The Atari to ethics pipeline

A robot staning in front of a stack of atari games. The robot is holding a fan saying "gay af" in rainbow colours

Most success stories start with two tech bros in a garage. This one starts with a group of queers at a conference. „NeurIPS 2017“, Willie says „I was a little first-year grad student.'' We are on a Zoom call and it’s 10 am on his side and 7 pm on mine. Willie has a looming deadline and doesn’t know what being fully rested even feels like anymore. Once the paper is submitted it’ll be 8 hours of sleep again, but for now, the priority is getting everything done in time. I am all the more grateful that he found the time to sit down and talk about the beginnings of Queer in AI. „They had the Whova app“, he says „which is like a forum. I just made a „Queer at NuerIPS” forum thing and people started posting and doing introductions. Eduardo did a little coffee meetup. And we were like: we have to do more things, this is so great, this is so wonderful.”

This first meeting at NeurIPS became a crystallization point. The time between NeurIPS 2017 and Neurips 2018 Willie spent collaborating with the initial group: emailing, meeting, and working on documents, not only to plan future activities but also to establish what form Queer in AI should take. „At NeurIPS 2018 we had our first workshop“, Willie says „At the time NeurIPS didn’t give us a venue so we rented a hotel across the street. We were also novices at workshop organizing, and it was 8 am on the last day of the conference, so it was rough for everyone to be there, and it was rough for me, too. We had breakfast, we had speakers, and we had panelists. It was really exciting seeing all the people there, navigating through the hotel. A lot of people I’ve met there I have been seeing for years and years afterward. It was a very foundational moment.“

Shortly after the workshop, the Slack workspace came into being. „It’s our only way to function since we’re so spread out.“ Willie sweeps his arms around. „Everyone is in different time zones and busy with different things all the time. It’s a very nice asynchronous space where we can communicate effectively.“With this tool in place, growth happened quickly. 2019 saw Queer in AI events at most major AI conferences. The organizational structure of Queer in AI had to change to accommodate that growth. „We started as a petty standard organization: There were officers, and presidents and vice presidents, a person for finances“, Willie says „But then we figured out that it doesn’t make sense to the way we function. There are just too many events and too many different people and communities and it doesn’t make sense to say that one person is in charge of it. Going into 2020 we got rid of every role except for core organizer.“

Getting rid of hierarchical titles fits better with the earth-spanning nature of the community and it also opens the door for people to approach organizing faster. „If someone wants to organize a social event at AAAI, we can say: you can have money, here are guides, here are people who can mentor you, but really, go and do whatever you want. We have standard practices and a code of conduct regarding the kind of inclusion we are looking for in our speakers. But this has been the only way we have been able to handle twelve-plus events every year all over the world.“ Willie sees learning to decentralize organizing as one of his biggest lessons from Queer in AI. It requires trusting in everyone involved and also tolerating tensions. And full decentralization is not entirely possible: People can request money, but it still involves going through the sheet and sending it via PayPal. Willie's Name on the Slack channel is ‚Willie from accounting‘, poking fun at this last element of a traditional structure in an otherwise flat collective. „These points of centralization do produce tension,” Willie says „But that is where real insight and learning comes from”.

While Willie is still a Ph.D. student at the University of Washington, many things have changed since NurIPS 2017. The growth of Queer in AI is mirrored in his academic trajectory „I started my Ph.D. with a pretty uncritical view of AI“, he says. He worked on object-oriented reinforcement learning for Atari, a problem he now calls „narrow and unrelated to the real world“. „I was convinced that that is the way to do good in the world, to do at Atari better. I think with a lot of people who started machine learning at that time, you can see the same journey again and again. They start in a very technical area and gradually become an AI ethics person.“ While the roots that AI ethics draws upon go back in time, the field itself is quite new and Willie feels good about the role Queer in AI played to support its emergence: hosting workshops and paying speakers, especially because Queer in AI’s work also shows the harm that can be done to queer people by AI systems.

Willie now sees his research and advocacy more broadly, both of them interlocking and overlapping in different ways. „Existing questions that I am interested in now are model ownership and data sovereignty, and are definitely inspired by queer in AI”, he says „In a lot of cases queer people don’t decide how they are represented in models and databases. You should always be able to own your data, own your identity.“ For his academic future, Willie wants to stay on that track: „Radical AI ethics is a place where there aren't a lot of people at the moment. Being a professor, establishing a lab, creating a place where Ph.D. students can come to get training, get mentorship, write good papers, a place where postdocs and visiting researchers can come - that would be very helpful to the broader community. To support people and to have this professorship title that would allow me to get these big grants.”

Achieving this next step would allow his research work and Queer in AI to come even closer. „It would help in making this organization more sustainable in terms of making it more rewarding to the volunteers, to give them the concrete professional benefit of, here is a paper. A lot is going on with queer data, a lot is going on with queer models and algorithms using that data. I am very excited to take all of the expertise we have in Queer in AI and apply it to these problems, to develop frameworks and models for queer data ownership, and what governance of queer data collectives could look like. Those are problems that I am extremely excited about.”

Additionally to being a research hub, Queer in AI fulfills other functions in Willie's life, too. Queer in AI came about out of a sense of isolation for him, especially coming from Georgia Tech, where Willie didn't know a lot of queer people and where there weren't any out queer professors. He hopes that for queer people entering the AI field Queer in AI can be the resource that his younger self needed. „Another big thing is that it’s keeping me grounded.”, says Willie. „AI is a wild wild place. Start-ups are spending 10 Million dollars training a slightly better model and they think that it will revolutionize the world. Especially being involved in our aid programs and seeing: 500 dollars is going to transform this person's life, it will let them apply to and get into an amazing grad school. Staying grounded in that amount of money and that amount of aid really helps you see how strange some things are that are going on in AI. I am very grateful for this for keeping at least part of me in the real world.“


You can find more of Willie’s research here


A picture of a white person wearing a blue and white patterned shirt

This post was written by Sabine Weber. Sabine is a queer person who just finished their PhD at the University of Edinburgh. They are interested in multilingual NLP, AI ethics, science communication and art. They organized Queer in AI socials and were one of the Social Chairs at NAACL 2021. You can find them on twitter as @multilingual_s

Previous
Previous

The opening paragraph of your paper sucks, and here is why.