Brussels, Berlin Companies should now label content created with artificial intelligence (AI). The EU Commission wants to prevent so-called “deepfakes” from spreading on the Internet. These are texts, images or videos that appear realistic but are completely invented.
On Monday, EU Commission Vice President Vera Jourova and Internal Market Commissioner Thierry Breton met with business representatives in Brussels to present their plan. Companies that build so-called generative AI into their search engines must ensure that they cannot be misused by malicious actors, Jourova said.
This applies to Microsoft and Google, for example, whose new search engines Bingchat and Bard are based on this technology. “Generative AI” refers to software that can independently create text and images from data.
Jourova said companies should label their AI content — starting now. Google boss Sundar Pichai recently assured her that this was technically possible. Users should be able to recognize immediately that a text or video was created by a robot.
Powerful AI applications like ChatGPT exacerbate the problem of deepfakes because a scam is becoming increasingly difficult to detect. They are still often harmless Internet jokes: in March, for example, a fake picture of Pope Francis in a hip down jacket caused a stir.
>> Read here: Like the grandchild trick, only with AI – deepfake scammers blackmail companies with fake boss voice
But the AI could also cause real damage: It would be conceivable that stock market speculators with a fake video of a company boss trigger a fall in prices – and benefit from the falling prices.
The new AI applications could create images of events that never happened, Jourova said. Politicians have to react to that.
First of all, the labeling should be done voluntarily. The Commission is simply adding a new AI passage to the EU code of conduct against disinformation that has existed for a year. 40 companies and organizations have signed the code, including Microsoft, Google, Meta and Tiktok.
SPD leader Esken calls for legal regulation
In Berlin, the voluntary regulation is met with skepticism. “Those who want to unsettle and divide our free and democratic societies with their disinformation will follow neither a voluntary nor a legal obligation to label AI-generated content,” SPD leader Saskia Esken told the Handelsblatt.
“In this respect, it would be wiser and more targeted to mark the authenticity of digital media content such as text, sound and image at the originator in a forgery-proof manner and thus strengthen the recognizability of reliable information.”
According to the SPD chairman, a voluntary agreement cannot replace statutory regulation.
(Photo: dpa)
Esken therefore only sees the EU initiative as a first step. A voluntary agreement cannot replace legal regulation, she said. Schleswig-Holstein’s data protection officer, Marit Hansen, also said that legal requirements were necessary “promptly”.
However, until the AI Act comes into force, it is “very sensible to implement measures such as labeling, but also impact assessments and protective measures” as a bridging measure. It is already clear that not all companies would participate. “So we won’t be able to rely on deepfakes and disinformation powered by AI being detectable.”
The general manager of the digital association Bitkom, Bernhard Rohleder, sees another problem. Almost half of Germans don’t know what AI actually is, he said. “Without this basic knowledge, a labeling obligation would be useless.”
>> Read here: EU urges rapid self-commitment by AI companies
The Commission sees the voluntary commitment only as a provisional measure. The advantage is that the code of conduct can be changed quickly without a legislative procedure, said Jourova. In the longer term, AI labeling is to be enshrined in law.
In an international comparison, the EU is a pioneer in AI regulation. The planned European AI law, the “AI Act”, is currently still being discussed in the European Parliament. The Council of Member States still has to agree. It will therefore probably only come into force in a few years.
EU Commission warns Twitter boss Elon Musk
The “Digital Services Act” (DSA), on the other hand, will apply from August 25th. With the law, the EU wants to curb the problem of hate speech and fake news on the Internet in general. Until then, all social networks must disclose the criteria according to which they display content to their users.
So far, the balance of the fight against fake news has been sobering. “There’s still far too much dangerous disinformation out there,” Jourova said. The big platforms would have to build more capacity to check facts and add warnings to content. As an example, she cited the disinformation by Russia in the Ukraine war.
It’s not about bans for us, it’s about reducing risks. EU Commissioner for Internal Market and Services, Thierry Breton
The Kremlin is trying to use its propaganda to undermine democracy in the EU, she said. Especially in Eastern European countries, the online platforms would have to check and sort out more closely. The EU suffered a setback a few weeks ago when Twitter, under the leadership of new owner Elon Musk, withdrew from the voluntary EU code of conduct. A big mistake from Jourova’s point of view.
“Twitter chose confrontation,” said the Commission Vice-President. When the DSA comes into force in August, the commission will look very closely to see whether Twitter is complying with the regulations. The signers of the code of conduct, on the other hand, could hope for the goodwill of the supervisors. In June, the Commission wants to start stress tests with Twitter and other platforms to see to what extent they are already DSA-compliant.
Breton travels to Silicon Valley
Commissioner Breton will open the EU liaison office in Silicon Valley in two weeks. Talks are also planned with OpenAI, the company behind the supposed all-rounder ChatGPT, and Nvidia, the leading manufacturer of AI chips. Breton sees AI as a “fantastic innovation”, but his message to US companies is: If you want to be active in the European market, you have to comply with European rules.
“It’s not about bans, it’s about reducing risks.” Breton compares the development of ever more powerful AI systems with the invention of the automobile – and the EU rules with the introduction of seat belts.
Just as a seat belt does not prevent accidents, the Brussels mix of voluntary codes of conduct and regulatory provisions will not reduce the AI risk to zero. But the commissioner hopes that the risk of injury – which in the case of AI affects society as a whole – will drop significantly.
More: ChatGPT – what you should know about OpenAI’s AI