Social responsibility
There are many ways to act responsibly as a researcher. I think this is one of the many places where the ‘boy scout rule’ is applicable:
Leave every place better than you found it.
In other words, when feeling overwhelmed by the challenge of doing everything right, just start with something and make it a little bit better. (Cultivate virtues.)
Some things us machine learning researchers can do:
- evaluate our models thoroughly to prevent exaggerated performance claims [1], [2], [3]
- follow the TRIPOD guidelines when reporting on clinical risk prediction models, publish model cards or datasheets for datasets
- collaborate with domain experts and stakeholders to create ML systems that generate meaningful impact, instead of advancing 1% on a benchmark
- understand whether a problem really requires (yet another) ML “solution”
- perform an algorithmic impact assessment [1], [2], [3]
- assess the climate impact of our ML projects, and support the Climate Change AI Initiative
- engage critically in discussions regarding ethical usage of AI/ML
Some things all researchers can do:
- be kind to yourself, your students, and your colleagues, and nurture a supportive and encouraging workplace for everyone
- work towards meaningful progress on important challenges, instead of working towards that next paper as an end in itself
- contribute to the flourishing of your research community by reviewing, organizing events, teaching, supervising, mentoring, volunteering, doing public outreach work, …
- make research openly accessible (paper, code), correct, and reproducible (if interested, see some very basic best practices for scientific software development that I wrote quite a while ago)
- aim for design justice by giving those who will be affected by our research a voice in it
- support the Scientists for future and, in particular, their I won’t do it under 1000km initiative
Let me know if you think something crucial is missing that I should add!