Beijing dreams of a ChatGPT that respects “the values ​​of socialism”

Is artificial intelligence à la ChatGPT soluble in Chinese-style socialism? This is what Beijing wants to believe, which has just published new regulations to oversee and promote the development of this technology in accordance with the doctrine of the regime, reported The New York Times, Wednesday, April 26.

These measures, submitted to the authorities of the Chinese Communist Party on April 11l, make China the first country to impose rules on this sector, which has exploded since the release, in November 2022, of ChatGPT, the conversational agent developed by the company OpenAI.

Fear of AI slippages

Beijing wanted to react as quickly as possible because “as in all countries, Chinese leaders have understood that this technology could have a profound impact on society”, assures Guangyu Qiao-Franco, specialist in emerging technologies in China at the University of Radboud (Netherlands).

Impossible, therefore, for a regime so eager for control, to leave these conversational agents in total freedom. Especially since Beijing knows what it could give: in February 2022, ChatYuan, the first Chinese alternative to ChatGPT,had to be removed urgently after having stated, in particular, that the Chinese economy was in “poor state” and that the conflict in Ukraine was a “war of Russian aggression”, while China takes up the Russian argument of a “special military operation” .

The new rules aim to avoid these slippages of the AI. Concretely, Chinese companies developing alternatives to ChatGPT must ensure that their algorithms do not produce discriminatory content, do not violate the right to privacy of Internet users, and do not spread false information, lists le South China Morning Post.

Concerns that concern governments around the world in the face of the challenges of AI.

But article 4 of these new rules also contains an obligation to design AIs “that respect the central values ​​of socialism [chinois]”. A call to put chatbots on the page of Mao Zedong’s Little Red Book?

In fact, “there is a list of concepts that are included in the central values ​​of socialism”, assures Guangyu Qiao-Franco. We find there, in particular, democracy, prosperity, equality and justice, patriotism, the sense of duty or even the attachment to the value of work.

Pleasing Xi Jinping?

A real potpourri of great values ​​that could be shared by most Western countries. But it is difficult to deduce from this the rules to follow in terms of the development of algorithms for an AI compatible with this “socialism”. “It is true that these are rather vague concepts”, recognizes Guanyu Qiao-Franco.

In reality, “we must conceive of this idea of ​​’central values ​​of socialism’ as an umbrella concept which allows the regime a certain flexibility as to what is authorized or not”, assures Xin Sun, a specialist in Chinese economic and industrial policy.

Faced with a technology that evolves as quickly as these conversational AIs, Beijing does not know exactly where to put the cursor of its censorship. The “core values ​​of socialism” are sufficiently vague to adapt to all situations.

>> To read also: ChatGPT: putting AI on hold, “an existential issue”?

The use of this reference is no less surprising. Indeed, authorities have already enacted rules for “good AI governance” in 2019 and as part of an “AI code of ethics” in 2021. No sign of “core values ​​of socialism” in these two cases.

In 2023, the regulator therefore decides to go further. It could have been inspired by “Xi Jinping who makes more and more references to these values. It would then be a way of showing the leader that we are trying to follow his example”, suggests Guanyu Qiao-Franco.

In this hypothesis, AI would join the growing number of parts of the economy subject to a ideological takeoverbecame one of the priorities of the president.

This will not make it easier for Chinese companies to compete with ChatGPT. How to create a “socially compatible” AI? “A priori, it will be necessary to create a filter at the entrance of the algorithm and another at the exit”, estimates Xin Sun.

Doomed to be behind the United States?

This double censorship will first justify a sorting of information from the database in which the AI ​​will draw to find its answers. Then, “it will probably be necessary to put in place a system of verification of the final answers to ensure that they do not contain anything contrary to the regulations”, specifies Xin Sun.

There remains the great mystery of these AIs. “The problem is that we have absolutely no idea why these machines opt for one answer rather than another based on the data that we provide them”, explained, at the end of March, to France 24, Joseph Sifakis, director of research at the University of Grenoble and the only French holder of a Turing Prize (equivalent to the Nobel Prize for artificial intelligence). In other words, no engineer will be able to 100% guarantee to the CCP that their chatbot will remain strictly within the framework of the “core values ​​of socialism”.

To this first uncertainty, we must also add that surrounding the new regulations. “What Chinese entrepreneurs are asking for the most is clarity in what they have the right to do or not”, assures Guanyu Qiao-Franco. This vagueness around the “central values ​​of socialism” forces “companies to move forward with great caution, which risks putting them behind the competition, especially in North America”, maintains Xin Sun.

What put the Chinese authorities in front of their own contradictions. Indeed, AI is on the list of priority technology sectors for Beijing in its battle for global leadership. It is even central because the authorities “consider AI a bit like electricity, that is to say a technology that allows development in other priority areas [comme la robotique ou les biotechs, NDLR]”, souligne Guanyu Qiao-Franco.

Beijing would therefore be shooting itself in the foot with these new rules, which risk slowing down innovation in a critical area. But “the other countries will also think about the best way to regulate this sector”, assures Guanyo Qiao-Franco, which would limit China’s initial delay

It is still necessary that the authorities and the professionals of the sector agree on what is a conversational agent compatible with Chinese-style socialism. This will be the subject of consultations that the government intends to conduct with Chinese groups that want to overshadow ChatGPT. And between Alibaba, Baidu or Tencent, they are numerous and powerful.