At the world’s largest artificial-intelligence conference, known as NeurIPS, Weichen Huang almost blended in. Last month in Vancouver, he was one of numerous researchers explaining their machine-learning projects to crowds of professors and engineers. But Huang wasn’t a professor, an engineer, or even a graduate student. In fact, he’s still applying to college.
The 18-year-old was among the winners of NeurIPS’s first research competition for high schoolers. He was thrilled to be among 17,000 computer scientists and
Or subscribe now to read with unlimited access for less than $10/month.
Don’t have an account? Sign up now.
A free account provides you access to a limited number of free articles each month, plus newsletters, job postings, salary data, and exclusive store discounts.
If you need assistance, please contact us at 202-466-1032 or help@chronicle.com.
At the world’s largest artificial-intelligence conference, known as NeurIPS, Weichen Huang almost blended in. Last month in Vancouver, he was one of numerous researchers explaining their machine-learning projects to crowds of professors and engineers. But Huang wasn’t a professor, an engineer, or even a graduate student. In fact, he’s still applying to college.
The 18-year-old was among the winners of NeurIPS’s first research competition for high schoolers. He was thrilled to be among 17,000 computer scientists and a few walking robots, even if he was too young for the data-themed cocktail parties. “I do like reading machine-learning papers sometimes in my free time,” Huang said on a Zoom call from his hotel room, after flying in from Dublin, Ireland. “I really wanted the opportunity to be able to see all of these papers in person and to connect with other researchers.”
In the era of ChatGPT and hypercompetitive college admissions, smart, driven Gen Z students want to make their mark in AI. But many in the computing community raised an eyebrow at the news of a high-school contest, expressing concern that it would set unrealistic expectations for, and exacerbate inequities among, students entering a field with racial and gender disparities.
NeurIPS organizers said the goal was to “get the next generation excited” about machine learning and its societal benefits. But to critics, the competition epitomized the annual conference’s transformation into a showcase for the AI gold rush, which has left many feeling like its core purpose — sharing research — has been diluted. “To be a little bit colloquial, I feel like they slapped on a science-fair aspect to the entire conference,” said Gautam Kamath, an assistant professor of computer science at the University of Waterloo in Ontario, Canada. “It raised the question of what exactly is the purpose of NeurIPS? Which I think has changed a lot over the last decade.”
NeurIPS, which stands for the Conference on Neural Information Processing Systems, began in 1987 as a small gathering of academics toiling in obscurity. In 2017, it drew 8,000 registrants. In 2024, two years after ChatGPT was unleashed, that number more than doubled. Demand was so great that attendee hopefuls had to enter a lottery if they weren’t presenting research of their own.
ADVERTISEMENT
Inside the Vancouver Convention Centre, OpenAI’s cofounder spoke, as did superstars from Stanford University and the Massachusetts Institute of Technology. The heavy hitters — Meta, Alphabet, Microsoft, Nvidia — were there recruiting and schmoozing. Getting a paper accepted at the conference is computer science’s equivalent to publishing in a prestigious journal. But participants complained that with NeurIPS serving as a job fair and industry expo, its peer-to-peer process for vetting papers — a record 15,671 were submitted — seemed stretched thinner than ever. “Everybody says, ‘The peer-review system for those giant conferences is broken. We need to find something more sustainable,’” said Clément Canonne, a senior lecturer in computer science at the University of Sydney. “And just when that happens, we hear, ‘Oh, by the way, we found the time and resources to actually somehow ask high-school students to also submit.’”
In April, NeurIPS invited high schoolers to send in papers about how machine learning could bring about “positive social impact.” Their projects could be about issues like agriculture, climate change, or education, as long as they were independently and wholly done by the students. More than 330 entries were whittled down to four winners, including Huang’s, and 21 runners-up, a selection rate of roughly 8 percent. That made the high-school track even tougher than the actual conference, which accepted one-quarter of its submissions. (NeurIPS chairs did not answer questions about how projects were evaluated.)
For most of the average high schoolers, there’s no way you can afford that kind of computing.
Kamath worried about the message being sent to students (and parents) desperate to impress admissions officers. “This just seems like another thing that raises the stakes even more, another thing for people to compete about,” he said.
He and other published scientists said that advancing the frontiers of AI requires resources not usually available until college or graduate school: a grasp of high-level math and programming languages, the understanding of how to conduct an experiment and write a technical paper. Then there’s a need for computing power to train a model on tons of data. “For most of the average high schoolers, there’s no way you can afford that kind of computing,” said Fred Zhangzhi Peng, a Duke University graduate student who researches machine-learning applications for biomedical engineering.
Researchers questioned the benefits of elevating select high schoolers with access to those resources, and wondered how innovative most such work could be. Shriram Krishnamurthi, a computer-science professor at Brown University, has noticed that as more high schools teach programming with wildly varying degrees of rigor, incoming students are increasingly showing up thinking they know more than they do. “There’s this weird thing where they are very competent at patching together some things and producing graphs that look nice,” Krishnamurthi said, “but their understanding of what they did is pretty low.” (He added that he wasn’t casting judgment on the individual winners at NeurIPS. “There has always been and will always be a sliver of students that are extraordinarily capable,” he acknowledged.) Outside of NeurIPS, high schoolers can pay companies a handsome fee to co-author academic papers, a cottage industry that’s widely criticized.
