I began conducting research at Columbia University in the beginning of 2021.
Here are my regrets.
I used natural language processing to deduplicate medical data processed by my school's medical center.
I designed pivot tables that could process 1,000,000 records at once because we had a lot of medical data to process but a slow user interface.
I came into research with the assumption that my work would be free and open to all. My goal was, and still is, to use AI to grow and nurture global communities.
I became attracted to AI - deep learning in particular - by playing with deep generative models and reading illustrated blog posts about machine learning. I did, and still do believe, that AI can promote creative self-expression.
I became interested in federated learning because of its seemingly humane goals. Federated learning respects the privacy of the user's data that is used to train a model.
I want to use AI to help people. My research from 2021 hurt people.
In the process of labeling and training on medical data, I reinforced social constructs like gender and race to a disgusting extent. I was disgusted during team meetings where we would discuss using (in my mind) somewhat derogatory labels for patient data, but I held my tongue.
The results of the work I performed for my university are not open to the public. The code is very much closed-source.
Today, it is December 17 of 2021. I want to talk about the ideas I have for the rest of this year and the next.
Mary Grey says that hope stretches the limits of what is possible. I want to make humane AI possible.