Communique for #BlackMastodon and Black folk only:
(White folk can listen too if they want, but this conversation is not for them).
The people telling you to be very afraid of Artificial General Intelligence don't know what they're talking about. Remember, their last big predictions were:
* Monkey jpegs are now money. Buy crypto.
* Elon will be great for Twitter
* Listening to VCs talk on Clubhouse is the next big social network
* Adam Neuman is a genius, and we should give him more money
These people are very very rich, not very very smart. The track record of their judgment speaks for itself.
There are very real, very serious, risks from Machine Learning in general, but they are not the risks that these delusional dudes are talking about.
The real risks are not "coming in the near future." They are here with us today, and they affect systems that impact marginalized communities the most.
Most of the experts on the real risks are from marginalized communities.
All this talk of artificial general intelligence is a head fake to draw attention from the massive and real harms that can be caused by ML systems today.
Today's systems can:
* Issue a warrant for your arrest for a crime that you didn't even do based on facial recognition
* Decide that you are a pre-trial flight risk and deny you pre-trial release
* Give a false diagnosis at a hospital, deciding that you are not worth putting on life support
* Tell a car to run you over as you cross the street
Start with
@timnitGebru @abebab
@alex
@emilymbender
And work out from there to people that they collaborate with and recommend. 👍🏿
Don't take your AI Ethics and societal impact advice from VCs that hang out in Nazi chatrooms. These people aren't that smart, but most importantly, they don't care about you. At all.
A few years ago, I said don't get your financial advice from VCs that hang out in Nazi chatrooms. Some of y'all listened. Others lost their shirts on crypto.
Yes, they have sharp criticism of big tech companies that work in this space. That's a feature not a bug.
If you ever have the immense privilege to work on something that impacts all of us, positively or negatively, expect criticism.
I'm not Miles Dyson. I'm not going to accidentally create T2. That's not a real risk.
But I might accidentally build a system that doesn't work for Black folk. Your city might turn over law enforcement decisions to a system incapable of making unbiased decisions.
@mekkaokereke Law enforcement decisions are ALREADY made by a system incapable of making unbiased decisions, but some of that system is made of humans you can talk to who MIGHT change their minds.
Obscuring biases with bogus AI is just more layers of BS smeared on systems that have already proven extremely hard to hold accountable in any way.
@AutisticMumTo3 @mekkaokereke I found a classic video of it. Here's a video from when facial recog was new. It was clear that the data used to train the software didn't have enough black people. This sort of design goes back to
This goes back to the ~~nature~~ origin of film itself and this shows up again and again and again with every technology. We're literally afterthoughts. Make camera, make it beautiful, then make it work for black people.
@marypcbuk @wolfkin @AutisticMumTo3
(Unapologetic self-promo)
I'm really proud of the work that the Inclusive Imagery group at Google did on this front.
https://m.youtube.com/watch?v=2DXY9cR7vN4
We don't have to accept that photography products just don't work as well for darker skin. That's a choice. And given the skin tones of most of the world's 8 billion people, not a very good choice.
We can make the other choice, to build more inclusive imagery products that work for everyone.
@marypcbuk @wolfkin @AutisticMumTo3 @mekkaokereke I first realized that it's a choice as an anglo getting film developed in south India - there, they default to darker skin (as makes sense), so I looked radioactive in most of the photos. The rolls I developed in the US, my Tamil friends were extra dark and indistinct.
From what I hear, modern digital photography has made great strides on working for everyone.
@marypcbuk @[email protected] @[email protected] (Unapologetic self-promo) I'm really proud of the work that the Inclusive Imagery group at Google did on this front. https://m.youtube.com/watch?v=2DXY9cR7vN4 We don't have to accept that photography products just don't work as well for darker skin. That's a choice. And given the skin tones of most of the world's 8 billion people, not a very good choice. We can make the other choice, to build more inclusive imagery products that work for everyone.
@amaditalks @mekkaokereke
https://www.wired.com/story/algorithms-supposed-fix-bail-system-they-havent/
These algorithmic approaches were supposed to remove prejudice from the system, but instead have ended up encoding it into the system in a layer of opaque abstraction.
More generally, replacing human staff, who can understand mistakes and special situations and work around them, by an "AI" system that works fine in 99% of the cases but cannot do any of that when faced with the 1%.
(And if each of us has a 1% chance of getting screwed on any interaction with such systems, in a couple of years everyone will have got screwed...)
@mekkaokereke Hard disagree! Today's systems are insanely dangerous and should be regulated and called out, no doubt about that.
Being concerned about AGI *as well* is not a head-fake. We can be concerned about more than one thing.
Just like being concerned about the issues with current-day ML is not a head-fake to draw attention from climate change: both issues are real and deserve attention.
@moshez @mekkaokereke Rob Miles has some really interesting content wrt. why it's important to take AGI alignment issues seriously.
That being said, the techbro hivemind has absolutely tried to shift the discussion over to _only_ discussing the ramifications of technology that doesn't exist yet. The misuse of existing AI tech is already happening, at scale, right now, and it will only get worse as we climb the exponential curve.
Meanwhile, Microsoft just laid off an AI ethics team.
Go figure.
@duk @mekkaokereke If it's any consolation, MS also does pretty badly on long-term AI Safety, so if nothing else, they're consistent in putting profits over people.
Which is kind of my point. This isn't an either-or situation, just like "stop structural racism" and "stop structural sexism" aren't either-or, and anyone who tells you differently is probably a fascist.
We can care about both and believe both are problems we absolutely should solve.
@mekkaokereke >>There are very real, very serious, risks from Machine Learning in general, but they are not the risks that these delusional dudes are talking about.<<
This was going to be my only comment when i saw that first post. That there are things to fear from AI but just not things like a cyberwar in 2029 with autonomous robotic soldiers ala Terminator 2. More like AI stealing all your art styles w/no attribution or compensation for one.
mekka okereke 
dizzyd
Akshay | అక్షయ్
Robert J. Berger
Pär Björklund
Kofi Loves Efia 
D. G. Fitch
@PublicHealthMediaNetwork
wolfkin
Ricardo Harvin
Analog AI
AutisticMumTo3
Mary Branscombe
Dangandblast
Amadi Lovelace
The Febrile Muse
Tim Kellogg
T.C. Harris
drawks
Wayne Werner
klausfiend has moved
Jorge Stolfi
Jake 
Moshe Zadka
duk