I have a strong dystopian imagination but had not anticipated algorithmic horrors like what Israel has unleashed in Gaza with its target generation systems
Read
@marwasf
& me on how systems that should be banned in peacetime are repurposed in war for murderous purposes
Recent investigations have revealed how Israel’s AI systems “Lavender” + “The Gospel” are causing mass destruction in Gaza, raising ethical concerns about AI in warfare.
@marwasf
and
@djleufer
explain how these systems are unleashing tech-driven horrors:
Very cool, creepy project that uses AI to scan CCTV footage to match with Instagram influencers' pics & track them
Easy availability of AI tools like this means we need to reevaluate the risks posed by CCTV, as the scope for stalking & tracking has exploded
🔥NEW PROJECT!🔥
'The Follower' is software searching how an Instagram photo was taken with the help of AI and open cameras.
Project page:
YouTube video:
Support my work:
🧵👇
Today I launch my
@mozilla
fellowship project on
#aimyths
!
I’ve put together the resources on this website to tackle 8 of the most insidious myths, misconceptions and inaccuracies about AI
The aim is to disect myths and guide folks to great existing work
These updates to give users more control over their data are 100% a result of pressure on GDPR compliance from
@GPDP_IT
But hacks will say "oh Italy banned ChatGPT" or "GDPR is out of touch"
No, GDPR just made OpenAI improve ChatGPT for all users
My (perhaps) controversial AI opinion is that I am utterly, profoundly uninterested in whether it is, or could be sentient
In 8 yrs of academic philosophy I never had a spark of interest in a text about defining “what is consciousness” etc
100% don’t care
🚨🚨Big (bad) news on the EU AI Act🚨🚨
@POLITICOEurope
leaked the latest compromise proposal for a loophole in Article 6
This would allow developers to unilaterally exempt themselves from all obligations based on 3 very wide criteria
A long thread on why this is really bad
🚨🚨BREAKING: AI Act Plenary Vote🚨🚨
The European Parliament voted to affirm the full ban on real-time remote biometric identification (RBI) & almost full ban on post RBI 🎉
This is incredible, but they failed to add bans on dangerous AI systems in migration & border contexts
🚨 AI Act 🚨
Seeing some very worrying amendments being proposed to Article 6 of the AI Act that would effectively make it a set of voluntary guidelines
In the original draft, your system is high risk if it's used for one of the purposes listed in Annex III
This new proposal...
🚨EU AI Act🚨
As others have said, the Parliament is under extreme pressure to accept a package deal on the prohibitions in Article 5
I have the document & it is an absolute disgrace
Parliament cannot accept this, no deal can be made based on this, here's why...
Great news that
@Microsoft
is dropping its dodgy face analysis tools, including emotion, gender and age recognition
@accessnow
& others have been pushing industry to do for years
But we can't rely on the goodwill of companies alone, these things need to be legally prohibited
Every 6 months there's a piece about a mask, makeup or tshirt that makes you 'invisible' to facial recognition
It stops systems for about 24hrs, the systems adapt, they work again etc etc
This is not a solution, the solution is to ban facial recognition
#banBS
#reclaimyourface
Researchers have designed a shirt that renders the wearer undetectable to AI. It's part of a growing number of "adversarial examples"—physical objects designed to counteract the creep of digital surveillance: Via WIRED UK (From 2020)
📸: Benedict Evans
Worried about
#GenerativeAI
, but not quite sure what to be worried about?
We've got you covered in our new FAQ on generative AI &
#humanrights
where we cut through the robocrap about what it is, what it isn't, and who you should be listening to
Check out my new blog post on
@mozilla
on hype, myths and inaccuracies about AI.
For the next 9 months, I'll be working with experts to develop resources to challenge these narratives and myths.
Let me know what myths need busting, and who to work with!
For all you AI policy nerds, I can report that the official number of amendments in the IMCO-LIBE committees on the AI Act is 3312 😭
Wish us all luck in reading through the good, the bad, and the nonsensical
(p.s. there's a full ban on emotion recognition in there)
Hot take on the letter for a moratorium on training systems more powerful than GPT-4
Anything worthwhile in it was already said in
@timnitGebru
@emilymbender
@mmitchell_ai
et al.'s paper, Stochastic Parrots, over two years ago, but without all the hype
For AI policy aficionados looking to keep working during the summer holidays...don't!
