Tech companies have the skills and technology to detect abuse, and they should be responsible for reporting it.
Our new proposal sets obligations for companies to detect and report the abuse of children, with strong safeguards guaranteeing privacy of all. #EUvsChildSexualAbuse
I’m not sure that’s strictly true. Detecting abuse is INCREDIBLY difficult to do reliably. If you depend on automated “AI” systems perpetrators will learn to get around them. If you use human moderators there is too much content to manually review and the mental health tole on moderators is devastating.
More needs to be done, but we can’t flippantly make directives & then make it tech companies problem to implement. You must with *with* them.
The Dutch tax authority ruined thousands of lives after using an algorithm to spot suspected benefits fraud — and critics say there is little stopping it from happening again.
I wonder if the @EU_Commission knows that #Tech companies can plant undetectable #backdoors in these #AI that let them control any output without any observer being able to detect the manipulation.
Given the computational cost and technical expertise required to train
machine learning models, users may delegate the task of learning to a service
provider. We show how a malicious learner can plant an undetectable backdoor
into a classifier.
A new EU Centre to #StopChildAbuse online will help EU countries, companies and local authorities to: 🔹 Establish robust prevention measures 🔹 Ensure that offenders are brought to justice 🔹 Support victims
Crimes don't occur "online", crimes are perpetrated by real offenders.
You can increase staffing and budget of specific police units and general prevention if you really think there's a problem, but policing by "AI" is a sure way to automatic false accusations, with fatal consequences you will be responsible for.
But I assume you've been told that a thousand times already, and have developed a thick skin instead of a conscience. Clowns
Given the computational cost and technical expertise required to train
machine learning models, users may delegate the task of learning to a service
provider. We show how a malicious learner can plant an undetectable backdoor
into a classifier.
It has even more far-reaching consequences than the inevitable accidental false positives that will *already* destroy the "AI detection" dreams 😃 It basically destroys the whole ML models as a service industry.
Or it would, if marketing didn't trump logic every single time ...
Except they don't because their algorithms rely on known content that is spread by abusers, but it won't detect any new content that might even be being shared through other channels to begin with.
Please cease this.
The only thing those algorithms are good for, is to open backdoors for third parties to spy on our chat logs. Pretty much the European version of pegasus, but legal and implemented as a feature on all IM software.
The best way to prevent child abuse is good parenting, instead of parents giving phones with internet access to 5yo children who might become prey to predators on the internet.
Given the computational cost and technical expertise required to train
machine learning models, users may delegate the task of learning to a service
provider. We show how a malicious learner can plant an undetectable backdoor
into a classifier.
it's not tech companies that should make the police, and EU should better invest in it's democratic police system rather than privatizing it like this.
surveillance state is not the solution, and you can't protect privacy while in the same time watching everybody. #chatcontrol is wrong !
Gilbert Busana
in reply to Gilbert Busana • • •Our new proposal sets obligations for companies to detect and report the abuse of children, with strong safeguards guaranteeing privacy of all.
#EUvsChildSexualAbuse
Gilbert Busana
in reply to Gilbert Busana • • •More needs to be done, but we can’t flippantly make directives & then make it tech companies problem to implement. You must with *with* them.
Gilbert Busana
in reply to Gilbert Busana • • •Dutch scandal serves as a warning for Europe over risks of using algorithms
Melissa Heikkilä (POLITICO)Shamar
in reply to Gilbert Busana • • •Paper here: https://arxiv.org/abs/2204.06974
@wiredfire
Planting Undetectable Backdoors in Machine Learning Models
arXiv.orgGilbert Busana
in reply to Gilbert Busana • • •🔹 Establish robust prevention measures
🔹 Ensure that offenders are brought to justice
🔹 Support victims
Learn more⮕ http://europa.eu/!4CDyNK
#EUvsChildSexualAbuse
Press corner
European Commission - European CommissionGilbert Busana
in reply to Gilbert Busana • • •You can increase staffing and budget of specific police units and general prevention if you really think there's a problem, but policing by "AI" is a sure way to automatic false accusations, with fatal consequences you will be responsible for.
But I assume you've been told that a thousand times already, and have developed a thick skin instead of a conscience. Clowns
Shamar
in reply to Gilbert Busana • • •https://arxiv.org/abs/2204.06974
@EU_Commission
Planting Undetectable Backdoors in Machine Learning Models
arXiv.orgShamar reshared this.
Gilbert Busana
in reply to Shamar • • •It has even more far-reaching consequences than the inevitable accidental false positives that will *already* destroy the "AI detection" dreams 😃 It basically destroys the whole ML models as a service industry.
Or it would, if marketing didn't trump logic every single time ...
Gilbert Busana
in reply to Gilbert Busana • • •Please cease this.
The only thing those algorithms are good for, is to open backdoors for third parties to spy on our chat logs. Pretty much the European version of pegasus, but legal and implemented as a feature on all IM software.
The best way to prevent child abuse is good parenting, instead of parents giving phones with internet access to 5yo children who might become prey to predators on the internet.
Shamar
in reply to Gilbert Busana • • •Planting Undetectable #Backdoors in #MachineLearning Models
https://arxiv.org/abs/2204.06974
And if you think this is paranoic, read something about #Assange.
@EU_Commission
Planting Undetectable Backdoors in Machine Learning Models
arXiv.orgGilbert Busana
in reply to Gilbert Busana • • •surveillance state is not the solution, and you can't protect privacy while in the same time watching everybody.
#chatcontrol is wrong !