The Biases of AI

[Gord_in_Toronto]

Here's an interesting project.

ImageNet Roulette, a digital art project and viral selfie app, exposes how biases have crept into the artificial-intelligence technologies changing our lives.

https://www.nytimes.com/2019/09/20/arts/design/imagenet-trevor-paglen-ai-facial-recognition.html

The page is here: https://imagenet-roulette.paglen.com/

I tried a few photos I had of myself and it appears the White Male Privilege certainly works.

My results were:

• Counsel to the Crown: a barrister selected to serve as counsel to the British ruler
• biographer: someone who writes an account of a person's life, and
• newsreader, news reader: someone who reads out broadcast news bulletin

The last was actually a picture of me and Randi(qv) and identifies him as:

• queen mother: a queen dowager who is mother of the reigning sovereign

Houston we have a problem.

Amusingly enough (and we all know how easily I am amused), trying a number of pictures of family members turned up some quite accurate results such as granddaughter and scientist (the Barnum effect at work?). Though a picture of my middle-aged son and my wife identified them both as nuns.

Note that: ImageNet Roulette has made its point - it has inspired a long-overdue public conversation about the politics of training data, and we hope it acts as a call to action for the AI community to contend with the potential harms of classifying people.

And so as of Friday, September 27th, 2019 we’re taking it off the internet.

 

[Belz...]

In the end AI is made by humans and often it uses human data to reach conclusions. If the data is corrupted or biased, then so's the machine's conclusions. It has little ability to spot problems with the data, unlike humans, who already struggle with their own bias.

 

[dann]

The last was actually a picture of me and Randi(qv) and identifies him as:

• queen mother: a queen dowager who is mother of the reigning sovereign

Come on, he is bald and has a long beard! Anybody could make that mistake!