Rudimentary implementation of algorithm development. Far too little information provided to be a real world comparison. The general idea is presented however, liberal bias is very heavy handed. More about political positioning rather than scientific absolution in algorithmic efficiency and accuracy.
I did a fair first playthrough then decided to do some unfair ones to see how good/bad the algorithim is, when I kept giving maximum semtences without even looking at the full profile it said I had biases which made no sense. The idea is good but the execution of it wasn't the best.
This game says that it is removing the irrelevant features but it removes the features that I use the most, then at the end says that I biased to the three things that were left (age, education and offense) even though I never even looked at two of them. On the second playthrough I only looked at prior felonys until that went away then I just swiped on offense and at the end it said that I was biased on (offence, violent felony, and employment) I did not use any of those
Ive got a legitimate complaint about this, first of all the algorithm set forth is based upon the information YOU as the creator provided to the characters submitted therefore it doesn't even matter what I sentence the result will be the same! Black and Hispanic judged harsher and Caucasian judged lighter, I played 5 times with different ways 1.sympathetic 2.cruel 3.factual 4. Harsh and 5. Random ALL came out the same black and Hispanic judged harsh and whites judged less. So I'm only led to believe that this game is a blatant propaganda attempt
Ironic a game devised to show bias could have such a blatant agenda/bias itself. It listed my biases as prior violent offenses (1 or more tended to get heavier sentence), more or less even treatment, but still detected bias on offense type (I tended to not sentence drug crimes as much), and more or less even treatment on race, which I assume was statistical noise. Here is the thing though: 1. Real world learning algorithms can be TOLD what inputs to use, so you won't get things like age, gender, race bias. 2. 50 trials is a pathetically small data sample, and not enough to extrapolate into a strong algorithm. 3. The simulation is oversimplified.
The algorithm was not very accurate but I feel that maybe that was the point, to show us how messed up it could all become.
Improved player feedback.