AI’s Brain Drain
Tech companies are hoovering up academics. So who’s going to teach the next generation?
AI wasn’t always the sector that everyone in tech wanted to work in. In fact, as recently as a decade ago, it was as uncool as it got. (The 90’s and early 2000’s were even referred to as an AI Winter.) So who’s behind the ideas and infrastructure that go into large language models like ChatGPT? Academic researchers, who have been building the foundations of AI since the 1950’s.
Now those academics are finding themselves outside the gates, as lots of compute and data are locked up in big companies that have billions in funding for the massive data scrapes, specialized computer chips (GPU’s) and other components necessary to grow at today’s scale. (For example, Stanford’s Natural Language Processing Group has something like 60 GPUs for themselves, while Meta plans to procure 35,000.)
In order to continue their work, some academics partner with tech employees and accept industry grants in order to do their research. Others focus their scholarship on work that has commercial applications. But many are simply jumping ship to take jobs at tech companies. Sure, twist their arms: The salaries start at over $300,000 and can reach $20m for four years. Today, 70% of phDs work in private industry, compared to 21% just two decades ago.
On Episode Six of Season One of the podcast, I spoke with NYU professor Kyunghyun Cho, who told me that while 50% of his time is spent teaching, the other half is spent working for a biotech company that uses machine learning to design new antibodies.
If this talent drain continues, who’s going to do the independent academic research needed to look at (and create) the future of technology without being motivated purely by profit? Imagine if all cancer research were done by drug companies, or all astronomers were on the SpaceX payroll?
What’s next? The companies aren’t incentivized to support academic research: Their sole function is to turn it into products and ROI, not support the public good. We as a country need to give academics freedom and support to focus on this fundamental work. That is the place of government investment — investment that ultimately benefits our economy and helps us to remain a tech leader, sure, but it’s also an investment in national security and more.
Fei-Fei Li is a Stanford professor and co-director of the Stanford Institute for Human-Centered AI who’s been called the godmother of artificial intelligence. During Biden’s last State of the Union address, she directly pleaded for a national warehouse of data sets and compute — part of what the Washington Post called a “moonshot investment” to help keep independent researchers in the game.
There are definitely glimmers of hope: Microsoft donated $20m in computing credits to the National AI Research Resource (basically a star researcher’s salary, but it’s a start). Last year, the National Science Foundation invested $140m in university-driven research institutes dedicated to studying how AI could help offset the effects of climate change and other topics. And the Create AI Act, which has bipartisan support in the House and Senate, will come up for a vote soon.
But there’s another problem that’s intrinsically linked with academic capture, and that is immigration policy. Should the academics choose to stay in academia, they will be teaching students, some of whom will be here on visas. Every one of those students should be given a job to remain in the States. The problem is that our immigration system is so outdated that it hasn’t caught up with the need for immigrants in tech — especially AI — making it painfully time-consuming for workers to get green cards. The talent pool is so small that Meta has allegedly hired AI researchers and engineers without even interviewing them.
The potential losses are huge: Just think about the Turkish researcher who had to return home after attending MIT because he couldn’t get funding for his work on information theory or an academic posting that would sponsor his visa. He ended up selling his discovery to China…which used it to make the leap from 4G to 5G.
Google recently wrote a letter to the Department of Labor, begging for updates to Schedule A, a list of occupations that the government certifies as not having enough American workers, to include AI and cybersecurity. (Those occupations haven’t been revised in 20 years!) The government, too, sees the threat that the current immigration process poses to America’s ability to lead the world in AI, not to mention protect itself from cyber attacks: Biden’s executive order on AI asks the Department of Homeland Security and the Department of State to work to streamline the immigration process for highly skilled workers in AI and other technologies.
Allow academics to follow their curiosity, not profit. And allow top talent to live here to help bring those discoveries to light.
Worth the Read
The Wall Street Journal tracks how tech workers are retooling to work in AI.
When Google revealed its AI search the other week, things instantly got weird. I mean, should you put glue on your pizza to help the cheese stick? More fun here. (Google quickly backtracked, eliminating five of the six questions it had launched with for the public.)
The Atlantic claims that media companies are making yet another mistake by capitulating to AI.
AOC gets behind the bipartisan Defiance Act, which would allow nonconsensual victims of AI porn to sue in court.