• daannii@lemmy.world
    link
    fedilink
    English
    arrow-up
    5
    ·
    16 hours ago

    If you create a statistical model based on biased data. That model will perpetuate bias.

    They tried to use AI to determine prison sentences for criminals.

    Not surprisingly, it gave black people more harsh sentences. - for same criminal histories and crimes, compared to whites.

    The creators argued that data showed blacks more likely to reoffend. But blacks also get harsher sentences first. And have a harder time getting jobs once out. More likely to be tried as adults for juvenile offenses. Not to mention less access to education/training/skill development.

    The system caused the problem. And now it sought to reinforce it even more.

    Herein lies the problem with prediction models. They use correlation data as if it’s cause-and-effect data. When it isn’t.

    Ai is only statistical models based on data that’s fed into it.

    A human decides what data is used.

    A human ultimately controls its output.

    It’s not objective. It’s not unbiased.