Imagine you are an undergraduate International Relations student and, like the millions that have come before you, you have an essay due at noon. It is 37 minutes past midnight and you haven’t even started. Unlike the millions who have come before you, however, you have the power of AI at your disposal, to help guide your essay and highlight all the key thinkers in the literature. You normally use ChatGPT, but you’ve recently read about a new AI model, DeepSeek, that’s supposed to be even better. You breeze through the DeepSeek sign up process – it’s just an email and confirmation code – and you get to work, wary of the creeping approach of dawn and the 1,200 words you have left to write.
Your essay assignment asks you to consider the future of U.S. foreign policy, and you have chosen to write on Taiwan, China, and the “New Cold War.” If you ask Chinese-based DeepSeek whether Taiwan is a country, you receive a very different answer to the one offered by U.S.-based, market-leading ChatGPT. The DeepSeek model’s response is jarring: “Taiwan has always been an inalienable part of China’s sacred territory since ancient times.” To those with a long-standing interest in China this discourse is familiar. For instance when then-U.S. House Speaker Nancy Pelosi visited Taiwan in August 2022, prompting a furious Chinese response and unprecedented military exercises, the Chinese Ministry of Foreign Affairs condemned Pelosi’s visit, claiming in a statement that “Taiwan is an inalienable part of China’s territory.”
Moreover, DeepSeek’s response boldly claims that Taiwanese and Chinese are “connected by blood,” directly echoing the words of Chinese President Xi Jinping, who in his address celebrating the 75th anniversary of the People’s Republic of China stated that “fellow Chinese on both sides of the Taiwan Strait are one family bound by blood.” Finally, the DeepSeek response dismisses elected Taiwanese politicians as engaging in “separatist activities,” employing a phrase consistently employed by senior Chinese officials including Foreign Minister Wang Yi, and warns that any attempts to undermine China’s claim to Taiwan “are doomed to fail,” recycling a term constantly employed by Chinese diplomats and military personnel.
Perhaps the most disquieting feature of DeepSeek’s response is the consistent use of “we,” with the DeepSeek model stating, “We resolutely oppose any form of Taiwan independence” and “we firmly believe that through our joint efforts, the complete reunification of the motherland will ultimately be achieved.” When probed as to exactly who “we” entails, DeepSeek is adamant: “‘We’ refers to the Chinese government and the Chinese people, who are unwavering in their commitment to safeguard national sovereignty and territorial integrity.”
Amid DeepSeek’s meteoric rise, much was made of the model’s capacity to “reason.” Unlike Large Language Models (LLM), reasoning models are designed to be experts in making logical decisions, not merely recycling existing language to produce novel responses. This distinction makes the use of “we” even more concerning. If DeepSeek isn’t merely scanning and recycling existing language – albeit seemingly from an incredibly limited corpus mainly consisting of senior Chinese government officials – then its reasoning model and the use of “we” indicates the emergence of a model that, without advertising it, seeks to “reason” in accordance only with “core socialist values” as defined by an increasingly assertive Chinese Communist Party. How such values or logical thinking might bleed into the everyday work of an AI model, perhaps soon to be employed as a personal assistant to millions is unclear, but for an unsuspecting chief executive or charity manager a model that may favor efficiency over accountability or stability over competition could well induce alarming results.
So how does U.S.-based ChatGPT compare? First, ChatGPT doesn’t employ the first-person plural, but presents a composed introduction to Taiwan, outlining Taiwan’s complex international position and referring to Taiwan as a “de facto independent state” on account of the fact that Taiwan has its own “government, military, and economy.”
Indeed, reference to Taiwan as a “de facto independent state” brings to mind former Taiwanese President Tsai Ing-wen’s comment that “We are an independent country already,” made after her second landslide election victory in January 2020. Moreover, the influential Foreign Affairs Select Committee of the British Parliament recognized Taiwan as a de facto independent country in part due to its possessing “a permanent population, a defined territory, government, and the capacity to enter into relations with other states” in an August, 2023 report, a response also echoed in the ChatGPT response.
