How a major AI conference is attempting to tackle its culture problem
The run up to one of the biggest artificial-intelligence (AI) conferences of 2018, held in early December in Montreal, Canada, was marred by controversy. After last year’s event, some people criticized the culture of the conference and other machine-learning meetings held in previous years; reports emerged of harassment, discrimination and an unwelcoming environment for women.
The members of the board of trustees and advisory board that oversee the Neural Information Processing Systems conference, previously known as NIPS, took several actions to try to improve the culture of the event. But a furious debate erupted after they initially decided not to rename the conference, despite complaints that the acronym was offensive. Almost 2,000 researchers signed a petition calling for a rethink and questioning the board’s choice to use a survey to guide the decision.
The board relented just weeks before the event kicked off, and adopted NeurIPS as its shortened name. Trustees and advisory boards had also put in place a number of measures for the meeting, including an updated code of conduct and the creation of diversity and inclusion chair posts on the organizing committee.
Machine-learning researchers Katherine Heller at Duke University in Durham, North Carolina, and Hal Daumé at the University of Maryland in College Park took on these roles for 2018, introducing measures that many attendees said have helped to improve the culture of the event. These included a town-hall meeting on diversity, workshops for people in under-represented groups, and gender-neutral bathrooms.
The two researchers share their experiences with Nature.
Katherine Heller and Hal Daumé.
How do you think the conference went this year?
Katherine Heller: It has gone very well. We have managed to accomplish many things for the first time this year — there is a long way to go, but hopefully we will get there.
Hal Daumé: The thing that I am the most happy about is that people are talking about this stuff. We have had little bits of positive feedback. Some people were excited that they got to wear a “first time” sticker. We got an e-mail from some folks who had been avoiding coming to the conference and came here for the first time this year. Nothing is perfect, but in general, it has been a pretty positive outcome.
How do you feel about the debate over the name of the conference?
K. H.: The name change was really a rallying point for people to express the fact that they think some of these diversity and inclusion issues are very important and have not gotten enough attention in the past. What you saw was that part of the community stepping forward to say, “Hey look, you really need to start paying attention to us and how we feel.”
H. D.: The board did decide to change the acronym and that is a reflection that they heard this. It is important for everyone to feel heard, and there is now an increased willingness to do that. It is a positive thing.
Was it frustrating, given your aims to improve inclusion, that the board was slow to respond to concerns about the name?
K. H.: It is safe to say that everyone is well intended. There are a lot of logistical issues that come out of this situation. There is a lack of transparency around the roles of the organizing committee, the board of trustees and the advisory board of NeurIPS. Not necessarily because anyone intended it; it’s something that most of the community just doesn’t have good sight of. So when the executive board made a decision about the name change, people would often look to the organizing committee and say, “Hey, why did that decision get made?” and we were like, “we had nothing to do with that decision or that process”.
What have been the biggest challenges to boosting inclusion and diversity?
K. H.: Communicating with people who are not used to this being an issue that is first and foremost on their plate. Trying to restructure how they think about diversity and inclusion, and where they expect to spend their time. The conference organization has to get used to this being an issue that they need to dedicate some amount of time and attention to.
H. D.: One of the challenges is the growth of the conference. If things were relatively stable, making course shifts would be a lot easier. At the moment, everyone is struggling to keep up and do these things at the same time. There are just not enough people to go around.
K. H.: There have also been other things. When you are dealing with a new position, you make a lot of people unhappy. There has been a lot of unhappiness created on all sides for all things that we have done — for example, the machine-learning ally pledge (a promise that attendees can make to take concrete actions to stop others from being oppressed at the conference). Everyone agreed with it, and everyone wanted to sign off on it, but nobody liked the wording of it and everybody would re-word it. So there were a lot of phone calls with people who were upset because they wanted something different to be going on, even if they agreed with the overall scheme.
H. D.: We have our own advisory board, which currently consists of four people. We probably should expand that because Kat and my experiences don’t reflect the experience of a significant fraction of attendees of this conference. Expanding our view has been important and I think we can do a lot more of that.
What is at stake if this progress doesn’t continue at the rate the community needs?
K. H.: We have seen progress in society as a whole on these issues. All of the people who are involved in the affinity groups have really felt under-represented in the community, and their voices have not been heard. I think it is really important that it starts to change now, and that diversity begins to be more integrated into the technical aspects of what we do. There is also an AI ethics component, so it affects what we put forward as a community, as well as how we behave and treat others. It is super important.