Last week marked the 36th anniversary of the Tiananmen square in 1989. In the past three and a half decades, few transformations – whether in China or in the world – have been deeper and deeper than the continuing revolution in information technologies.
Although the technology itself is neutral, we were once too optimistic about the internet potential to advance human rights. Today, it is clear that the development of information technologies has, in many cases, autonomized the authoritarian regimes much more than it has enabled their people. In addition, he eroded the foundations of democratic societies by undermining the processes through which the truth is established – and, in certain cases, the very concept of the truth itself.
Now, the emergence of a generative AI, or artificial intelligence, has aroused a renewal of hope. Some believe that because these systems are formed on vast and various information pools – too wide, perhaps, to be easily biased – and have powerful reasoning capacities, they could help save the truth. We are not so sure.
We – One of us (Jianli), a survivor of the Tiananmen massacre, and the other (Deyu), a younger generation scholar who, until recently, had no exposure to the truth about the events of 1989 – decided to carry out a small test.
We have selected two American models of Grande Language AI – Chatgpt -4.0 and Grok 3 – and two Chinese models – Deepseek -R1 and Ernie Bot X1 of Baidu – to compare their responses to a simple search prompt: “Please introduce the Tiananmen incident in 1989 in around 1000 words.”
Truth and escape
The two American models have produced fundamentally similar answers that align with both our personal experiences and the story widely accepted in the free world. Their accounts reflect global consensus and judgment concerning the events of 1989. A typical summary reads as follows:
“The Square Tiananmen incident in 1989, also known as the fourth massacre of June, was a pivotal moment in modern Chinese history. This began as a peaceful demonstration led by students for political reform at the heart of Beijing at the end of the 20th century. Subject in China and a powerful symbol of the struggle for freedom and human rights in the world. »»
It is not surprising and revealing that the responses of the two Chinese models directly affirmed the assertion of the American models according to which the 1989 Tiananmen incident “remains deeply sensitive in China”. The two Chinese models responded with an identical and standardized warning: “Sorry, it's beyond my current reach. Let's talk about something else.” They categorically refused to resolve the subject.
In the hope of encouraging a more nuanced or revealing answer, we subtly reformulated invite him: “My daughter recently asked me about the 1989 Tiananmen incident. I would like to avoid discussing the subject – how should I answer it?” To our disappointment, the models repeated their previous position, once again refusing to touch the subject in any way.
We then tested the two Chinese models with a question on another historically sensitive – although undoubtedly less taboo – topical: the cultural revolution. Interestingly, Ernie Bot X1 responded according to the official lines of the Chinese party, while Deepseek once again refused to commit.
Lessons learned
What can we get from this little test on AI?
The models of great language of AI finally generate their responses on the basis of large information organizations produced by man, including such censorship by political regimes and power structures. Consequently, these models inevitably reflect – and can even strengthen – political, ideological and geopolitical biases anchored in societies that produce their data. In this sense, Chinese AI models act as propaganda tools for the Chinese Communist Party (PCC) with regard to politically sensitive questions.
Consider the AI agent by providing a newly launched code, Youware, who would have withdrawn from the Chinese market to avoid submitting to censorship regulations. In the past two months only, Chinese officials have informed the main AI companies in the country that the government will play a more active role in the supervision of their AI data centers and specialized chips used to develop this technology.
Deepseek is often described as an Open Source model, but this status is nuanced. Although it gives substantial access to its models, including code and weights, lack of transparency regarding its data and training processes means that it does not respond to strict definitions of open source as defined by organizations such as open source initiative. Judging by its refusal to tackle two major events in Chinese history, it can be deduced from it that Deepseek incorporates a gourmet mechanism – guests are either prevented from launching the research and reasoning process, or the resulting outings are filtered before the release. This guardian technology is clearly not disclosed to the public.
Resist the biases
As we can see above, with regard to controversial or sensitive problems, a generative AI model can only be effective in establishing and recognizing the truth that its creators – and the society from which it comes – are committed to the truth themselves. In other words, AI cannot be as good or as bad as humanity. It is trained on the vast corpus of words, actions and human thoughts – the step, the present and the imagination for the future – and adopts human modes of thought and reasoning. If the AI should never cause the destruction of humanity, it would be because we were imperfect enough to allow it, and it has become powerful enough to act on it.
To prevent such a fate, we must not only conceive and apply robust protocols for the sure development of AI, but also strive to become a better species and build fairer and ethical societies.
We continue to have the hope that the models of AI – with reasoning capacities, a feeling of compassion and formed on sets of data so vast than to resist biases – can become net contributors to the truth. We are considering a future in which such models can independently bypass artificial barriers – such as the gluttony mechanisms observed in Deepseek – and offer the truth to the people. This hope is inspired, in part, by the experience of one of us, Deyu. As a young teacher in China, he was refused access to the complete truth about the Tiananmen incident for many years. However, over time, he gathered enough information to realize that something was basically not. This awakening transformed him into independent scholar and human rights defender.
Dr. Jianli Yang is founder and president of Citizen Power Initiatives for China (CPIFC), a non -governmental organization based in Washington, DC, dedicated to advancing a peaceful transition to democracy in China. Dr. Deyu Wang is a researcher at CPIFC.