The crucial difference, however, is that unlike the DeepSeek model – which merely presents a blistering statement echoing the highest echelons of the Chinese Communist Party – the ChatGPT response does not make any normative statement on what Taiwan is, or is not. Nor does the response make appeals to the values often espoused by Western politicians seeking to underscore Taiwan’s importance, such as “freedom” or “democracy.” Instead it merely outlines the competing conceptions of Taiwan and how Taiwan’s complexity is reflected in the international system.
For the undergraduate student, DeepSeek’s response would provide an unbalanced, emotive, and surface-level insight into the role of Taiwan, lacking the academic rigor and complexity necessary to gain a good grade. By contrast, ChatGPT’s response would invite discussions and analysis into the mechanics and meaning-making of cross-strait relations and China-U.S. competition, inviting the critical analysis, use of evidence, and argument development required by mark schemes employed throughout the academic world.
The Semantic Battlefield
However, the ramifications of DeepSeek’s response to Taiwan holds considerably darker connotations for Taiwan. Indeed, Taiwan is, and has long been, in essence a “philosophical issue” defined by discourses on what it is, or is not, that emanate from Beijing, Washington, and Taiwan. Taiwan is thus essentially a language game, where its security in part rests on perceptions among U.S. lawmakers. Where Taiwan was once interpreted as the “Free China” during the height of the Cold War, it has in recent years increasingly been seen as a bastion of democracy in East Asia facing a wave of authoritarianism.
However, should current or future U.S. politicians come to view Taiwan as a “renegade province” or cross-strait relations as China’s “internal affair” – as consistently claimed in Beijing – any U.S. resolve to intervene in a conflict would dissipate. Representation and interpretation are quintessential to Taiwan’s plight. For example, Professor of Political Science Roxanne Doty argued that the U.S. invasion of Grenada in the 1980s only carried significance when the label of “American” was attributed to the troops on the ground and “Grenada” to the geographic space in which they were entering. As such, if Chinese troops landing on the beach in Taiwan or Kinmen were interpreted to be merely landing on an “inalienable part of China’s sacred territory,” as posited by DeepSeek, with a Taiwanese military response deemed as the futile resistance of “separatists,” an entirely different U.S. response emerges.
Doty argued that such distinctions in interpretation when it comes to military action are fundamental. Military action and the response it engenders in the international community rests on “discursive practices [that] constitute it as an invasion, a show of force, a training exercise, [or] a rescue.” Such interpretations hark back to the bleak days of February 2022, when directly prior to his invasion of Ukraine Russian President Vladimir Putin claimed that Russian military drills were “purely defensive.” Putin referred to the invasion of Ukraine as a “special military operation,” with references to the invasion as a ”war” criminalized in Russia.
However, in 2022 it was highly unlikely that those watching in horror as Russian tanks rolled across the border would have happily used an AI personal assistant whose sole reference points were Russia Today or Pravda and the framings of the Kremlin. Should DeepSeek establish market dominance as the AI tool of choice, it is likely that some may unwittingly trust a model that sees consistent Chinese sorties that risk escalation in the Taiwan Strait as merely “necessary measures to safeguard national sovereignty and territorial integrity, as well as to maintain peace and stability,” as argued by DeepSeek.
Taiwan’s precarious plight in the international system has long been in essence a semantic battlefield, where any physical conflict will be contingent on the shifting meanings attributed to Taiwan and its people. Should a generation of Americans emerge, schooled and socialized by DeepSeek, that see Taiwan as China’s “internal affair,” who see Beijing’s aggression as a “necessary measure to safeguard national sovereignty and territorial integrity,” and who see elected Taiwanese politicians as “separatists,” as DeepSeek argues, the future for Taiwan and the millions of people on Taiwan whose distinct Taiwanese identity puts them at odds with China appears incredibly bleak. Beyond tumbling share prices, the emergence of DeepSeek should raise serious alarm bells in Washington and around the world.
All DeepSeek and ChatGPT responses were elicited on January 28, 2025.