AI Researchers Fight Over Four Letters: NIPS

The future of humanity will be shaped by artificial intelligence. Now some of the best brains working on the technology are riven by a debate about a four-letter acronym that some say contributes to the field's well-documented diversity problems.

NIPS is the name of AI’s most prominent conference, a venue for machine learning research formally known as the Annual Conference on Neural Information Processing Systems. Researchers at tech companies including Google and leading universities allege that the name contributes to an atmosphere unwelcoming to women. The acronym has long inspired anatomical jokes about nipples; others dislike the word’s racist connotations.

“It encourages juvenile behavior and it’s a distraction from the science we’re trying to do,” says Anima Anandkumar, a Caltech professor and director of research at chipmaker Nvidia. Late Wednesday she tweeted a link to an online petition asking the NIPS board to rename the event. There are now more than 800 signatories, including researchers at Amazon, Microsoft, and Google.

They include Jeff Dean, who as Google’s top AI boss leads one of the world’s largest and most influential AI research groups, dubbed Google Brain. Dean told WIRED Thursday that he plans to raise his concerns with members of the NIPS board when he gets the chance. “I think enough people are made to feel uncomfortable by the current name that the NIPS board should change the name,” he tweeted Wednesday.

University of Washington grad student Maarten Sap’s attempt to follow the brouhaha helped illustrate the complaints from Dean and others: When he plugged “nips” into Twitter’s search function, it led him to pornographic tweets.

AI is projected to reshape everything from health care to war, but the community of people working on the technology is markedly different from the society it is supposed to serve. WIRED and startup Element AI found that in recent years at NIPS and two other leading academic conferences, only about 12 percent of people presenting work were women. That suggests the field is even less diverse than the notoriously monocultural tech industry. Some researchers fear this raises the risk of incidents like those in which image recognition systems have been found to have skewed views of women or black people.

The name NIPS became a flashpoint in the debate over how to make AI more inclusive earlier this year after complaints about sexist behavior at the conference, which takes place each December. A blog post about an on-stage remark about sexual harassment at last year’s conference led to investigations into accusations of physical harassment by Google’s director of statistics research Steven Scott, and University of Minnesota professor Brad Carlin. Both subsequently left their their jobs.

That incident helped inspire an open letter in March from 112 Johns Hopkins University faculty asking the NIPS board to seek a name less “vulnerable to sexual puns.” In April the board said it would consider alternatives, and in August it surveyed attendees from the past five years on crowdsourced options including SNIPS or ICOLS.

Monday, the board announced that it would not be changing the name after all, prompting the new round of protests. “The poll itself did not yield a clear consensus on a name change or a well-regarded alternative name,” a board statement said. Of the 2,270 respondents to the poll, 294 identified themselves as female. Overall, roughly 30 percent of respondents wanted to change the name; among those who identified as female, the proportion was 44 percent. In a statement on Friday, Terrence Sejnowski, president of the NIPS board and a professor at the Salk Institute, said the group had not ruled out reconsidering the name at some point in the future. "However the board wants to focus on taking substantive actions to make the conference more inclusive, and it welcomes suggestions for other actions," he said.¹

Elana Fertig, an associate professor at Hopkins, can take some credit for forcing the NIPS board to consider the issue. She helped organize the Hopkins letter asking for the name change, after a grad student confided that she wanted to attend NIPS, but was concerned about the atmosphere there for women.

In an interview Thursday, Fertig questioned the conference board’s decision to survey only past attendees of the event. Surveying people who had been going to NIPS already isn’t a good way to help those who feel excluded from the community, she says.

Fertig and others unhappy with the name say there has been other progress in making NIPS and the wider field more inclusive. After last year’s conference, University of Maryland professor Hal Daumé pledged to boycott the event if it didn’t clean up its act. This year he’s one of two new “diversity and inclusion” chairs on the organizing committee, which has introduced a detailed new code of conduct, and support for childcare. “We’ve made some positive steps this year although there are a bunch more to go,” says Katherine Heller, a Duke professor who is the second diversity and inclusion chair, and has also spoken out about sexual harassment in the field. She and Daumé both say they “care a lot” about the name of the conference, but declined to elaborate.

Anandkumar hopes she and her fellow protestors can convince the NIPS board to reconsider its decision to let the current acronym stand, and begin a broader program of consultation. “It reflects on the deeper issues,” she says. “If we can’t resolve even a simple name change how can we deal with the more serious ones?”

Others who wish the acronym had been retired are doubtful the board will change course in the short term. “Hopefully we’ll revisit this at another time,” says Allison Chaney, a postdoctoral researcher at Princeton, who will present work at this year’s NIPS conference, which takes place in Montreal in December. “For now we can make the best efforts we can to make NIPS and other machine learning conferences as inclusive as possible.”

Fertig at Johns Hopkins says there’s plenty else to do. Her experience in biological research suggests that machine learning lags other areas of science in welcoming women and other groups that have historically found it hard to advance in science and technology. “The conversations there are just leaps and bounds beyond the conversations we’re having in machine learning,” she says.

¹ UPDATED, Oct. 26, 1PM ET: This article has been updated to include comment from Terrence Sejnowski, president of the NIPS board.

• How the US fought China's cybertheft—with a Chinese spy
• Robocars could make humans unhealthier than ever
• Turning California’s weed into the champagne of cannabis
• Welcome to Voldemorting, the ultimate SEO dis
• PHOTOS: From Mars, Pennsylvania to the Red Planet
• Get even more of our inside scoops with our weekly Backchannel newsletter