Demonstrators at the University of California, Berkeley, rally against funding cuts and attacks on free speech.Credit: Santiago Mejia/San Francisco Chronicle/Getty
In February, I learnt that a National Science Foundation grant funding my US laboratory, along with more than 3,000 other such grants, had been flagged for federal government scrutiny because, according to the US Senate Commerce Committee, it “promoted Diversity, Equity, and Inclusion (DEI) or advanced neo-Marxist class warfare propaganda”.
I confess I am not sure what “neo-Marxist” means. I am a mathematics nerd, not an ideologue. I started at Stanford University in California in 2009 as a physics major, entranced by the theoretical elegance of variational principles and Riemannian manifolds. What eventually drove me into my chosen field, artificial intelligence (AI), was not political. I learnt, at 20, that my mother, my grandfather and I all carried a genetic mutation, more common in people with Ashkenazi Jewish ancestry than in other ethnic groups, that confers a very high risk of cancer. As I struggled to cope with the bleak news, I found that one of the few things that made me feel better was a scientific paper that used an AI model to find overlooked clues for predicting survival1, suggesting entirely new directions for research.
This paper was a revelation to me; it gave me hope that I could protect my family, and myself, with AI. I wrote to the researcher who had led the study, Daphne Koller, asking if I could work in her lab. And when she wrote back, on New Year’s Day, I knew that AI research would define the rest of my life. Neo-Marxism didn’t come up.
Trump 2.0: an assault on science anywhere is an assault on science everywhere
AI models are extraordinary, but they can also fail for some groups of people: if a model is trained only on data from men, for example, it might perform poorly when looking at women. So an important component of my work is building AI models that work for everyone. Researchers in my field ensure that algorithms don’t underestimate cancer risks for Black people2; that speech-recognition systems understand you, whether you speak with a Texan accent or a New York accent3; that self-driving cars aren’t blind to pedestrians with darker skin4; and that health-care models perform well in both rural and urban parts of the United States5.
It’s this aspect of my work that probably got me branded by the Commerce Committee as a possible neo-Marxist. But polls show that many people share my concerns about AI bias, and hundreds of AI experts recently signed an open letter reaffirming this concern. More broadly, ensuring that technologies work for everyone isn’t a controversial idea: it’s why we test drugs on diverse populations and use crash-test dummies of different sizes.
I shall not be moved
In the weeks since my grant was flagged for scrutiny, many students and colleagues have asked me what I will do. Here are two things I won’t do.
