Predictive Policing Software Terrible At Predicting Crimes – The Markup
themarkup.org
A software company sold a New Jersey police department an algorithm that was right less than 1% of the time
  • Perroboc@lemmy.worldEnglish
    561·
    1 year ago

    That’s because you’re not engraving the suspects name in wooden balls based on the dreams of 3 people sleeping in some weird hot tubs.

  • flamingo_pinyata@sopuli.xyzEnglish
    32·
    1 year ago

    Less than 1%? Did they forget to flip a boolean condition?
    Like that’s worse than random, it’s worse than if you intentionally wanted to be wrong.

    • deranger@lemmy.worldEnglish
      112·
      1 year ago

      How do you figure that’s worse than random? Randomly attempting to predict crimes would likely be 0% accurate. I’m not supporting predictive policing at all, just curious what brought you to that conclusion.

      There are near infinite failure conditions and few successful conditions.

      • Edgelord_Of_Tomorrow@lemmy.worldEnglish
        3·
        1 year ago

        If you randomly selected a citizen as the culprit every time a crime was committed the only percentage of accuracy it wouldn’t be is 0%, because it’s inevitable you would be right at least once.

  • _haha_oh_wow_@sh.itjust.worksEnglish
    25·
    1 year ago

    Police are notorious for using bullshit tech to try and justify their “investigations”. Remember Voice Stress Analysis? Total bullshit, but thousands of departments bought into it. There are probably still innocent people in prison because of it.

  • hperrin@lemmy.worldEnglish
    221·
    1 year ago

    I know how they could make it thousands of times more accurate. Just rewrite it to always point at Wall Street.

  • ShaggySnacks@lemmy.myserv.oneEnglish
    17·
    1 year ago

    An algorithm needs good data, I would wager a bet that the Police are very good at keeping data that is racist and terrible.

  • profdc9@lemmy.worldEnglish
    193·
    1 year ago

    The police need crimes and criminals to justify their existence. If the criminals are selected by a computer program, that is sufficient for their purposes.

  • waterbogan@lemmy.worldEnglish
    5·
    1 year ago

    How did they manage to do so spectacularly badly? I think part of the problem is that they were trying to predict times and locations, rather than focusing on individual offenders. Past record is highly predicitive of future behaviour, i.e. if an offender has committed assault half a dozen times, it is highly probable that they will commit another assault or similar violent offence again, we just dont know when or where. Poor quality data may also be part of it - garbage in, garbage out