Thread by @ClareAngelyn

I’m honored to release a new report with @GeorgetownCPT “A Forensic Without the Science: Face Recognition in U.S. Criminal Investigations” 1/n
forensicwithoutscience.org

This report is a very long look at a very basic question: How reliable is a police face recognition search? In other words: How often is a search right—or wrong? 2/n

It’s a vital question to ask and answer. Face recognition has been used in policing since 2001. And in the 7+ years researching this tech, we’ve begun seeing a pattern of people arrested primarily—if not solely—on the basis of a face rec match. 3/n

https://twitter.com/GeorgetownCPT/status/1600156731039838208

Nijeer Parks was arrested in NJ based on a #FaceRecognition lead plus a non-witness officer “confirming” the match. He was innocent. 4/n
nytimes.com/2020/12/29/tec…

Same thing happened in 2020 in another jurisdiction (more details in the paper):

Michael Oliver was wrongfully arrested in MI based on a #FaceRecognition lead and a faulty eyewitness ID.
freep.com/story/news/loc… 6/n

NYPD arrested a man in 2018 based on a #FaceRecognition lead & a photo sent to a witness w/ the text: “Is this the guy you know from going into the store many times before?”
7/n

And so on. In sum, it appears that in at least some cases police have assumed face recognition search results are highly reliable. But are they? 8/n

::Sidebar:: I get some pushback on using these examples as evidence of a problem—
1) These are just anecdotes, not quantitative data
2) Out of 10,000s cases, this is a small error %
3) This is not a #FaceRecognition problem—it’s an investigation problem. Etc.
9/n

1) Ok in one sense, yes, these are anecdotes. Due to a persistent absence of transparency around #FaceRecognition use, we quite simply do not have quantitative data about its use. No one does. This is… not a good thing.
perpetuallineup.org/report 10/n

1) But in another sense no, these are not just anecdotes. These are real people whose rights, liberties, lives have been directly, negatively impacted by the technology.

https://twitter.com/ClareAngelyn/status/1394730509977198600

2) This is not an experiment—this is the real world. The Constitution is absolute in its grant of rights & liberties. It doesn’t allow for an acceptable number of people wrongfully arrested. That’s not how this works (or should). 12/n

3) This is a #FaceRecognition AND a police investigative problem. These cases stem from a lack of understanding how the technology should—or should not—be used in policing. This is the heart of what this paper is exploring.
forensicwithoutscience.org 13/n

Spoiler alert. This report doesn’t have a definitive, empirical answer to this question. But after an intensive review of police use of the tech and CS, psych & forensic science literature, it concludes #FaceRecognition leads should not by default be considered reliable. 14/n

Why? For a lot of reasons—it’s why this paper is so long.
15/n

Oh you want the TL;DR? That’s fair
16/n

Police use of #FaceRecognition is a due process issue. This is about the right to a fair defense—for 10,000s or 100,000s of people since 2001 and probably 100,000s of people moving forward. 17/n

Police use of #FaceRecognition is a forensic science issue. We have a moment—a brief one—to prevent the judicial certification of an unproven forensic technique. We should take it. 18/n

Police use of #FaceRecognition is a policing issue. Why are we ok with relying on a tool whose reliability we don’t know? 19/n

Police use of #FaceRecognition is a civil rights, liberties, and justice issue. We should ALL *want* to prevent what happened to Michael Oliver, Robert Williams, Nijeer Parks, and others from happening again.

That’s all for now. Happy to answer questions! And I’ll dig into some of the other parts of the research—like the inherent biases of faces as biometrics—later on. In the meantime, the paper is here! /fin
forensicwithoutscience.org