Companies and universities are sabotaging each other by trying to corner the proceeds of joint research into artificial intelligence, a conference has heard.
Microsoft machine learning expert Kuansan Wang said that a “more aggressive” stance from universities was forcing his company to pay more attention to intellectual property rights assertion when it hosted doctoral students.
“If we are not careful, the university would want a claim on the IP,” Dr Wang told Times Higher Education’s Research Excellence Summit: Asia-Pacific, held at the University of New South Wales. “That creates lots of complications. It’s certainly not helpful.”
Pascale Fung, director of the Centre for AI Research at Hong Kong University of Science and Technology, said that industry should share the proceeds. “Why should the companies own all the IP rights when the students are trained by us?” she asked.
Professor Fung said that Bell Labs, where she was a doctoral student in the 1990s, had been “a lot more relaxed” about intellectual property than companies today. She had published jointly with Bell Labs at the time, but such opportunities were now unusual, and both sides – companies and universities – needed to show more flexibility.
She said that intellectual property protection was “useless unless you make something out of it”, becoming a “malicious kind of competition” that thwarted rather than encouraged technology development. “It should be seen as some kind of seed to future innovation,” she said. “We allow the students to do start-ups, and maybe the companies can have a stake.”
Toby Walsh, Scientia professor of artificial intelligence at UNSW, said that he supported the University of California, Berkeley’s approach – making research widely available while providing recognition to individual researchers – over intellectual property protection that generated minuscule earnings and did not incentivise technology development.
Berkeley had “found in hindsight that they get far more in return in philanthropy than they would if they’d tried to hold on to the IP themselves”, he said.
Professor Fung said that the biggest challenge facing academic artificial intelligence research was access to the massive datasets needed to improve the technology. “Universities today cannot compete against the Facebooks, the Googles, the Microsofts and the Baidus of the world because we don’t have access to that huge amount of data,” she said.
Hunger for data was making the laboratories of the internet companies so large that they monopolised researchers, Professor Fung said. “Universities are having to compete with these industries to get talent. We have no problem getting students, but we don’t have enough AI professors. They’re all in industry,” she said.
Dr Wang, who heads Microsoft Research Outreach Academic Services, said that universities needed access to big datasets so that they could generate the industry’s future workers. But privacy and copyright issues precluded companies from simply handing over data.
Microsoft, he said, had searched for years before settling on a data source that “our lawyers think we can safely share” – scholarly communication, a phenomenon with “the opposite of a privacy problem” because authors wanted as much exposure as possible.
Dr Wang said Microsoft had used machine reading to curate a dataset of more than 200 million academic publications extracted from the internet. The company is making it available as a teaching and research resource, he said.