Thread by @ginasue

We called BS on AI.
Here's a thread of what we learned.
(1/15)

In 30 minutes we came up with ~28 cases of AI not working. There are lots more to map out.
(2/15)

Some cases are clearly where AI is BS while other cases reflect "growing pains" of a developing technology -- this insight from @dashungwang.
(3/15)

Lots of these cases are well known. Amazon Rekognition misgendering dark skinned women: wired.com/story/amazon-f… (Thanks to @jovialjoy's and @timnitGebru's amazing work)(4/15)

Amazon's AI hiring tool scrapped for learning to be sexist. reuters.com/article/us-ama…
(5/15)

COMPASS recidivism software is biased against blacks.
propublica.org/article/machin…
(6/15)

Others of these cases were less well known: Bless @msalganik's heart he called BS on himself.
(Read his great book Bit by Bit)
press.princeton.edu/titles/11057.h…
(7/15)

If more people working in maching learning could call BS on themselves occasionally, the world would be much better off.
(8/15)

We certaintly would be more circumspect in the face of calls for 'The Singularity': Kim Stanley Robinson called BS.
(9/15)

And calling BS on AI is not new - @jseelybrown reminds us.
(10/15)

Brilliant @merbroussard calls BS by calling it 'artificial UNintelligence'. mitpress.mit.edu/books/artifici…
(11/15)

What to do? Keep mapping these cases of AI BS. And...
(12/15)

UW's @billghowe from @uwescience and @rogerm suggested synthetic data can help when today's data make biased inputs (GIGO).
(13/15)

Thanks to all who helped us call BS on AI, including my amazing co-host @jevinwest (Buy his forthcoming book Calling Bullsh*t)
(14/15)

In the hopeful words of @movie_pundit, why are we using data from our world today to build models for the world we want tomorrow?
15/15. <fin>