But if you do, check out our new EU AI Act timeline page that keeps track of
@accessnow
's work on the AI Act
You'll find all our blogs & civil society coalition work
😃😭😡AI Act Prohibitions Pt.3😡😭😃
Emotion recognition & AI polygraphs (lie detectors)
Today MEPs will discuss an ambitious range of topics, including Article 5's list of bans
I want to highlight here why that list must include emotion recognition & AI polygraphs
Long 🧵
EU AI Act update
December 6 is supposed to be the 'final' trilogue negotiation for the AI Act
But huge issues remain unresolved, from police use of AI to foundation models
There is a big risk that negotiators will announce that a 'deal has been reached' when it has in fact not
🚨AI Act Deal🚨
There's been a deal, at least on most of the text
The issue is that we don't know what was sacrificed
The PR is a big danger
They will say "we banned the most dangerous uses & added robust safeguards for other uses"
But details matter
At the end AI act trilogue, press conference with Spanish Secretary of State Carme Artigas Brugal at arount 23.50 (provisional timing).
Follow the livestream here🎥👇
Subprime automation, a 🧵
This startup erases the accents of call centre workers so they sound white & American
Its 'solution' ignores the real issues of worker surveillance & abuse
But the problem is deeper & part of a pattern in AI 1/6
There's often a false dichotomy made between promoting innovation & protecting people's rights
Today, a group of 149 investors representing over US$1.6T in assets issued this letter on the AI Act asking for greatly increased protections for human rights
🚨 AI Act loophole update🚨
Well, well, well...
Unequivocal legal opinion from the Parliament's legal service that the Article 6 'filter' aka loophole introduced by EP and Council is not legally possible & undermines the core of the AI Act
We've said this since June 2022
#AI
Act: The EU Parliament's legal office issued a damning opinion of the proposed exemption conditions that would allow AI developers to avoid falling into the high-risk category. This approach is seen as in contrast with the AI law's regulatory aim.
For all you AI Act enthusiasts, I can confirm that at least one amendment will be tabled that uses the word 'metaverse'
I had proposed this as a joke, and now it's a reality
If you were wondering why so-called 'general-purpose AI systems' pose a risk to human rights in the very way they are designed and trained, here's a good example
Stable diffusion, the image generation model that is wowing the world, was trained on a database full of porn
This site let's you search the giant database behind image-making AI systems like Stable Diffusion. It's supposed to be for artists to see if their art is in the data, but it also shows the sheer volume of NSFW/toxic stuff that's behind these AI tools.
According to Euractiv, MEPs agree to push for a FULL BAN on remote biometric identification in the AI Act
Unlike the original draft, this would cover real-time and post (i.e. Clearview AI), public and private actors, and online and offline spaces
#banBS
#reclaimyourface
#AI
Act: For the first time, MEPs discussed politically sensitive topics like biometric recognition and the regulation's scope in a shadow meeting yesterday. Compromise AMs were also put forth on the high-risk database.
This is absolutely huge!
"EDPB and the EDPS call for a general ban on any use of AI for automated recognition of human features in publicly accessible spaces"
Not just live facial recognition, but ALL automated recognition of human features
#banBS
#reclaimyourface
This has always been an extremely important distinction in surveillance
You'll often hear facial recognition advocates say stuff like "oh but your data is deleted if there's no match, so your privacy is not at risk" or "you weren't identified if there's no match"
Whereas...
People—including me, sometimes—often say "privacy" when they mean "data protection".
With AI, this distinction really matters.
ChatGPT presents some privacy risks.
But there are bigger issues around accuracy, transparency, fairness—data protection concepts. Not really privacy.
Just had the realisation that it's a matter of months before I bump into some unbearable founder of a robot rights NGO that's doing some kind of
#freeSydney
type campaign against content moderation on LLM chatbots
Already deeply annoyed about a thing that doesn't exist yet
"less existential" = less ridiculous, sci-fi
Also, as someone with a PhD in existentialist philosophy, I can say that
@timnitGebru
's work is more existentially serious than any of this AI bro faff about superintelligence
I asked
@jaketapper
to ask Hinton about his lack of support for
@timnitGebru
,
@mmitchell_ai
@mer__edith
and other whistleblowers.
