Patterico's Pontifications

8/13/2019

Snarky Comments are required on this Post

Filed under: Politics — DRJ @ 6:49 am



[Headlines from DRJ]

Amazon’s Face Recognition Falsely Matched 28 Members of Congress With Mugshots:

Amazon’s face surveillance technology is the target of growing opposition nationwide, and today, there are 28 more causes for concern. In a test the ACLU recently conducted of the facial recognition tool, called “Rekognition,” the software incorrectly matched 28 members of Congress, identifying them as other people who have been arrested for a crime. 

The members of Congress who were falsely matched with the mugshot database we used in the test include Republicans and Democrats, men and women, and legislators of all ages, from all across the country.

And at the LA Times — Facial recognition software mistook 1 in 5 California lawmakers for criminals, says ACLU:

Facial recognition software mistook 1 in 5 California lawmakers for criminals, says ACLU. In a recent test, facial recognition software incorrectly matched 26 California legislators with mug shots of people who had been arrested. California is considering banning such software from being used with police body cameras.

— DRJ

30 Responses to “Snarky Comments are required on this Post”

  1. Serious comments encouraged, too.

    DRJ (15874d)

  2. Gimmicks!

    The Illinois Secretary of State now takes biometric photos for drivers licenses and state IDs. It requires a special camera and a special program to code and store the image. That’s the real deal.

    Amazon’s gimmick is for entertainment purposes only.

    nk (dbc370)

  3. Note to the La Times: you misspelled “only identified” as “mistook”. That’s a big typo, and a big issue with facial recognition software in general.

    Not that voters are any better at recognizing them as criminals.

    PrincetonAl (0744e4)

  4. I read that they identified these Congressmen as criminals in adatabase because they put the setting at 80% accuracy, but at 99% none of them were matched.

    Sammy Finkelman (324ec1)

  5. If you want pure industrial-size snark you’re gonna need to do a Fredo Post.

    harkin (58d012)

  6. Are we sure it was falsely?

    Gryph (08c844)

  7. If you want pure industrial-size snark you’re gonna need to do a Fredo Post.

    harkin

    Heh. True. I have a feeling someone else will handle that.

    DRJ (15874d)

  8. Non-snarky response.

    The issue is, though, what would such an identification be used for? Obviously, not to convict someone. Perhaps probably cause to stop and question the person? Or arrest them (assuming there is an outstanding warrant)?

    So is this software more or less accurate than, say, human identification. Suppose someone was mugged, the cops interview her. Then the next day she tells the cops, “That’s the guy, he mugged me!” Clearly, the cops would be justified in arresting him. Has anyone done studies as to how accurate such an identification would be?

    Bored Lawyer (998177)

  9. “Clearly, the cops would be justified in arresting him”

    Questioning, yes. Not sure on arresting without more info.

    harkin (58d012)

  10. What’s the fuss all about?
    Never used Amazon.
    Don’t use Twitter.

    Welcome to 1998. 😉

    DCSCA (797bc0)

  11. I’d think these Caliunicornia legislators would be “prime” suspects even without the use of software…

    Colonel Haiku (2601c0)

  12. Could have been worse: some normal salt-of-the-earth American citizen might have been misidentified as a member of the United States Congress.

    JVW (54fd0b)

  13. “Questioning, yes. Not sure on arresting without more info.”

    The criminal law mavens can chime in. Not sure.

    My point is, human identification is also flawed, yet we do allow the police to rely on it for some purposes.

    Bored Lawyer (998177)

  14. Take the legislature members photo and enter it into the database. Swab them for DNA too.
    Notify the IRS that their net worth will be growing faster than their paycheck would seemingly allow and small businesses owned by family members will be miraculously booking profits in excess of revenues.
    PS: remember to check the office freezer for cash

    steveg (354706)

  15. Only 28? And was it really “incorrect”? Possibly just un-indicted coconspirators.

    Stacy0311 (3d63e6)

  16. More than 1000 public employee pensions in CA exceed IRS thresholds.
    https://www.sacbee.com/news/politics-government/the-state-worker/article233539467.html

    Long past time to bust the public employee unions

    steveg (354706)

  17. 16… amen! If Caliunicornians don’t understand that public employee retirees will get their pensions one way or another, they are sadly mistaken. The State will be paid, even if takes force of arms.

