Een interessant boek over de werking van geautomatiseerde modellen en databases die ondanks de soms goede bedoelingen (grotere objectiviteit) en vaak dankzij de minder goede bedoelingen van kille kostenbesparingen automatisch leiden tot zeer ongewenste uitkomsten, vooral voor financieel zwakkeren in de maatschappij. De tekst is goed leesbaar en de gegeven voorbeelden zijn duidelijk en zeer illustratief.
"A model takes what we know and uses it to predict responses in various situations.
Models are opinions embedded in mathematics.
A signature quality of "Weapons of Math Destruction": when the model itself contributes to a toxic cycle and helps to sustain it [=> e.g. diverting police resources to a particular neighbourhood and due to that increased police presence automatically finding more instances of criminality which then seems to justify the inital diversion and results in even a larger diversion of police forces to the same area which results in higher crime figures, etc..].
Three elements of WMD: opacity, scale and damage.
Although some people benefit (when using supposedly impartial models rather than opiniated humans in selection or awarding processes), too many suffer from the consequences of using WMD's.
Big data processes codify the past and they do not invent the future. Human decison making can evolve; learning from mistakes and adapt. Automated systems do not, and changes depend on when they are reengineered. "
Aucun commentaire:
Enregistrer un commentaire