchers believe Kyuma pressed Waters to restructure the research lab around product development projects specific to Mitsubishi.
“Management [in Japan] changed, and there was a 180-degree turnaround on the whole idea of external engagement, including licensing and even collaboration, and I was summarily fired,” Marks says.
Marks says he was literally escorted from the building on the day of his firing last October. “It was an unceremonious kick out the door,” he says with noticeable annoyance. “It’s very unusual for a member of the research community to be treated that way, especially after five years of hard work directing the lab and 12 years at MERL altogether.”
When Xconomy spoke with CEO Waters, he confirmed some parts of the history related by Marks and other former MERL researchers and strongly disputed others. Waters says there is no disagreement within MERL or Mitsubishi Electric over the need for external engagements with collaborators. “That is a thing that MERL has done in the past and is doing now, and we certainly intend to continue doing,” he says.
The real difference of philosophy, according to Waters, was about the mix of research projects Marks authorized as research director. Waters and managers in Japan felt that too few of these projects had a chance of paying off for Mitsubishi Electric. “The reason companies pay for research labs is so that the labs will do them some good,” says Waters. “The old joke about the drunk who only looks for his missing keys under the lamppost isn’t so stupid. If that’s the only place you can see, that’s where you’d better be looking. Doing good research is fantastic, but you need to do it near the lamppost, not across the street where it’s never going to do the company any good.”
Waters also said Marks’s termination and the recent restructuring were inevitable and would have happened even if someone other than Kyuma had been in charge of Mitsubushi Electric’s corporate R&D division. “In general MERL has reported to a new person about every four years, and yes, those people are different,” Waters says. “Some are more active than others, and the rate of change may be higher or lower. But the fundamental direction isn’t different.”
Waters declined to speak about the specific circumstances of Marks’s firing. But whatever the reason, the firing shook the senior staff in the research division to the core, Xconomy’s sources say. It was “a complete shock to everyone—it was just out of the blue,” says one former MERL researcher. “A lot of people I have talked to think that letting Joe go was really a mistake. Joe had assembled a world-class cast of people in computer graphics and machine learning, and those are the people who are leaving now.”
With all the departures, MERL “is already not much of a force” in computer science research, this source says. And with conditions growing increasingly uncomfortable for the remaining researchers, another wave of resignations may be imminent. “The uncertainty is not being remedied,” the researcher says. He adds, “Some people are saying that it’s only a matter of the academic cycle starting this fall before most people leave.”
Waters, however, says he believes that the overhauled MERL is stronger than before. “Change is always upsetting. But the essential core values of MERL which make it a great place to work—things like open publication and being part of the research community, and having funding to support lots of students—have not changed,” Waters says. “There are certainly a few people who I strongly regret have left, but most of our prominent people are still here and they are doing some great stuff.” The lab has already recruited researchers to replace some of the departed staff, “and we will be hiring more,” Waters says.
Marks, for his part, says that despite his acrimonious departure, he is proud of the accomplishments at MERL during his tenure. “My biggest regret is that the experiment that we had started, to try to come up with a way of doing industrial research that was obviously more beneficial to the company, was cut off,” he says. “That is very disappointing. However, the people who were part of MERL Research who are now going to other places are going to do fantastic work. Their influence will be felt throughout the various industries they’re working in. So the ghost of MERL Research may have just as much impact [as MERL would have had if it had continued]—maybe more so. There’s something gratifying about that.”
UPDATE 08/01/07 11:40 a.m.: This article has been picked up by Slashdot, where there is a vigorous discussion going on about MERL’s record and the place of basic research in the world of corporate R&D. -W.R.
UPDATE 08/01/07 5:25 p.m.: We’ve just published a followup story with more details about the researchers known to have left MERL. -W.R.