Comment Re:alogrithms aren't racist (Score 2) 352
I don't believe the algorithm is impugning the humanity of my offspring, I just think it is far-from-perfect.
But is the algorithm even wrong? I think the question to the Google recognizer is "of the images in my collection which ones look most like a seal"? If the collection is mostly all pictures of your kids, it'll show you the pictures of your kids that it thinks have the most in common with what it has as an idea about what seals look like. This isn't to make fun of your kids, of course, it's just its best guess due to the nature of the question that was asked of it. A human could make a similar selection when posed with the same question.
So as to the point of TFA, the searcher asked Google for the pictures that are most likely to be pictures of a gorilla inside his photo collection. If we assume that there were not actual pictures of gorillas in the collection, then the guess might not be a bad one. If you gave any human a set of pictures of a speedboat, a skyscraper, a turtle, a box of cereal, and a woman, and asked the human which one of those things looks most like a gorilla, there's only one truthful answer. It might be an offensive one to some people but that doesn't make the guess mathematically incorrect.
Reading the Twitter stream, Google has decided to censor such results. Their first attempt was to say "if somebody searches for gorilla and it matches a picture with a face in it, don't show that". That failed on two pics where a face wasn't recognized, so they added even more filtering and now they're building an i18n wordlist of "offensive" words to restrict the algorithm's output depending on locale.
Being a for-profit company, one of Google's primary concerns is to not alienate its users, so for them I'm sure it's the right move. But we need to be aware that it is imposing censorship (on itself) and that the output of the algorithm is becoming less useful to some degree to avoid offending some people. It's their trade-off to make, for sure, but for the larger computing community it's a valuable lesson to keep in mind. Such trade-offs need to be made carefully and consciously.