Recently, deep learning researcher Yann LeCun wrote a tweet that sparked an important debate in the academic community. As someone who does research and development in the industry, I can say that this discussion reaches beyond academia and affects all of us.
At its core, it started out as a debate between whether a model’s architecture (and related design decisions) or the data its trained on ought to be the object of our attention when talking about biased results in the real world. Over time, however, it became a referendum on how this debate is carried out and the way in which Professor LeCun communicates with those who disagree with him in general.
This is especially true because of the statements that Professor LeCun made about the responsibilities of researchers vs. engineers:
The consequences of bias are considerably more dire in a deployed product than in an academic paper. - @ylecun, 2020/06/21
Here is a thorough breakdown of the main spine of the discussion.
It’s a well-written post that’s worth a read. It presents multiple sides and still stays focused on the main throughline: what are our responsibilities as technologists and how do we conduct ourselves to inflict the least amount of harm on each other and the world? It is still absolutely an open question, and I’m not sure when we’ll get a satisfying answer. Our field is comparatively young, so I am hopeful that we will gain enough maturity so we won’t have to keep having these conversations.