Algorithmic Choice
Twitter launched their responsible machine learning initiative.
I like that they give concrete examples of projects they are planning and that they promise to share their insights (as well as potential failures!) A company like Twitter has the potential to be a leading example in how general promises of "AI transparency" can be concretized and operationalized. Or... where the limits of this attempt are.
They also mention the term "algorithmic choice":
We’re also building explainable ML solutions so you can better understand our algorithms, what informs them, and how they impact what you see on Twitter. Similarly, algorithmic choice will allow people to have more input and control in shaping what they want Twitter to be for them. We’re currently in the early stages of exploring this and will share more soon.
What I find interesting here is that this is a particular way of answering the question (here implicit): "Why do we want our ML to be explainable?"
Is explainability a core value in itself?
I'd say explainability is instrumental to some value we are trying to uphold.
The mention of algorithmic choice is very interesting in this regard, because this suggests that we do not just want explainability to explain AI-based decisions that affect people. The point of making ML explainable is to give people more control over AI, i.e. it is instrumental to respecting human autonomy and a step towards more responsible AI.
TODO: is that a common terminology?
They list as the responsible ML pillars:
- Taking responsibility for our algorithmic decisions
- Equity and fairness of outcomes
- Transparency about our decisions and how we arrived at them
- Enabling agency and algorithmic choice