Thread · 4 tweets · 10 Oct 2022

This is probably not the answer you're looking for, but in the case of Meta… I'm not sure anyone knows. Including Meta. I'm not being flippant, the Vice doc from not long ago says something very similar. vice.com/en/article/akv…

To be frank, and without detracting from Meta's turpitude, this is exactly what you would expect from experience with software/data projects at a large and complex company, especially a fast-growing one. It's not Meta-specific.

That's why I put very little stock in the ability of procedural requirements like Art30 or DPIAs to tidy things up. That's sort of an orderly imagined view of data processes as basically very big spreadsheets. The reality is that computer things become very chaotic very fast.

The only way this gets sorted is, in most parts, by preventing collection. Once it's in the system, you have to rely on provenance, lineage, and an understanding of processing purpose. That gets hairy faster than most can manage.