7 comments

[ 3.2 ms ] story [ 27.2 ms ] thread
From the article, "Google’s system may not have 'seen' enough pictures of gorillas to learn the differences – and wouldn’t understand the significance of such a mistake, he said."

It seems to recognize gorillas just fine. I'm guessing it's more like, "may not have 'seen' enough pictures of black or dark-skinned people."

Some of Google's best friends are dark-skinned people!
(comment deleted)
Reminds me about the article explaining how a couch textured with leopard skin got confused as a leopard by an AI algorithm. I assume something similar is happening here.
People textured with gorilla skin getting confused for gorillas? /joke
I remember hearing a story (probably apocryphal) of one of the armed forces trying to use image recognition to identify tanks. So the took a bunch of scenery pictures with and without tanks, then fed those pictures to the computer.

After some processing the computer was identifying pictures of tanks with 100% accuracy. So the group started showing off the software. Some guy proceeded to take a picture outside that had a tank in it, then fed that picture to the software. No tank found. More images given, more problems found.

So the group went back and look at the their tank image stockpile. All the pictures without tanks were taken in the morning, all the pictures with tanks in the afternoon.

When they thought they were training the software in identifying tanks, really it was identifying time of day.