Headlines about the University of Wyoming poisoning case tell a confidentiality breach story, built around a Chinese national and a cache of hard drives seized at Los Angeles airport.

The far more important story is trusted input tampering. Prosecutors allege graduate student Aili Wang put a neurotoxin, acrylamide, in the water a fellow researcher used every day to make tea at his desk. His wife and colleagues boiled his kettle’s water down to 63 milligrams of white powder, and a chemistry professor identified it. The affidavit says a hidden camera then recorded Wang rinsing his mug with liquid from her own, and university records tied her uwyo.edu account to acrylamide bought on eBay.

Their professor, Mark Gomelsky, told the AP that Wang failed her first dissertation defense and grew envious of her colleague’s success using bacteria to deliver RNA into human cells. The alleged goal was to stop a competitor from producing results. Tampering with his trusted input to get there is an integrity breach.

It reminds me of Fei-Fei Li.

By her own account, a Princeton professor told her the ImageNet work was unethical. She moved to Stanford in 2009 and kept building it there. The ImageNet leaderboard became the field’s definition of success, and her Stanford profile calls ImageNet one of the three driving forces of the deep learning revolution.

The leaderboard rewarded gaming.

In 2015 the challenge banned Baidu for 12 months after organizers found its team had used multiple accounts to run far more test evaluations than the rules allowed. Baidu fired the team’s leader, who denied cheating.

The leaderboard scored outputs.

The input got its check in 2019, when an art project exposed bias in the person labels. Princeton and Stanford researchers then traced the bias to how ImageNet was built. By then Li co-directed Stanford’s new Institute for Human-Centered AI, promoting ethics for a field that trained on her dataset.

Nationality matters here as competition culture.

China has a name for its grind, neijuan, which young Chinese began using in 2020 for the hypercompetitive and often self-defeating chase after markers of success. American AI runs on its own win-at-all-cost version, and Li’s career runs through both. The headlines tie nationality to espionage. The motive, for what it is worth on explaining behavior, is competition.

Fei-Fei built massive influence and power on a faulty notion of success. Her dataset served biased person labels to the public for a decade before her own field checked them. Same poisoned ramp as Wyoming: win at all cost.