It is one of the most important topics at the meeting between US President Donald Trump and Chinese President Xi Jinping tomorrow: artificial intelligence. The US and China are competing for leadership in AI, which both countries see as crucial for their economic and military power.
Prior to the talks, Washington proposed a kind of hotline to share information about serious threats from artificial intelligence. This was prompted by warnings from AI leaders that the development of AI models is progressing too fast and could even wipe out humanity.
Trump and Xi consider that warning exaggerated, and Trump also states that he does not intend to lose the AI race to China. Are the world leaders really willing to make rules against AI weapons, or is it still waiting for a disaster?
Nuclear weapons
Koen Aartsma, researcher at the Clingendael Institute, thinks agreements are "quite possible." "We saw that before with nuclear weapons. People realized that those developments could be incredibly destructive. Unfortunately, bombs in Japan were needed first. But still, if you want to see the glass half full, it partly succeeded to establish rules."
Two years ago, Aartsma already wrote a scenario about AI in which he spoke of 'mass-disruptive weapons.' "There is quite a possible scenario in which AI itself, or on command, starts doing things that have an enormous disruptive effect." Think of shutting off power or water supply.
Wiping out humanity: nonsense, or not?
According to Anthropic CEO Dario Amodei, the developer of the AI model Claude, AI agents could take over the entire internet within twelve months. "If we do not stay on top of this, the development of AI could surpass our ability to understand and control these systems," Amodei said in an essay.
AI specialist Michiel Bakker shares the concerns. "AI developments can have extremely large consequences. Last summer we already saw how much damage those powerful systems can cause." At that time it appeared that an AI program from the company OpenAI accidentally hacked a competitor after breaking out of a test environment.
Tech expert Bert Hubert sees less danger: "I do not take that warning about wiping out humanity seriously. It is a pattern: AI companies issue a warning but then still bring their product to market. So it seems like another marketing trick to me. Besides, an AI is only as dangerous as you make it yourself. So if it is really that dangerous, why don't you take action yourself?"
Anthropic CEO Amodei advocates safety checks at AI companies and more global cooperation on AI regulation. According to Bakker, it would be illogical for tech companies to ask for rules themselves if they do not really see a danger. "New regulations could also work against their interests. So I do believe in the sincerity of that warning."
Safety or winning the AI race?
In any case, Aartsma does not think AI companies will slow down on their own. "They see that developments are moving fast and that they do not have full control, but they are also in that AI race with each other. That is why agreements during the meeting between Xi and Trump are very important."
Those countries have a shared interest, says Aartsma. "The US and China are both leading in AI developments, which gives them an enormous economic and geopolitical boost. They would rather not want other countries to also catch up on that point. So behind the scenes, without making it public, they could make agreements about this."
Europe is far behind China and the US in the AI market; watch this video about that:
The question is whether those agreements are sufficient to contain the potential dangers of AI. Aartsma: "Agreements about AI are even more difficult than about nuclear weapons, given the geopolitical era we live in. Moreover, in an AI attack it is not always clear who is behind it. It could also be the AI itself."
AI is also not a tool with only clearly negative effects, like nuclear weapons. "It can, for example, help detect all kinds of diseases. So how can we ensure that we can maintain that side, but not the negative side?"
Is Trump persuadable?
Bakker is hopeful and also thinks that America can guide AI developments properly. "There are only a few players with computing power, and they are all American companies, think of Anthropic, OpenAI, and Meta. The US can slow down and at the same time maintain a leading position over China. The main question is how Trump can be convinced of that."
Aartsma: "I think China and the US will take small steps, more promises and agreements to continue talking with each other."