His response is that their concerns were less existential than his and that it’s easier to voice concerns if you leave first
According to
@POLITICOEurope
Portugal is releasing a legal chatbot running on
@OpenAI
models to “answer citizens’ questions about the judicial system”
Many things to say about this but let me zoom in on this (hopefully incorrect) comment that it won’t be high-risk under AI Act:
Attention AI Act aficionados!
From this summary of Brazil’s draft AI Law, it seems the Brussels effect is already in full swing, i.e many similarities to the AI Act, including the same 4 prohibitions
But lots of interesting differences too, looking forward to more info
The Brazilian Senate has released a 900+ page report & draft AI Law, drawing inspiration from earlier proposals & research into how OECD countries regulate AI. Here are some aspects I read in the draft AI law...
A UK police force is looking to trial a facial recognition system to detect emotions
Problems include that emotion detection is scientific garbage & also what possible use is a "system to search [CCTV footage] for certain moods and facial expressions"
No, just have fewer crappy meetings
Such classic AI solutionism:
"There's a structural problem with X, how about we fix it by not tackling the problem, but by adding AI"
Research has shown that meetings correlate with a decline in workplace happiness, productivity, and even company market share. Can tech, like automated scheduling tools & facial recognition that measures who’s paying attention, make them better (From 2020)
I've always been pretty confident that I can't be replaced by AI, because my job is mostly analysing how AI impacts people's rights...
...but ChatGPT got interviewed for this piece & I didn't
😂
The tide is turning on the rollout of facial recognition & biometric surveillance
Today, over 170 civil society organisations and activists from 55 countries around the world release this statement to call for an outright ban on biometric surveillance in public spaces!
The world needs to end biometric surveillance💥
We’re joining a global call w/
@AmnestyTech
@internetfreedom
@idec
@edri
@hrw
& 170+ human rights advocates to ban this harmful tech that fuels discrimination & violates civil liberties.
#BanBS
You cannot let developers be the ones to determine whether they should adhere to legal obligations
If your system is designed for one of the high-risk purposes in Annex III, it's high risk
Let's not turn the AIA into a worthless set of voluntary guidelines
#AIAct
updates
@POLITICOEurope
just leaked what appears to be the most recent "4 column document" on the AI Act
This is the almost unreadable 1000 page behemoth with all the alternative versions of articles & what should be the final version adopted it in the deal
After going through the full 80+ pages of this leaked draft of the
@EU_Commission
's regulatory proposal on AI, here are some initial comments:
- First, this leak is of a draft from January, so it's likely, and hopeful, that the draft has significantly progressed since then
SCOOP! I got a leak of the European Commission's upcoming AI laws. The Commission wants to ban certain uses of high-risk artificial intelligence systems and only allow risky AI tech into the EU’s internal market if it has been vetted
⚡️AI Act Article 5⚡️
As discussions in Parliament progress, we're getting close to the debate around prohibitions, so I'll do a few threads on what needs to be banned in the AI Act
First up: real-time AND post remote biometric identification (RBI) in publicly accessible spaces
👩🏾⚧️🏳️🌈 AI Act bans Pt.2 🏳️🌈⚧️🧔🏼♂️
Biometric Categorisation
As Parliament starts to discuss bans, here’s a thread on why inherently discriminatory forms of biometric categorisation must be banned
AI should be not be used to infer race, gender, sexuality & other sensitive data
Invariably some people thinking I am saying it's cool you can do this 🤡
The project is a cool/good/great/innovative (whatever word you want) way to alert people to the surveillance nightmare we are living in, but it is definitely not cool that you can do this.
#AIAct
The AI Act is not a done deal
The text isn't finalised & key provisions on loopholes & exceptions must still be agreed
Governments, industry & police will all be lobbying these remaining 'technical' meetings, so scrutiny must be maintained
But loopholes will remain...
The EU is celebrating reaching a deal on the AI Act –– but its human rights protections have major exceptions.
Here's
@djleufer
,
@infofannny
, +
@CaterinaRodelli
on the deal and why the AI Act has always been a concession to industry + law enforcement.
