I have written previously about the need to regulate the use of facial recognition technology. The calls for regulation have become stronger in the wake of Kashmir Hill’s New York Times article about Clearview AI. It is extremely important to get this right. Here are a few more thoughts based on the article to add to my prior post.

First, it should be clear by now that it has become almost trivial to build a system like this. A lot of open source frameworks and neural networks have been made available that can be trained for face recognition. Clearview AI did not have to come up with some technological breakthrough, they just had to point existing technology at image sources (web scraping is also widely available).

Second, not surprisingly, the article mentions instances of the system being used successfully. Recognizing who someone is constitutes an essential part of investigative police work. If you live in New York, as I do, hardly a day goes by where there isn’t video footage of a crime such as a robbery show on television with a call to the public to identify a face.

What then does this mean? We have technology that can easily be built and that can readily help with legitimate police investigations. This are the good use cases. Then there are of course tons of ways to abuse this very same technology. For instance a police officer might use it to track down whom a former girlfriend (or boyfriend) is now dating. And by now we also have overwhelming evidence that the technology in its present state has big bias problems.

So the challenge is can we find a way to improve and apply this technology that maximizes its positive uses and minimizes its abuses? I believe this is possible. I also believe this will take time and deliberation. One idea is to set up a new agency specifically to deal with these kind of technologies used for domestic law enforcement. This could be at the state level. The law creating such an agency should focus on protecting due process (e.g. requiring a real court order, not something secret like FISA) and should set a high bar for transparency, such as how many requests were made in a time period as well as accuracy.

I am suggesting this because the existing agencies, including police departments, have all been set up before this technology has been available and have no legal safeguards and processes built in. Retrofitting them will be nearly impossible. And we shouldn’t try. We have seen what happens when we give new unchecked capabilities to police departments in recent years with equipping ever more police departments with SWAT teams.

PS What to do about individuals having facial recognition apps on their phones is a whole different story that I will write about separately. The reason to draw a clear distinction here is that individuals do not have the power of the state.