back to article Which job is AI going to eat next? Step forward, CCTV operators

NEC Corporation, one of Japan’s biggest IT providers, says it has built an AI that can rapidly search CCTV footage and spot a specific person out of a million or more faces. The application – snappily titled NeoFace Image data mining – can find wanted criminals, missing kids, and so on, all from video surveillance. We're told …

  1. Voland's right hand Silver badge

    suspicious == non-conformant

    Based on the description of the algorithm suspicious looks like anything which sticks out statistically out of the overall mass. This marks my entire family suspicious as we are not going to visit McDonalds or any of the other junk food outlets in the mall. There is further evidence of nefarious thought as the kids have walked past the Barbie and Star Wars parlor and ignored it.

    I will probably have to start training them in simulated conformance now.

    So, conform, citizen.

    Or have a discussion about conformance with Ed 209.

    1. Pascal Monett Silver badge

      One could eventually hope that these surveillance powers (need to re-watch that cracking documentary called Person Of Interest) will be used as they should, meaning to detect hoodie-wearing individuals loitering near a store for a while, then darting in and out and running away. For me, that is suspicious behavior. Actually, running away from a store is suspicious in itself, hoodie or not.

      The reality of this world, and the continued demonstration that police forces everywhere abuse the tools at their disposal with depressing regularity, means that these surveillance abilities will be used to allow a cop to follow his wife while she shops, along with anyone she appears to talk to. Alternately, some cops will certainly prefer to go crowd-surfing, picking some good-looking girl and following her just for visual gratification. In short, the system will mostly be used for anything except catching actual suspects - unless a draconian usage surveillance system is put in place around it. A system that would, for example, require a request form including a picture of the person to watch - form which would need approval in the system before the cop would be able to launch surveillance on that picture and that one only. During the procedure, other individuals could be tagged and additional requests made, each one having to await approval before being followed.

      But, of course, all that would create unacceptable delays for the one time in the year when the system will actually be used to follow a would-be terrorist, so no usage surveillance will ever be implemented.

      When I retire I will stay at home and get everything delivered.

    2. Magani
      Linux

      Re: suspicious == non-conformant

      Don't forget this is Japan we're talking about here.

      Guess where the saying "The nail that sticks out will be hammered down" originated?

      Penguinistas are also considered non-conformists.

      1. Anonymous Coward
        Anonymous Coward

        Re: suspicious == non-conformant

        As I recall, in China it's "the tallest tree will be cut first" in addition to the hammer-nail parable. I almost never leave home these days. Even though I'm known by almost evey cop in town (the good way), I'll take bets on how often the AI's will kick me out as an unknown /suspicious type.

        1. BebopWeBop

          Re: suspicious == non-conformant

          It is 'tall poppies' in Aus

  2. TRT Silver badge

    Meh.

    They've being doing this on NCIS for years.

  3. breakfast Silver badge

    I can see AI being good enough to match a human doing a shitty job at this kind of thing, but probably not someone with real skill. The problem being that most humans of skill start out doing a shitty job but get better with experience, so perhaps over time the outcome of AI in many of these currently human-occupied roles is that you have something cheaper than humans but the standard drops in general because the AI becomes the benchmark of doing the job.

    Also there is the whole thing about how AI is often trained with it's creators racial biases and you potentially end up with automated systems that might behave in ways that are effectively racist and who do you call out on that? The people running the systems? The people creating them? Establishing the human responsible would be an interesting challenge.

    1. Anonymous Coward
      Anonymous Coward

      As far as looking for a wanted criminal or missing child, it depends on how much of the AI's 1% left over from its 99% success rate is false positive versus false negative. If it is almost all false positive, which is typical with facial recognition, then it massively reduces the search space for humans to give it a look and see which leads are worth pursuing. A computer can search a day's footage on a whole city's camera network in the same amount of time it takes a CCTV operator to grab a coffee and a donut.

      That sort of use for facial recognition doesn't bother me, but stuff like "let's identify everyone we can so we can track their movements in public" and "let's let a computer make judgments over what sort of activities look 'suspicious' is where I draw the line!

      As far as racial profiling, it is a lot easier to program a computer to not be racist than it is to get a human to not be racist...

  4. allthecoolshortnamesweretaken

    Um, how is facial recognition "spotting suspicious behavior"?

    (Other than recognising known bad guys who may or may not be up to something?)

    1. diodesign (Written by Reg staff) Silver badge

      Re: allthecoolshortnamesweretaken

      >Um, how is facial recognition "spotting suspicious behavior"?

      Easily. Detect when certain people - or unidentified people - are seen hanging around a restricted area, for example.

      C.

  5. martinusher Silver badge

    Bit late for 'concern'

    This has been brewing up for some time, its just that now its ready for mass deployment.

    Think of it as ANPR for people. It will free everyone -- just as ANPR catches vehicles that aren't supposed to be on the road or are being driven inappropriately this person ANPR will be able to not just track individuals but locate anyone of interest (you *have* paid your taxes, haven't you?) and remember where they've been for the last 5 years (or however long we keep the data for). Collating individual sightings to determine associations, especially if combined with purchase history, searches and communications, could be quite the information bonanza for both commercial and governmental organizations.

    Oh, you're not happy with this? Something about privacy? Sorry, that's so last year.....

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