It seems lately I see an article like this about once a week - "an algorithm did it" and 1 person's livelihood in shambles, with said person needing to prove they're real/not a criminal/not violating policies/not fired/ etc.

Just off the top of my head - [1] https://news.ycombinator.com/item?id=17350645

We need to think critically about how to propel legislative action requiring every company/organization using machine learning to have a human-in-the-loop that can override the mistake. It doesn't matter how good an algorithm is - thanks to the nature of probability and a large enough sample size, it will always make a mistake. The onus to correct that mistake should not fall to the victim.

Currently, if, say, one's Google's account is suspended because of a mistake, there seems to be no recourse for a victim except to stir up a large enough PR scandal. In essence - a "guilty until proven innocent" approach. Our courts don't work that way and neither should the closure of our accounts (which are, increasingly, our livelihoods).