posted by Ana Rita Pena (2019 cohort)
The ACM Conference on Fairness, Accountability and Transparency (FAccT 2021) is an interdisciplinary conference with an interest in research on “ethical” socio-technical systems. Hosted entirely online the 2021 edition was the 4th edition of the conference, which started with fairly small in 2018 but has received a growing amount of interest in the last couple of editions.
The conference started on the 3rd of March with the Doctoral Colloquium, following with a Tutorial’s day (divided into three tracks: Technology; Philosophy/Law/Society and Practise) and a CRAFT day.
Before the consortium we were asked to prepare an informal presentation on our PhD work to present to the other participants in small groups. Having small breakout groups led to very engaging back and forth discussions on everyone’s work. Following on from that we had the choice of several discussion topics each in a different breakout room, the topics ranged from research interests to career advice to current world events. For the last activity of the consortium, we were divided into similar research interests and each group was allocated a mentor. The discussions we had ranged from understanding how all of the attendees’ research fitted together within a higher ecosystem to discussions on various approaches to incorporating our world/political views within our research. At times when focusing on our work it is easy to lose sight of the higher picture and even critically evaluate our own approach to our work, so being able to have a space to discuss it with a varied group of people working in a similar area was one of the most enriching experiences of the conference.
Another personal highlight of the conference was the CRAFT session “An Equality Opportunity: Combating Disability Discrimination in AI” which was presented by Lydia X. Z. Brown, Hannah Quay-de la Vallee, and Stan Adams (Center for Democracy & Technology). The CRAFT sessions are specifically designed to bring academics of different disciplines together to discuss current open problems. While algorithmic bias and discrimination regarding race and gender are more widely studied, disability bias has been severely understudied, this in part caused by the difficulty to summarise the varied disability spectrum in discrete labels. The session’s discussion was to imagine and think about possible ways to address disability bias, while still giving a voice to people with lived experiences.
After the weekend, there were three full days of paper presentations. Each day there was a panel session with a given topic followed up with the keynote. On day one the panel topic was “Health Inequities, Machine Learning, and the Covid Looking Glass “ followed by an excellent keynote by Yeshimabeit Milner from Data For Black Lives on Health, Technology, and Race (https://www.youtube.com/watch?v=CmaNsbB-bIo for the keynote video). The second day discussion was around the topics of the flaws of mathematical models of causality and fairness approaches. To end the conference on a bit of a more optimistic note the final discussions were possible future directions and the role of journalism and the importance of good journalism to audit algorithms and make them accountable to the public. The keynote speaker was Julia Angwin who was the first journalist to report on the COMPAS recidivism prediction tool bias. The COMPAS dataset bias was one of the issues that made the topic of algorithmic fairness gain some traction and that is still commonly used in the literature of Fairness in Machine Learning. Julia is currently in charge of The Markup, an independent and not-for-profit newsroom that focuses on data-driven journalism.
The different discussions enabled in the conference gave me some space to look at my own work and critically reflect on what I am doing, why I am doing it and the approach that I am taking, which is a conversation with myself that is still in process. It was not necessarily the very interesting research that was presented, but the deep discussions that had taken place that made my attendance of FAccT 2021 an enriching experience.
Here are some of my favourite papers of the conference:
Representativeness in Statistics, Politics, and Machine Learning
(https://dl.acm.org/doi/10.1145/3442188.3445872)
Epistemic values in feature importance methods: Lessons from feminist epistemology (https://dl.acm.org/doi/10.1145/3442188.3445943-)