Guarantee that if you’re asked this question on an interview the response video will be analysed by some totally pseudoscientific AI system that does something stupid like infer your Big 5 type from how you move your jaw
#AIAct
Based on what we're hearing from the room, the deal that is being offered to Parliament on prohibitions is an absolute disgrace
It can't be accepted. Council, pressured by police, is pushing to be allowed to use AI TO RACIALLY PROFILE PEOPLE
This is AI physiognomy
#AI
Act: the Council is pushing for allowing law enforcement to conduct biometric categorisation based on race. Pressure is on Parliament to accept the package deal on prohibitions.
Having a a Fundamental Rights Impact Assessment
#FRIA
in the final deal of the
#AIAct
is a huge milestone, reached thanks to the Co-Rapporteur
@brandobenifei
and thanks to the many senior colleagues who two months ago have accepted to sign the
@BPrivacyHub
’s appeal launched by
Today we launch our new report on human rights impact assessments for AI systems, by the fantastic
@PhilipDawson
&
@BNonnecke
As the use of AI systems impacts more aspects of our lives, the need to audit them & measure their impact becomes more pressing
There are an incredible amount of pressing issues at the intersection of AI & philosophy, but for me they are about its politics, about infrastructure, how ambiguous aspects of our existence are encoded & inferred, how people are oppressed by AI systems & resist that oppression
At the AI Act press conference just now,
@ThierryBreton
says "on remote biometric identification, there is a full ban...with just 3 exceptions"
That's not a full ban. It's a guidebook on how to use a technology that has no place in a democratic, rights-based society
After being heavily quoted in multiple languages calling the Guardian GPT3 op-ed "an absolute joke", I can now add to my list of published GPT3 put-downs calling it "a very fancy autocomplete"
@OpenAI
must love me
The world's first play written and performed with an AI debuts at
@youngvictheatre
tonight. (It may contain racism.)
I chatted to the team behind the play, with extra insights from
@Abebab
and
@djleufer
Last week
@accessnow
submitted our response to the
@EU_Commission
's consultation on the draft AI Act.
A thread 🧵 focusing on our recommendation on prohibited AI practices...
In an enormous win for human rights, the
@EP_SingleMarket
&
@EP_Justice
committees have voted in favour of key amendments to improve the AI Act
Key gaps remain, but this is big progress
Here's our press release:
Bit of a hot take but it's interesting that educational institutions have been investing massively in AI to SURVEIL students for years (AI proctoring etc) but the widespread moral panic only happens with
#ChatGPT
because it's an application of AI that helps students
In our new blog we look at
@Facebook
&
@ray_ban
's new 'smart' (= surveillance) glasses
These glasses are a first step towards full blown
#AugmentedReality
glasses & they show that Facebook does not take people's rights as seriously as its own profits
AI policy nerds!
The full list of 3312 amendments from the
@Europarl_EN
lead committees are available online here:
And from looking at the prohibitions, there are 251 amendments alone on Article 5
The EP is really on the banwagon
This
@guardian
#GPT3
article is an absolute joke. It would have been actually interesting to see the 8 essays the system actually produced, but editing and splicing them like this does nothing but contribute to hype and misinform people who aren't going to read the fine print
this is nonsense,
@guardian
without a) a register of edits and/or b) publishing the original machine-composed essays, the journalistic value of this is 0
Me &
@sarahchander
spoke to Billy about this very interesting document from OpenAI
In autumn 2022 a disturbing loophole appeared in the Council text of the AI Act, exempting "purely accessory" AI systems from all requirements
It's very in line with what OpenAI demanded
🚨 SCOOP: OpenAI lobbied the E.U. to weaken forthcoming AI regulation, even as in public it calls for stronger AI guardrails, documents obtained by TIME show.
We're publishing the key document in full alongside my story 👇
And finally the ban on real-time remote biometric identification is not a ban at all, it's a guidebook on how to legitimise a dystopian surveillance technology that should be fully banned with no exceptions
This package is a disgrace,
@brandobenifei
&
@IoanDragosT
must reject it
The clock is ticking on Clearview AI
A growing number of regulators have confirmed what activists have said for years: it's database was compiled illegally, and needs to be fully deleted
More than that, the entire suite of products built off that data needs to be deleted
“If your training data is illegitimately collected, not only should you have to delete it, you should delete models that were built on it,” Access Now's
@djleufer
says of Clearview AI.
