In recent years, attracted by salary, benefits and hardware conditions, more and more AI scholars choose to join the industry, joining various technology giants or startups. But according to a study by the University of Rochester, this trend has had a negative impact on university innovation: the more AI professors who work in the industry, the fewer AI companies their students create after graduation and the less funding they receive for their start-ups.
Human capital is critical to AI-driven innovation. Between 2004 and 2018, the lack of human capital required for AI research and development caused a serious brain drain among AI professors at North American universities, who joined the industry. In terms of intensity, the loss of these AI professors also reduces the amount of early funding graduates receive to start their own businesses, according to the University of Rochester study. The interruption of knowledge from professors to students makes the redistribution of human capital negatively affect innovation.
In 2018, nearly 20 percent of academics at US universities left their faculties to join the AI industry. In the past 15 years, 153 ai professors have joined the industry at North American universities, while another 68 have moved into industry while retaining university positions. The biggest departures in recent years have been at Carnegie Mellon University, the University of Washington and the University of California, Berkeley.
And the rising tide looks set to continue.
More and more AI professors are working in the industry
The study, which looked at the loss of AI professors at several universities between 2004 and 2018, found that since 2009, the number of AI professors at North American universities leaving academia for the industry has increased exponentially.
Among them, the big tech companies employ a large number of academics who are skilled in a particular technology. For example, two of the three senior scholars who won the 2018 Turing Prize are already in the industry, Geoffrey Hinton at Google and Yann LeCun at Facebook. And finally, MILA head Yoshua Bengio, a professor at Montreal University, became an adviser to Microsoft when the company announced its acquisition of Maluuba in 2017.
As the chart above shows, in 2004, no AI professors left the field, but in 2018, 39 joined the field. In addition, in 2010, only 4 percent of a university’s professors engaged in industry were cited on average, but this figure rose to around 20 percent in 2018.
The chart below shows the top 15 North American universities for AI professor loss between 2004 and 2018. The top three universities for AI professor loss are Carnegie Mellon University (CMU), the University of Washington, and UC Berkeley. Carnegie Mellon lost 17 tenured professors, but no more than tenured; The University of Washington lost seven tenured professors and four assistant professors. In the Canadian sample, the University of Toronto lost the most AI professors, including six tenured professors and three assistant professors.
So where are all the professors coming out of academia?
The chart below shows the companies that hired the most AI professors between 2004 and 2018, with Google, Amazon, and Microsoft poaching the most AI professors from North American universities. During that time, Google and its DeepMind subsidiary hired 23 tenured and probationary AI professors at North American universities; Amazon and Microsoft employ 17 and 13 AI professors, respectively. In addition to tech companies, big names in the financial industry such as Morgan Stanley, American Express and jpmorgan Chase are also poaching AI professors. It is also worth noting that of the 221 professors who joined the profession, about 45% went to public companies.
Professors influence students to start businesses
The University of Rochester study found that having an AI professor in the field had a negative impact on students’ post-graduation success. But how did this conclusion come about?
The authors do not directly analyze the quality of education students receive after these professors leave, but focus on whether their departure leads to a decrease in the number of student entrepreneurship. Studies show that students are less likely to start their own businesses when these top professors are replaced by professors from lower-ranking schools. Moreover, four to six years after tenured AI professors leave, graduates are less likely to become AI entrepreneurs and receive less early funding to start AI companies.
Starting a business in AI requires a high quality education
Unlike non-AI sectors, entrepreneurship in AI requires a higher level of expertise and skills.
The chart below shows the north American universities with the most AI companies founded by their graduates after their highest degrees, with MIT and Stanford leading the way with 77 and 72 graduates, respectively. Carnegie Mellon University ranked third in producing AI entrepreneurs, producing 39. The University of Waterloo is the Canadian university with the most ALUMNI of AI entrepreneurs, with 21 graduates founding AI companies.
The analysis shows that a high-quality university education is very important for AI entrepreneurs. AI professors play an important role in providing relevant education.
Next, the researchers explore the impact of the loss of AI professors on entrepreneurship.
Professor AI has the greatest impact on students four to six years after leaving
The researchers found that the loss of AI professors could have a negative impact on their universities’ graduates starting businesses in the field.
Those most affected by the loss of AI professors are those who graduate after four to six years, meaning they have little opportunity to interact with those who have left. At any one university, a significant increase in the number of departing professors reduced the number of AI entrepreneurs by 13%.
Ariel Procaccia, an associate professor of computer science at Carnegie Mellon University, worries about the rush of AI talent into the industry: “If the industry continues to poach top academics, who is going to train the next generation of AI innovators?”
But experts are divided on whether the downturn in the start-up economy has in turn hurt the development of AI.
Joshua Graff Zivin, an economics professor at the University of California, San Diego, said: “Just because students don’t start businesses after graduation doesn’t mean they aren’t working on AI. They may be contributing to the AI industry in other ways.”
But many experts think universities should increase their spending to make sure the next generation gets a proper education. “Machine learning and artificial intelligence are two of the most exciting and fast-growing areas of science,” says Scott Stern, a professor at the MIT Sloan School of Management. “We need to make sure we’re putting enough resources into them.”
Industry: we did not dig academic corner
The University of Rochester study has generated immediate buzz, but tech companies would disagree with the idea that industry is robbing academia. Google says it is the biggest supporter of academic research.
“Since 2005, we have invested more than $250 million in academic research, and Each year Google hosts more than 30 visiting scholars, dozens of doctoral students, and thousands of interns,” said Jason Freidenfelds, a Google spokesman. He says many professors who have worked at Google will return to universities.
Indeed, the most famous example is Li Feifei, a Stanford University professor and renowned artificial intelligence scholar. In November 2016, the news that Fei-fei Li, director of Stanford University’s ARTIFICIAL Intelligence Lab, and Jia Li, former head of Snapchat research, joined Google attracted wide attention. While serving as chief scientist for Google’s cloud AI, Li continues to serve as an associate professor at Stanford University and head of the Stanford AI Lab.
In September 2018, as her academic sabbatical came to an end, Li left Google and returned to Stanford University to teach full-time. She was replaced by Andrew Moore, the dean of Carnegie Mellon’s school of Computer science, who left his university post to pursue a full-time career in the industry.
In May, Li’s group’s work won the best paper of ICRA 2019, a top conference in the field of robotics. Her student and co-first author Yuke Zhu will soon teach at UT Austin.
In any case, the Rochester study tells us one thing: the growing presence of AI professors in the field is having an impact on student development. How can academia and industry benefit more when there are calls for closer collaboration? We won’t find out for a while.