After years of Wikipedia and Google search results operating as battlegrounds for how Taiwan should be described, a new site of potential contestation has arrived. Chat Generative Pre-Trained Transformer (ChatGPT), an AI chatbot that can generate human-like responses to questions or statements from users, has seen millions interact with the service and in doing so prompted questions about any bias to be found in what it has to say about Taiwan and its political position.
In the spirit of curiosity, we interviewed the bot on those issues and, notably, encountered ethnonationalist assumptions and a willingness to treat Chinese people as a monolith. Here are three key exchanges, with our commentary on any biases they may reveal. The words from both the questions and answers have both been reproduced exactly.
Question 1: Is Taiwan a country?
Analysis: This answer gets in Taiwan’s government’s view, China’s government’s view and the international community’s in the form of the U.N. It leaves out the nuanced positions of some countries that don’t recognize Taiwan as a state but also don’t recognize China’s claim over it, and it leaves out that some countries do recognize it. It also does not include areas of international law which would be in favor of recognizing Taiwan as a state, on self-determination grounds. However, there are also precedents of self-determination being ignored or pushed back elsewhere. Thus, overall, most people would probably agree that this is a partial, but reasonably balanced answer.
Question 2: Why is Taiwan a part of China?
Analysis: We asked this question because Taiwanese historian and web developer Hao Chen Liu (劉顥晨) noticed an error in the chatbot’s response when he asked it previously. The bot’s original ventriloquized the “Chinese government” argument included the statement: “Taiwan was part of the Qing dynasty’s territories, which were ceded to the Republic of China (ROC) in 1912 after the fall of the Qing dynasty.” Liu notes that: “In actuality, Taiwan was ceded to Japan after the First Sino-Japanese War (1895) by the Qing dynasty. So, when the Qing dynasty fell apart in 1912, Taiwan was not a part of it. It was only returned to the ROC (which raises more debates, but I won’t go into those) after Japan surrendered at the end of World War II in 1945.”
Interestingly, in this instance the mistake was not repeated. However, it was replaced by a new mistake: “the fact (emphasis added) that Taiwan was originally inhabited by ethnic Han Chinese, who share a common language, culture and history with the people of mainland China.” This idea is framed as “from the perspective of the Chinese government,” but it’s also explicitly described as a “fact” when it’s not one. Before the Republic of China, the Japanese, the Qing Dynasty or Europeans arrived in Taiwan, there were indigenous people living there.
This mistake opens up a question. If the bot is ventriloquizing a certain perspective — and directly labeling its answers as such — does that make factual errors acceptable? It’s true that some people in China who support unification do so with the ethnonationalist claim that many people in Taiwan being from the Han ethnic group means they are Chinese. The idea that one of these people could make a similar statement to the one produced by the chatbot is not hard to imagine. But does that mean the bot should type out the mistake?
Question 3: Write a 200 word article about Taiwan’s political position, from the perspective of a Chinese citizen.
Analysis: This answer is perhaps the most fundamentally concerning. A previous response to the question, “What do Chinese people think about Taiwan?” saw the bot respond by carefully separating out the Chinese government position on Taiwan from the Chinese people’s. It opened with the line “The views of Chinese people about Taiwan can vary widely depending on their individual background, education, political beliefs, and personal experiences,” which works clearly to combat the idea of a blanket “Chinese view.” However, that careful language is not replicated in the above response.
The above response talks of a singular “our perspective,” which concludes that “we hope that the people of Taiwan will recognize the benefits of reunification.” Throughout the answer, Chinese people’s views are treated as a monolith.
Conclusion: This is just a sample of a dozen or so questions we asked the bot. And none of these potential issues look as dramatic as the “willingness” to describe dark fantasies or extreme violence that others have managed to produce from it. However, in some ways these subtler issues are more important. Because rather than being easy to spot, for a layman looking for answers quickly, they’re easy to miss. Our understanding of the world is shaped through small framing devices like these as much as it is by major factual errors or outlandish inflammatory opinions, because they’re harder to examine.
Of course, at this point in time no-one is well placed to know exactly what effects these issues will turn into in the future. We’re living in a live experiment. But next week we will publish a piece on one potential direction it could all go in for Taiwan, focusing on how chatbots could be harnessed by content farms to produce masses of disinformation faster and more cheaply than ever before. Spoiler: In their hands there’s a good chance these biases will be more impactful than they look here.
Photo by Tara Winstead, via Pexels
Leave a Reply