4 ms·
It is a bit more complex. There are few scientific studies on this topic. See this one: http://proceedings.mlr.press/v81/buolamwini18a/buolamwini18a.pdf http://
by synctext 8y ago
It is a bit more complex. There are few scientific studies on this topic. See this one: http://proceedings.mlr.press/v81/buolamwini18a/buolamwini18a.pdf http://proceedings.mlr.press/v81/buolamwini18a/buolamwini18a...
"We find that these datasets are overwhelmingly composed of lighter-skinned subjects (79.6% for IJB-A and 86.2% for Adience) and introduce a new facial analysis dataset which is balanced by gender and skin type. We evaluate 3 commercial gender classification systems using our dataset and show that darker-skinned females are the most misclassified group (with error rates of up to 34.7%). The maximum error rate for lighter-skinned males is 0.8%.
- mc32 8y agoThat’s pretty interesting. Given those stats, why can’t it be selectively deployed on light skinned perps till the system is trained better on other skin color classifications in order to achieve similar accuracies? And in the mean time catch some baddies.
- solotronics 8y agoInteresting, so given your implementation only targeting light skinned "suspects" would using this be racist?
- mc32 8y agoI don’t think so. It would target more skin tones as it became better at identifying suspects accurately. The aim is to reduce crime regardless who commits it. We have a system which can help on a subset but not another. Why should we ignore the potential only because we cannot deploy it against everyone? It’s like saying, well, since we can’t successfully prosecute big bankers, well also ignore smaller fry bankers who are sloppier cuz it’s not fair to them. In the end everyone but the criminals benefit. Eventually one hopes the system is well enough trained so it can be deployed for all skin tones.