The states of the European Union adopted a historic document for the regulation of artificial intelligence, which should prevent it from replacing humans and destroying the credibility of digital content. Cybersecurity experts explained to “Adevărul” that the Directive will be an AI GDPR that will try to protect the way data circulates on the Internet to protect people from deepfake or fake news. In addition, employers will also have restrictions if they use both human labor and artificial intelligence.
The Law Governing AI – PHOTO Shutterstock
Last week, the member states of the European Union adopted a historic document for the regulation of artificial intelligence, which should prevent it from replacing humans.
“This “AI Law” is a fundamental step, establishing the first rules on the planet on artificial intelligence, to make it safer and respectful of human rights“, stated the Belgian presidency after the adoption of the law. For his part, European Commissioner Thierry Breton praised the initiative, which he considers “a historic regulation, a world premiere“.
The rules imposed in the field of AI by the EU
Among other things, the Directive will require employers to notify employees before implementing “high-risk artificial intelligence systems“, such as education, human resources or public order. In addition, it will require employers to implement their supervision by experts, immediately suspend the use of the system and notify the supplier, importer or distributor as well as the relevant market surveillance authority if an issue such as discrimination occurs and keep AI records for 6 months
In the case of generative artificial intelligences, which can produce texts, sounds or images on demand, the rules will be imposed to ensure the quality of the data used in establishing the algorithms and to verify compliance with European copyright legislation. Also, the developers will have to specify that the sounds, images and texts produced by the AI are artificial.
The document bans the use of artificial intelligence in social scoring systems, which rate people based on behavior or socio-economic status. Practices that covertly influence vulnerable people or groups to change their behavior will also be prohibited.
Deepfake protection
Alexandru Panait, cyber security expert, and Alexandru Budișteanu, inventor and AI expert, explained to “the truth” the danger of artificial intelligence and how the new law will help.
“For example, AI legislation can protect people from deepfake or fake news. Companies should be regulated – how they use AI and regulated how media such as social media or YouTube validate this information. But the internet is an open scene, where anyone could publish anything, so regulation should address companies developing AI, but also the way information is disseminated online. This is more importantly than regulating AI itself to limit the leverage through which bad actors can take advantage of this technology. You can design a computer program that doesn't let you use AI negatively And the best option is to move the regulations into a digital format, in the actual source code. “explained Alexandru Panait.
These programs can help create a digital identity. “Once you have a digital identity, either national, European or global, that you trust, you know that the source that published that video, text, is an authentic source and you can trace where the information came from in the first instance“, states the cyber security expert.
The adopted directive will basically be a kind of GDPR that aims to protect the way data flows on the Internet.
“GDPR is just a legal lever in case something bad happens. But nothing guarantees that your data is protected. On the AI side, you can regulate the way the AI is trained, the way it is disseminated, impose certain fines if the AI is used for negative purposes, but that's about it, because the online environment is extremely dynamic and volatile, and legislation for such a thing is very difficult to conceive“, says Alexandru Panait.
And Ionuț Budișteanu reinforces: “An algorithm or mathematical formula cannot be regulated. Any user can use it and make whatever kind of product they want, even the data center doesn't know if that AI model has been properly trained following GDPR policies or not. The biggest risk is that the Internet will not be what it is today. In the future, people will lose trust in digital content”.
Specifically, he believes that in the future the internet risks being full of fake news, written by bots, and fake reviews to fool people. “There will be fake accounts on social networks created with the help of AI, fake influencers used to trick people, and in this way, people's trust in what appears on the Internet will disappear“, he explains.
The fines provided for in the recently adopted directive range from 7.5 million euros or 1.5% of a company's annual turnover up to 35 million euros or 7% of a company's total annual turnover.
How will the legislation be enforced? “Every regulation comes with a lot of enforcement jobs, which means people have to check that they are being followed. Things would be simplified if the regulations are made at the code level. Somehow the goals of the AI law could be achieved through source code, not texts, because when you have texts you need a lot of resources to make sure the regulations are followed, to train people“, says Alex Panait.
Ionuț Budișteanu emphasizes: “Certain actors in the field of AI can be regulated very easily, but others are difficult to detect and regulate. In conclusion, things are complicated enough to be regulated“.
The AI Law, discussed 3 years ago
The AI law was discussed by the European Commission as early as three years ago, but the appearance of the ChatGPT bot at the end of 2022 upset the calculations of EU experts. The chatbot and other applications that can create deepfake sounds, images or videos have shown both the potential of AI and the dangers in this field. However, major countries such as France and Germany have put the AI law on hold. The two countries feared that the new regulations would cut off the momentum of European start-ups in the field, such as Mistral AI and Aleph Alpha.
“Regulation of AI is necessary, but I believe it will be regulated according to the interests of the political leaders who design these laws. I don't think the technological advance will be limited. (…) There is always an impact. The more restrictive a regulation is, the more difficult start-ups will be, burdened by that regulation. Each provision means new jobs, new costs, more work for a start-up. The simpler and fewer the regulations, the more efficient it is for a start-up to grow and develop, the more they are, the more difficult it is for this start-up to grow“, explains Alex Panait, who founded a software development company in 2019 and will soon establish a branch in the USA.
