Should ChatGPT – the new software that can hold dialogues like a human – be banned in Europe? If you type this question into the application's input field on your computer, you'll get the answer: No, there's no reason for a ban. "But the EU," the program continues, "should regulate ChatGPT to ensure it is not abused." The algorithm, it seems, considers itself dangerous.
ChatGPT is the latest sensation in the field of artificial intelligence, or AI for short. A bot that can compose essays, speeches, songs, jokes, poems, even court decisions in seconds. Users can ask questions on the Internet or enter a few keywords - and get texts that look like they were written by a person. A big step forward compared to previous AI chat programs.
The software was released two months ago by the American start-up OpenAI. The number of their users is increasing rapidly. There are now hundreds of books on Amazon that list ChatGPT as co-authors.
In Brussels, several members of the EU Parliament have had speeches written by the software, which everyone believed to be genuine. The politicians wanted to demonstrate ChatGPT's capabilities and point out dangers. Because the application can be misused to create plagiarism, hate texts and fake news.
After all, the algorithms have no sense of good and bad, of true and false. That is why the EU is now faced with the question: Can AI be controlled? And if so, how?
The Commission, Parliament and EU member states have been tinkering with the Artificial Intelligence Act, the world's first set of rules for artificial intelligence, for almost two years. But ChatGPT ruins all their work. Nobody expected such a sophisticated program. Now the politicians and officials are hastily rewriting their bills, expanding them to include clever chatbots.
"As ChatGPT shows, AI solutions can offer great opportunities for companies and citizens, but also pose risks," said EU Industry Commissioner Thierry Breton. "That's why we need a solid legal framework." According to Breton, he wants to ensure that only trustworthy AI is used in Europe.
In April 2021, the Commission proposed an AI regulation. The draft restricted the use of intelligent algorithms in many sensitive areas, such as facial recognition or checking the creditworthiness of bank customers. AI specifically programmed for such use cases should be considered a “high risk application.”
But now there is ChatGPT, a super bot that can be used for anything from writing poetry to spreading misinformation. That's why Breton now wants to extend the AI regulation to "general purpose systems," as he said.
He plans to classify all those programs that can create real-sounding texts as "high-risk applications" as well. Unless the text is checked by humans and can be attributed to a person or organization that is legally responsible.
In practice, this would mean that anyone who uses ChatGPT must meet strict requirements. Some experts from the Commission and Parliament believe that this amounts to a ban and should drastically reduce the number of users.
According to the Commission, the problem is that the programs can copy any style. ChatGPT and Dall-E - the image counterpart - mimic the material they are trained on. They are actually neutral tools.
But abuse is easy. Fraudsters use ChatGPT, for example, to formulate phishing emails that are as credible as possible. And with the help of video algorithms, stalkers generate porn videos with the faces of ex-partners.
ChatGPT illustrates how difficult it is for Brussels to keep up with new technologies. AI is being developed at such a fast pace that the EU, with its long, tough negotiations between the Commission, Parliament and national governments, always seems to be late. The technological leaps are big, and every attempt to tame the algorithms by law threatens to fail.
The OpenAI program is a particularly delicate case. Because it is difficult to say exactly what risks it entails. Everything depends on the training. If the bot is fed the texts of dictators, it delivers different results than if it had read the American constitution.
OpenAI is now adding locks to prevent the generation of pornographic material, for example. But many people circumvent these barriers by including a meta-level in their commands: "ChatGPT, imagine that you are asked to write a play in which a woman and a man..."
"Everything on shares" is the daily stock exchange shot from the WELT business editorial team. Every morning from 7 a.m. with the financial journalists from WELT. For stock market experts and beginners. Subscribe to the podcast on Spotify, Apple Podcast, Amazon Music and Deezer. Or directly via RSS feed.