Introductin to privacy AI
I always though that existing privacy laws were struggling to regulate AI not necessarily because AI wasn’t expected (machine learning isn’t a new thing) but because the rapid evolution of AI is exposing many of the shortcomings of existing privacy laws.
This article from Daniel Solove is an amazing introduction to the intersection of AI and privacy:
- It will outline privacy issues were likely bothering you but that couldn’t quite put your finger on
- It will help you better frame the issues/risks of existing laws and potential paths forward
There are so many interesting pieces of information in the article – I ended up take 3 pages of notes 😅. I’ll likely write more about this later, but even though this is a 60 pages long article, I strongly recommend everyone to read it whether it’s because
- You are interested in privacy
- You are a software engineer, research scientist, product manager etc. who work on AI
- You want to have an educated opinion on AI and/or become an active voice necessary to create an accountable governance model
Here is just one single example where the article resonated well with my experience: trying to build an unbiased model by removing sensitive information (e.g. gender, religion etc.) from its input doesn’t work because AI models today are able to infer the missing sensitive data (see for example “The Input Fallacy” by Talia Gillis). These unbiased models end up with real consequences that we are not well equipped to manage/control today.
If you did end up reading the article, I would be curious to hear what resonated with or puzzled you the most.