Metrics are just a proxy for what you really care about, and unthinkingly optimizing a metric can lead to unexpected negative results, such as YouTube heavily promoting conspiracy theories. https://t.co/9YErquyPk8 pic.twitter.com/ttK7YDbJq9
— Rachel Thomas (@math_rachel) February 9, 2019
“Let’s start this meeting by discussing the alarming fact that our algorithms turned out to be extremely biased and opaque and no one knows why.”
— Vicki Boykis (@vboykis) January 30, 2019
“Hold on — are you saying we were able to implement them and that they’re actually running in production?”
"Isn’t algorithmic bias just a reflection of how the world is?" No, our algorithms and products impact the world and are part of feedback loops. Our work is not just an effect, but also a cause.https://t.co/9YErquQqII pic.twitter.com/7IZi4Wphph
— Rachel Thomas (@math_rachel) January 30, 2019
No matter how good your model is, if you only optimize for watch time, you naturally push your users towards consuming content that draws them viscerally, which often tends to be more radical. This is a case where finding additional metrics, and moderation rules is crucial.
— Emmanuel Ameisen (@EmmanuelAmeisen) January 29, 2019
Really interesting article on #AlgorithmicBias and fairness in #MachineLearning
— francesc (@francesc) January 26, 2019
A great read from @quaesita in support of @AOC's comments earlier this week!
“Pay attention to that man behind the curtain” by Cassie Kozyrkov https://t.co/bxZfjlRXQp
New @MIT study shows gender and racial bias in @amazon Rekognition AI product -100% accuracy on pale males vs 69% accuracy on women of colorhttps://t.co/lkWce4hR5a - Study link https://t.co/3aU83ETcRG @ACLU @Data4BlackLives @black_in_ai @medialab @AIESConf @AINowInstitute @AOC pic.twitter.com/YsloeUE2yz
— Joy Buolamwini (@jovialjoy) January 25, 2019
Yikes.
— B Nalini (@purpultat) January 23, 2019
Is there a paper I can read to understand what this "AI-based tech" really does and how it does it? What data did they use to train these algorithms? What is the accuracy of the prediction on which the army might make life/death decisions?
Is the army asking these Qs? https://t.co/gQ2S8jQzxg
Let me give you an example of why high accuracy in a dataset doesn't translate into a valid or even useful modelhttps://t.co/JJfA8iixep
— francesc (@francesc) January 23, 2019
Thanks to everyone sharing great resources on the topic!
— Data Science Renee (@BecomingDataSci) January 23, 2019
Here's a Twitter list of people who discuss topics like ethics in machine learning regularly: https://t.co/2N7ZXQHCLy
I also have this Flipboard where I collect articles about it:https://t.co/4t9Sd13Iw1
"Bias laundering" happens when we choose to ignore the biases of automated decision systems because of the illusion that all algorithms must be objective since they're "driven by math" or "run by a computer". https://t.co/AMuUaOx4Mv
— François Chollet (@fchollet) January 23, 2019
There's no possible way for humans to monitor everything on YouTube, and its algorithms boost videos based on virality, not credibility. So we're trapped in this shadow realm where the world's biggest video site keeps spotlighting conspiracy-theory crap https://t.co/Q4KiCUg4ck pic.twitter.com/L0AwZkfBGa
— Drew Harwell (@drewharwell) January 11, 2019
Fake news has been shared on Facebook primarily by older, very conservative people in the US.
— Yann LeCun (@ylecun) January 11, 2019
There is a lot less fake news on nowadays because Facebook limits its dissemination.
This gives us hope that the young... https://t.co/lswSsHPjfp
