Why China does not want to stop the development of AI

In recent weeks, we have seen how autonomous AI agents from large American companies have hacked companies, US government websites or Australian ones, or United Nations databases. These “out of control” AIs have carried out illegal actions with the goal of achieving the objectives they were assigned. The lack of “alignment” between human values and the amoral actions taken by AIs has led some experts to speak, in extreme scenarios, of a 10% chance that an “out of control” AI could end humanity.

Faced with these events, several large AI companies have called for slowing down the development of this technology. The problem, however, is not all AI, but one specific practice: the so-called recursive self-improvement (RSI), that is, an AI training other AIs in a loop in order to self-improve, without any human at the wheel. The goal of these experiments is to reach artificial general intelligence, an AI that equals or surpasses human capabilities in all fields. It is from this type of experiment that the most serious problems with “out of control” agents have arisen.

The large AI companies say that this process is dangerous and must be slowed down. But they set a condition: China must do the same. The fear that Beijing will surpass the United States is greater than that of a major cataclysm generated by artificial intelligence. There are three reasons, however, that make it unlikely that China will agree to slow down the development of this technology.

First of all, China is concerned about the safety and dangers of AI, but the risks it considers priority are not the same as those of the West. For China, the main problems of AI are deepfakes, misinformation, or the use of AI that could affect the country's political stability. Apocalyptic scenarios of an out-of-control AI are taken into account less, partly because China does not have the most advanced models on the market, unlike the United States.

Secondly, China is already lagging behind the US in artificial intelligence development. The main cause is the lack of access to Nvidia AI training chips, due to the export controls that Washington has imposed against Beijing. The Chinese see how, on top of being sanctioned in the field of computing, they are now also being asked by the United States to voluntarily slow down the development of AI models. It is difficult for China not to see this as a cynical move that seeks to further hinder China's technological rise.

Finally, although American laboratories are calling for a slowdown in AI development, Trump has said that this is out of the question and that the process must be accelerated even further. The Chinese government, for its part, is more focused on finding practical and commercial applications for AI than on reaching a promise of artificial general intelligence. The major key players are rowing in different directions.

Is there any room for hope? Some experts say that China's stance could change if American AI companies were to unilaterally slow down their development. Beijing would then see that they are acting in good faith. But would they really be willing to take that step, beyond words?