Read the story by
@billyperrigo
via
@TIME
:
As Brussels AI policy nerds search high & low for copy of the IMCO-LIBE draft report,
@BertuzLuca
does us all a service with a sneak peek at what the report contains.
Here's a thread on some of the major things to watch out for as we wait to get our hands on it 🧵
@brandobenifei
&
@IoanDragosT
agreed on the draft report for the
#AI
Act. Highlights:
▶️Broad definition: no carve-outs
▶️Governance: Commission in charge of cross-border cases
▶️High-risk: extension to education, health, insurance, democracy applications
#AIAct
trilogue!!!
One ridiculous argument that keeps coming back from law enforcement, member states & Council is that "criminals have AI, so we need it too, you can't tie our hands"
This simply doesn't make sense
If it's about cars, ok. They have fast cars, you need them too
"While AI can bring many benefits to society, it also poses risks..."
Something very particular happens to my brain when I hear or see sentences like this
Fact: any report on AI will be 30% generic blurb on the 'potential' of AI, 30% about the need for 'trustworthy' systems, 30% stock photos of blue robotic hands, and 10% actual novel content.
#AIAct
trilogue
In this debate about prohibitions, civil society is not ‘extreme’ & the Parliament position is not something ‘extreme’ that needs to be toned down
Our position on bans is backed up by the
@EU_EDPS
@EU_EDPB
@volker_turk
UN OHCHR & numerous experts
No deal, thankfully. Parliament was being bullied into accepting a disgraceful package on the prohibitions.
Now time to regroup and for Parliament to go back in tomorrow and stand their ground for fundamental rights.
#NoBansNoDeal
🚨🚨AI Act Vote🚨🚨
A huge amount is at stake in tomorrow's vote in
@EP_SingleMarket
&
@EP_Justice
committees
At 9am MEPs will vote on 12 batches of amendments, incl. important provisions to increase accountability & transparency, but all eyes are on bans
Sounds great at the level of PR vagueness, but until we see how these are defined, what the exceptions are etc, we cannot celebrate
As an example, the most recent version of the predictive policing ban leaked was so vague it wouldn’t ban any existing system
Worth staying up for the applause and whooping when the inimitable
@BertuzLuca
asks his question at the press conference
Also, fantastically specific questions
Happy to announce that with
@ruchowdh
@sarahbmyers
& others I'll represent
@accessnow
on the
@OECDinnovation
AI Futures Working Group
There should be more civil society & global majority folks in it, but we'll do our best to push for a human rights focus
Whatever you may have thought about the academic discussion around robot rights, I think that any sane person must agree that actually campaigning on it, actively working on advocacy for "rights for AIs" is an outrageously stupid waste of time
Yet it's happening
🚨AI Act 'final' negotiation🚨
For those of you tuning in this morning, this is what's still on the agenda
This meeting has been going since 3pm YESTERDAY
All negotiators are exhausted & under pressure to seal a deal
But a deal must not happen at the expense of our rights
"We reject the Western-normative language of “ethical” AI and suggestions of “inclusivity” that do not destabilize current patterns of domination and address power asymmetries."
💭a Manyfesto of decolonial AI📝
"This manyfesto is a provocation, a question, an opening, a dance about a future of AI technologies that is decolonial."
#manyfesto
#AI
Delighted to share our latest work, 'Black-Boxed Politics:
Opacity is a Choice in AI Systems' on debunking the black box narrative around AI systems.
It was an absolute pleasure to write this with the incredible
@szymielewicz
and
@agatafshsh
!
Longer analysis coming when
@EU_Commission
publishes the final version of the AI proposal later today, but already want to flag issues with the treatment of 'biometric categorisation systems'
Current definition is: "an AI system for the purpose of assigning natural persons...
For anyone confused about the ridiculous EU AI Act trilogue negotiation process happening now, this explainer by Joe McNamee is fantastic, if depressing
This thing is a nightmare.
There is NO WAY they could have sufficiently complex, accurate & meaningful data to train an ML system to perform something as complex as this. It is pseudoscientifc crap.
