Raw LLM Responses

> So black people didn't reoffend at a higher rate, yet the AI still developed a bias? Am I reading you right? Not quite. "Re-offence rate" is actually impossible to know: the good criminals re-offend without getting caught; and who gets caught is also affected by where the police are looking. So for example if black people get longer terms of probation than white people, then they're likelier to be caught re-offending because a probation officer is watching them. And if cops are more likely to let white offenders go with a warning, then white re-offenders are less likely to be counted in the system. And if black neighborhoods are policed more intensively -- for example, if police have a "stop and frisk" policy that they carry out mostly on black people -- then again, black offenders are more likely to be caught than white ones. We actually _do_ have data indicating that black and white people offend at similar rates, so as far as we can tell for certain, you're right: black people don't re-offend at higher rates. But the way the system works today, black people are more likely to be caught re-offending, and the AI is fed data from the justice system. So the AI, like many Republicans, sincerely believes that black people re-offend at higher rates.