back to article Facial-recognition algos vary wildly, US Congress told, as politicians try to come up with new laws on advanced tech

A recent US government report investigating the accuracy of facial recognition systems across different demographic groups has sparked fresh questions on how the technology should be regulated. The House Committee on Oversight and Reform held a hearing to discuss the dossier and surrounding issues on Wednesday. “Despite the …

  1. johnrobyclayton

    There are no shortcuts


    They have 10000 different images of every person to be positively identified (scanned image is of this person)


    They have 10000 different images of every person that they want to negatively identify (scanned image is not of this person)


    They have 10000 different images of each person they are going to scan and attempt an identification on.


    They might have a chance of reliably identifying someone that they have scanned.

    It still will not be perfect though.

  2. Anonymous Coward
    Anonymous Coward

    Nothing to Hide nothing to Fear

    as Joseph Goebbels successors keep reminding us.

  3. ThatOne Silver badge


    Beware of those non-Caucasian, non-male persons: The proof they're up to no good is that they are trying to confuse our face recognition programs!

    (Sadly not a joke.)

    1. bombastic bob Silver badge

      Re: Unmasked

      YOURS is *PRECISELY* the kind of "feeling" (not THINKING) that we can expect from POLITICIANS that INJECT POLITICS into technology via STUPID-REGULATING it!!!

      The article sort of addressed this - you need sufficient TRAINING DATA. Otherwise, "all *INSERT DEMOGRAPHIC HERE* look the same" to the algorithm...

      What we *REALLY* need is an SJW detector, so we can stamp "un-hirable" on their applications... [thus avoiding MUCH legal expense and office drama that results FROM the SJW types]

      1. Fungus Bob

        Re: Unmasked

        Dude, there are many decaffeinated brands on the market that taste just as good as the real thing...

      2. ThatOne Silver badge

        Re: Unmasked

        > you need sufficient TRAINING DATA

        Unfortunately "needing" and "having" are not synonymous. Yes, you're right on this specific part, sufficient training data should do the trick, the only problem is nobody will bother, because "more/better training" translates as "more cost for the same profit", not to mention the time-to-market issue.

        As for the rest of your crazy rant, I'll leave you battling your windmills...

      3. Michael Wojcik Silver badge

        Re: Unmasked

        "Sufficient training data" is necessary (by definition), but not sufficient. Simply throwing more data at a model will often lead to over-fitting or other anomalies.

        There's a vast and rapidly-growing amount of research on this problem area, and forum posts from non-practitioners will not capture it in any useful fashion.

        Given unlimited resources - including time and expertise - it's possible to asymptotically approach perfect facial recognition. But resources are obviously not unlimited, and the practical limit on facial recognition for economically-viable use cases in the near future may remain mostly quite poor. Certainly in some problem domains, including the notorious job-applicant one, it appears to be little more than algorithmic dowsing.

  4. Mike 137 Silver badge

    "Most-accurate algorithms showed 'little to no bias', so nothing to fear, eh?"

    While I have deep sympathy for those affected by bias exhibited by facial recognition systems, bias is by no means the whole, or even the most significant problem. The biggest thing we should be afraid of (and resist to the utmost) is the virtual Panopticon - never being able to do anything unsupervised by covert unknown parties.

    CCTV already tracks almost all our movements in urban environments, but once an identity gets attached we will constantly live in fear of being challenged over any trivial thing the bureaucrats think up. The stifling effect of this (albeit accomplished by other means) can be seen in recent and current totalitarian states. Although these examples may seem extreme, it's inevitable that once the capacity is in place, ever more numerous uses will be found for it until we land in 1984 (but for everyone, not just the elite).

    1. Michael Wojcik Silver badge

      Re: "Most-accurate algorithms showed 'little to no bias', so nothing to fear, eh?"

      Yes, privacy is a major concern, as are attacks on group behavior. An argument can be made that crowd anonymity is a fundamental human right. I'm reminded of Poe's "The Man of the Crowd".1

      1I've always wondered if this story is the origin of the term "gumshoe" for a detective. A quick search didn't turn up an earlier use. I'd check the OED but I'm feeling lazy.

  5. Steven Guenther

    Already carry a tracker

    If you carry a phone, you are already being tracked. The govt is not allowed to track you, but they require the phone companies to hold the tracking data for them. So you ARE being tracked by Big Brother. The problem is when someone who looks like you does a crime close to your phone. Now the cops show up at your door and put the squeeze in you.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like