Also props to
@Gizmodo
for that headline
The 'ban' on biometric categorisation is a joke
It would 'ban' sth that is already illegal & give law enforcement a green light to profile people based on race, sexual & political preference & union membership if "those characteristics have a direct link with a specific crime...
@schock
The choice of terminology can have a huge influence on how credible or ridiculous statements sound.
To play around with that
@alexabruck
& I made a 'terminology swap' tool so you can change the word AI into something else:
- and so ultimately there's no improvement, we just automate & thereby reify the subprime situation & its conditions
- everything remains crappy, but with a veneer of 'fancy AI' to make it look innovative or disruptive, when it's actually reactionary, conservative & crap
5/6
I have to say, this letter is really nonsensical
"trying to enshrine the regulation of generative AI in a law and proceeding according to a rigid compliance logic is an approach that is as bureaucratic as it is inefficient"
Their alternative: let industry self regulate
Executives from over 150 European companies sign an open letter saying the draft AI Act will jeopardize the EU's competitiveness without addressing challenges (
@javierespft
/ Financial Times)
"...or threat for which the processing of data on those characteristics are of direct relevance and necessary to establish that link"
Aside from the fact that AI can't do anything like this reliably, this is pure dystopian surveillance, automating discrimination
Today,
@accessnow
& 10 other NGOs sent a letter to Czech Deputy Prime Minister for Digitisation
@PiratIvanBartos
asking for increased fundamental rights protections in the Council's AI Act position
Let's make sure the AI Act really protects people
What medical or safety reasons are there for the use of AI-based emotion recognition pseudoscience in schools & workplaces?
Will this safety exemption allow "aggression" detection, which has been shown to be biased against black people? Imagine the impact on students & employees
I called it! The Italian decision is just the beginning of legal woes for OpenAI and other developers of generative AI systems
Years of shoddy data protection practices are going to be paid for now, as there's no easy way for these systems to be brought into line
The EDPB confirmed that it has resolved the dispute on Meta's data transfer case related to the fine and remedies (details not yet known).
The Italian authority's decision against ChatGPT was also discussed, & a dedicated task force will be launched.
Thinking about this more, the crappy choices of tech bros not only monopolise our research agendas (and for civil society, our advocacy agendas), but also our imaginations
I find that my brain automatically thinks of the worst thing sby will do with new technology, rather than..
It really irritates me that my research agenda has to be based on what tech bros decide to chase. They get $$$ chase X, X gets proliferated, then we spend all our time analyzing and mitigating the harms of X. We never wanted to build X, that was never in our research agenda,
For all my large language model/generative AI people, we all know that 'hallucinate' is not a good term to describe when LLMs produce false information but what's a better term?
I haven't seen a good alternative, but I don't like the term
cc
@Abebab
@KrisShrishak
Here is our initial statement on the
@EU_Commission
's proposal for a regulation on AI
This proposal takes minimal steps to protect fundamental rights by introducing some prohibitions, but needs to go much further.
For me, few of the most philosophically important papers on AI were written by ‘academic philosophers’
They’re written by people working as activists, interdisciplinarily & who are existentially & politically driven to understand & deconstruct AI as a socio-political phenomenon
I’ve been involved in closed door conversations where the incredible, groundbreaking work of racialised & marginalised scholars on AI was dismissed as “unphilosophical” by mediocre white guys whose work is extremely “academically digestible” i.e. who write dull papers…
Every day I have a nice little agenda of things to do and every day there's NEW COMPROMISE AMENDMENTS ON THE AI ACT
I won't know what to do with myself when this thing is finally adopted
I couldn't think of a less
#trustworthy
application of AI
This is why last week
@accessnow
,
@edri
& others called for a ban on "the use of systems which make inferences and predictions about our most sensitive characteristics, behaviours and thoughts"
Bad week for Clearview AI!
Another 5.2 million euro fine from the French data protection authority, and...
- fingers crossed -
tomorrow their product, as well as their method of building their database, will be banned in the European Parliament position on the AI Act!!!
#FacialRecognition
The CNIL decided to impose an overdue penalty payment on CLEARVIEW AI. The company must pay 5,200,000 euros for not having complied with the order issued as part of the sanction decision of October 2022 👉