    Colonel Haiku (2601c0)

  18. the crunchy irony, is that ana Navarro, rick Wilson, and a host of other figures, have been crying ‘fredo in a public theatre,’ for years,

    narciso (d1f714)

  19. They need to fix their software. How did they miss the other 80%?

    NJRob (c6d8e8)

  20. The last I had anything to do with facial recognition software, which was about ten years ago or so, the faces it matched, based on ratios of distance between eyes, bottoms of ears, nose, etc. was very rough. Many of the faces that would match would appear to the human eye to be nothing like the source picture/face. Though at the time I’m pretty sure we didn’t have skin tone included, partly because of the limitations of the project (B&W photos, IIRC). Even for people of similar skin tone, the high probability matches were often head scratchers. But it was useful for narrowing down potential suspect lists. I gather from what I see on Facebook it’s much better today, but that is still working with a smaller pool of potential faces. Going out three degrees/links (or wth it’s called) from a source subject is still a much, much smaller probability pool. The CSI style TV shows set expectations way too high. The main purpose of these things (they were working on butt-impression matches last I heard) is to be used in conjunction with other biometric or non-biometric information.

    PTw (894877)

  21. Several good serious comments here, plus some good snark. Well done.

    DRJ (15874d)

  22. It’s like in Minority Report. Amazon has mastered precog.

    Paul Montagu (a2342d)

  23. 22, that’s good precedent, seeing as Trump sat out one of the Iowa debates to go to a veterans event across the street, this is actually being a veteran.

    urbanleftbehind (5eecdb)

  24. What was the “Born in East LA” % from Mississippi last week?

    urbanleftbehind (5eecdb)

  25. 20. PTw (894877) — 8/13/2019 @ 10:06 am

    I gather from what I see on Facebook it’s much better today,

    it was good enough, in 2011, for the Syrian government to fiind (at least some demonstrators (matching faces aaginst the database of state identity cards) aand arrest, torture, and kill some of them. This was one thing that really sparked the Syrian Civil War.

    Sammy Finkelman (324ec1)

  26. well the debates are ‘sound and fury, signifying nothing told by chuck todd (see what I did there)
    also Stefan halper says his bureau contract entitles him to lie, and greg craigs trial get pushed back, you thought there was any justice in the world ‘those words you are using;

    narciso (d1f714)

  27. I saw snarky comments and assumed this was the Fredo Cuomo thread. My bad.

    rcocean (1a839e)

  28. The implication of this complaint is that the use of such software and databases would be ok if it were accurate, and that it’s the inaccuracy which is the problem.

    But i’m not convinced it would be ok if it were accurate.

    aphrael (e0cdc9)

  29. I was a programmer for 35 years before I had to quit in 2015 because of Parkinson’s. Not only do I not trust self driving cars, I do not trust computers either. Just as people make mistakes, programs are written by people and they make mistakes too. I also believe people are fallible too, especially when they do not follow procedure as in this case where an innocent woman was arrested for having the same name as a criminal and not even living in the same area.

    ACLU lawyers file suit on behalf of innocent woman wrongfully arrested, jailed, due to Denver cop

    I remember a similar case that because Google search results are really bad now, I couldn’t find. In that case a woman was pulled over for a traffic stop and wrongly arrested because her name was the same name as a criminal. If they had simply compared pictures they would’ve seen that she did not look anything like the criminal.

    Tanny O'Haley (8a06bc)

  30. it was good enough, in 2011, for the Syrian government to fiind (at least some demonstrators (matching faces aaginst the database of state identity cards) aand arrest, torture, and kill some of them. This was one thing that really sparked the Syrian Civil War.

    The implication of this complaint is that the use of such software and databases would be ok if it were accurate, and that it’s the inaccuracy which is the problem.

    But i’m not convinced it would be ok if it were accurate.

    Not sure you’re understanding my point…though maybe. Facial recognition software is a tool. It is not something one should be using to definitively identify someone. It is useful for example if you have a picture of a specific someone, a specific suspect from a crime scene, or in the case of Syria, a culled subset of people at a demonstration. You can run that ONE (or two or three or so) picture against a database of millions. For each suspect/latent/crime scene picture, depending on how you may have finetuned your matcher, you get say five or six possible matches. To the human eye, the matches may very well look significantly different. You then have a human review those possible matches and throw out the obvious (to the human eye) misses. Maybe even all are misses. But likely you have something that looks right. But you don’t use this information to CONVICT anyone of a crime. You must then do further leg work to find more corroborating evidence, perhaps other biometrics such as fingerprints, credit card receipts, cell phone pings, whatever. You use it to narrow down your pool of suspects to better focus your investigation. I doubt one would ever use FRS in court, though perhaps. It would seem rather weak. But as an investigation tool, it is exceptionally useful.

    PTw (894877)

Leave a Reply

Comment moderation is enabled. Your comment may take some time to appear.


Powered by WordPress.

Page loaded in: 0.2318 secs.