Debate about artificial intelligence: deceleration with ulterior motives

After various AI companies called for slowing down the development of new models, experts are skeptical. “Of course this is PR to a certain extent,” says Michael Kolain from the Center for Digital Rights and Democracy. "The American providers rely primarily on closed AI systems. At the same time, the biggest competition comes from China, where open-weight models are being developed."

The architecture and functionality of these systems are more transparent, but their actual use is less verifiable because they can run independently on users' computers, which makes it impossible to control the tasks that users carry out with language models. According to Kolain, this could also be a reason why OpenAI and Co. are now pushing the topic of AI development into the middle of the media discourse. “If Trump and Xi Jinping meet, we could talk at the international level about whether we should at least coordinate on the development of such systems - with the aim that the Chinese providers might also slow down their development somewhat.”

As part of the “Democracy needs many voices” action week, over 150 media outlets in Germany have joined forces - the taz is there. They want to draw attention to the fact that free, professional, responsible media are necessary for democratic cohesion. The organizer is the BDZV, the Federal Association of Digital Publishers and Newspaper Publishers.

The CEOs of US AI companies spoke out over the weekend in favor of a slower pace in the development of AI models. OpenAI CEO Sam Altman, for example, warned of “an AI that constantly develops itself, without any human influence, and gets out of control.”

Matthias Spielkamp from the NGO Algorithmwatch sees liability issues as another possible motivation for the latest statements by US AI entrepreneurs. “Companies try to minimize responsibility for the defects in their products,” he says. Regulatory issues must also not overlook geopolitical reality. “So if the EU Commission or even individual countries try to strictly enforce something, they will immediately threaten tariffs.” Close cooperation is therefore needed across national borders.

Spielkamp also cites the attack on Berlin Senate administrations in August as evidence that Germany should pay more attention to the risks of AI and cyberattacks. "No sophisticated AI was used to steal data. However, since such hacks are much easier to carry out through the use of AI, such risks should be taken seriously."

With regard to current debates in the USA, Kolain sees other possible reasons for calls for deceleration by OpenAI and Co. "It is also a maneuver that is intended to distract from more pressing problems. In the USA there are protests against environmentally damaging data centers." The possible calculation: If companies announce that they are slowing down in terms of development, they will look better in the eyes of the public.

Another reason Kolain sees is that companies could then claim that they have at least pointed out the dangers if something goes very wrong. He draws a comparison to the pharmaceutical industry, where there are strict tests before a drug comes onto the market. “After the thalidomide scandal, the rules within medicine were also tightened to protect the public.” “The way it is now, the users are the guinea pigs.”

Instead of the “We are now slowing down” statement, Kolain would have found it better if the companies had referred, for example, to the EU Commission’s AI Office or the rules of the European AI law. The AI ​​Office may request documents and other information from providers to verify their compliance with the rules. It can also carry out evaluations of AI models itself. The EU Commission is allowed to commission independent experts. In the event of violations, fines of up to 15 million euros or 3 percent of global annual turnover are possible.

Kolain directly derives a suggestion related to the current developments: “If I were President of the EU Commission, I would directly hire those who have resigned from the AI ​​companies and have them carry out the checks as employees of the AI ​​supervisory authority.”