ADVERTISEMENT
When Canonne started graduate school, he hadn’t published any research. But these days, “people apply for Ph.D. programs with already one, two, ten papers in top conferences,” he said. “And now, at the high-school level, some people will have some research projects, so that’s going to make it even harder.” He added, “If your high-school teacher does not know about this, or doesn’t have the capacity to supervise you, or is just not a researcher, how do you do that? … I’m not sure it’s a well-thought experiment in fairness.”
NeurIPS organizers said in a press release that they received submissions from “around the globe.” Conference chairs did not answer questions about participants’ demographics.
Just because somebody’s young doesn’t mean they can’t do cutting-edge research.
Huang said he was glad for the opportunity. Around age 10, he started playing around with the family computer and trying to code. Upon getting an Amazon Echo Dot for Christmas, he wondered if he could engineer Alexa, its then-rudimentary voice assistant, to look up answers to questions online and summarize the text. “That,” he said, “was my first AI project.” He savors the high of cracking a tough problem — “the moment when you get that idea, you get a rush of adrenaline.”
Huang credited his father, a software-engineering director at Docusign, with helping him get started. But he said that he’s mostly taken the initiative to study machine learning on his own, by watching YouTube tutorials about programming in Python, participating in coding competitions, and taking online, open-to-the-public MIT courses in multivariable calculus and algorithm design. He’s done all this, he said, because St. Andrew’s College Dublin, the private high school where he’s a senior, didn’t teach computer science until recently.
For his “transition year,” in which Irish high schoolers are given time to gain real-world experience, Huang emailed local professors to see if they could use an intern. He ended up in the lab of Kathleen Curran, an associate professor at University College Dublin, who uses machine learning and AI to analyze medical imaging.
ADVERTISEMENT
Curran said she wouldn’t turn down someone because of their age, and considers instead “how well the student would fit into the group and how much additional work it would be.” She said she was impressed with Huang’s work ethic, enthusiasm, and coding skills from the start. “I think that he’s self-motivated, he’s genuinely interested,” she said.
Weichen Huang explains his machine-learning research to NeurIPS attendees in Vancouver. Courtesy of Yangcheng Huang
One of Curran’s master’s students was using a machine-learning tool to detect signs of Alzheimer’s disease in brain scans. Huang teamed up with him to research existing detection methods, and helped code a new model as well as experiments to test it, he said. That student graduated, but Huang remained curious. He went on to develop a model that could form a diagnosis from not just images, but also cognitive tests and medical histories.
That was the project he submitted to the NeurIPS contest, which he came across when he happened to look at its website. The other winners, hailing from Massachusetts, California, and New Mexico, used machine learning to diagnose leukemia, detect mountain lions, and identify rainwater-harvesting sites in Africa.
Huang said that he was interviewed by two conference organizers and didn’t receive written peer reviews, unlike other research submitted to NeurIPS. Finding out he’d made the cut was “a pleasant surprise,” albeit one that landed in the middle of college applications and exams. For the high-school showcase, he hauled in a poster filled with charts and text, and spent a good chunk of two-and-a-half hours explaining his work over and over to strangers. He’d been “a little nervous about presenting,” but said he enjoyed meeting the other high schoolers as well as established scholars.
Huang said he didn’t see the competition as inherently stressful, but rather as “an extra opportunity,” and acknowledged the privileges that he has. “There’s quite a big disparity within high schoolers who have the ability to do research, like me, and those who possibly don’t have the necessary resources,” he said.
But he noted that some tools are becoming widely available. To train his models for his Alzheimer’s project, he said, he mostly used Google Colab and Kaggle, which provide a limited amount of computing power for free.
ADVERTISEMENT
Curran said she would like to see conferences like NeurIPS seek out participants from low- and middle-income countries. But she supported the idea of carving out space for up-and-comers. Before NeurIPS, Huang had previously contributed to work submitted to three other conferences, and not in siloed high-school programs. “Just because somebody’s young doesn’t mean they can’t do cutting-edge research,” Curran said, “so why not?”
Instead of highlighting a few winners, Huang suggested, NeurIPS could invite dozens of students into the same room to present their work and meet each other. When the conference was held in New Orleans in 2022, organizers arranged for scientists and industry leaders to spend a day discussing AI with local high schoolers.
In a statement, the conference’s general and communications chairs said that “NeurIPS is committed to advancing inclusivity among high school and college students, both for those who are developing emerging interests in the industry and for those who already have an established passion for the field.” They also said that “the concerns about how this program might have privileged some students over others were heard, and this feedback will be factored into any future years.”
Huang said that he wants to keep researching AI, especially as it overlaps with neuroscience, and to stay in academe. “I guess it’s a little stressful, the first time I submitted to a conference,” he said. “But the thing is, what’s the worst that could really happen — you get rejected, you improve a bit on your project, and you try to submit it somewhere else, right?” As of early December, he was waiting to hear whether he’d gotten in via early action to his dream college, MIT. That, he said, was really stressful.
Stephanie M. Lee is a senior writer at The Chronicle covering research and society. Follow her on Twitter at @stephaniemlee, or email her at stephanie.lee@chronicle.com.