DALI 2019: Fairness and Transparency

Starting the year early, I had the privilege of hosting a session on Fairness and Transparency at the DALI meeting in George in January 2019. The session was co-organised with Moustapha Cisse.


My personal motivations for the session was an interest in the transparency side and how that could also lead to fairness. So how do we get people (who are not us the ML/DS community) to understand what it is we are doing. What are the models doing and how are they making decisions? How does this ultimately impact the decision making of others?

For this session we had 4 invited talks and the schedule is shown below.


10:00- 10:15 Opening Remarks Vukosi Marivate, University of Pretoria

10:15- 11:00 Interpreting machine learning using examples. Samni Koyejo, University of Illinois at Urbana-Champaign

11:30- 12:15 Responsible Belief as a Moral Obligation and the Quest for Fair and Transparent Machine Learning Emma Ruttkamp-Bloem, University of Pretoria

12:15- 13:00 Reliable Decision Support? Suchi Saria, JHU

17:00- 17:45 The wrong tools for the wrong job—damage control in a world awash with premature ML Zack Lipton, CMU


Slides are available here in a folder.

Some DALI off-siting. Our Jurassic park moments.


The session went well and generated good discussion (some bias here) but there were a number of thoughts I did come away with. Each presenter did challenge us in one way or another in our thinking.

  1. Explaining our input data is a good path to explaining what ultimately our models will learn (or not learn).
  2. Yes, our data is biased, but there is some power in knowing that. We can use the models to spot the bias and work on other fixes or feedback to society.
  3. What ultimately is our goal in our ML/DS work, is it just writing the next paper or making impact on society? The work that needs to be done for the latter needs to find a way to also make it into mainstream.
  4. It would be very interesting to explore these ideas in the NLP space and also as ways to make it easier to have conversations with public policy people about what their data means (Wink wink, look at Osonde Osoba's work)

There will be a number of these threads that I will be unravelling as the year goes on so looking forward to an interesting 2019!!!

Leave a Reply

Your email address will not be published. Required fields are marked *


This site uses Akismet to reduce spam. Learn how your comment data is processed.