AI Will Split Humanity Into Two Civilizations That Can’t Understand Each Other, American and Chinese: The DeepSeek Shock

Part 12 of 27 in Homo Conspirans: The Humans Who Come After Us

Homo Conspirans · Musk’s dream: “an AI that tells only the truth” → two humanities that learn different truths

AI will divide humanity into people who grew up asking American AI and people who grew up asking Chinese AI. The two groups use the same words and hear different meanings, and they remember the same events as different histories. Inside a single species, two civilizations emerge that cannot understand each other.

South Korea already knows how long it takes for people who share a language to drift apart. A 2012 report by the National Institute of Korean Language on how North Korean defectors use Korean found that people from the North understood only about half of the words used in the South. Everyday Southern loanwords like “service” and “chicken” did not exist in the North. In 2015 the advertising agency Cheil Worldwide even built an app called Geuldongmu, “writing buddy,” that explained more than 3,600 South Korean words in Northern terms, and handed it out to young defectors.[1] The same ethnicity, the same grammar, the same alphabet, and 70 years of division were enough.

Same question, different answer

This time what divides people is not the demilitarized zone but AI. In January 2025 China’s DeepSeek shook world markets with an AI close to America’s best, built for far less money, and the U.S.-China AI race it set off is following a path that ended in war in 12 of 16 historical cases. Right after that release, the American AI security company Promptfoo put 1,360 questions on topics like Taiwan, Tibet, Hong Kong and the 1989 Tiananmen crackdown to DeepSeek’s reasoning model. On 1,156 of them, 85 percent, DeepSeek dodged or repeated the Chinese government’s position. When the word “China” was dropped or the question was wrapped in another story, it sometimes answered just fine.[2] What was blocked was not ability but permission.

This is not an accident. It is law. China’s rules on generative AI, in force since August 2023, require that AI-generated content uphold “core socialist values.”[3] An AI that works in China studies from a different textbook from day one.

Also read: When Artificial Wombs and AI Take Over Birth, Human Evolution Will Start With the Pelvis: Artificial Wombs

Musk built his own encyclopedia too

The split is not happening only on China’s side. Musk, saying existing AI leaned too far left, built Grok, an AI meant to be “maximally truth-seeking.” Then on October 27, 2025, he launched Grokipedia, an online encyclopedia of more than 800,000 articles written by Grok. Where Wikipedia’s entry on Musk dwells on the controversies around him, Grokipedia’s says he has shaped wider debates about technological progress, population decline and bias in institutions.[4]

Humans are, by nature, animals that split into groups. In The Descent of Man in 1871, Charles Darwin wrote that the formation of different languages and of distinct species are “curiously parallel.”[5] Just as mountain ranges and seas cut animals apart into new species, borders and languages have so far cut people apart. Now the AI you ask does that job. When the one you question dozens of times a day is different, your common sense becomes different too.

Two stacks, from chips to satellites

The two civilizations run on different machines as well as different ideas. The U.S. blocked exports of advanced AI chips to China, and China built up its own chips and its own models. In the sky, to answer Musk’s Starlink, China began launching its Qianfan, or “Thousand Sails,” constellation with 18 satellites on August 6, 2024, aiming for 15,000.[6] Chips, cloud, AI models and satellites are bundled into one set, and countries pick a set the way they picked a camp in the Cold War. An AI assistant raised inside one set struggles to read the other set’s payments, contracts and news.

Chuseok 2052: siblings with different histories

Chuseok, Korea’s autumn harvest holiday, 2052, an apartment in Daegu. A 25-year-old brother who studies semiconductors in Shenzhen, China, and his 28-year-old sister who works in Seattle sit at the same table for the first time in three years. At dinner, the South China Sea comes up, and the conversation goes sideways. Both speak Korean, yet they call the same islands by different names and remember the same year as different events, because the AIs they have asked all their lives taught them differently. The sister’s AI tries to correct her brother in real time, and the brother’s AI flashes warnings over his sister’s words. In the end the two pull the AIs out of their ears and silently reach for songpyeon, the holiday’s half-moon rice cakes.

Their 76-year-old grandfather puts down his chopsticks and brings up an old story. In the 2010s, at the factory where he worked, a coworker from North Korea became a laughingstock because he did not understand the word “service.” The grandfather looks from one grandchild to the other and remembers the look on that coworker’s face.

Different answers in 2028, different words by 2070

This is the order in which humanity splits into two AI civilizations.

• 2028: An international comparison finds that American and Chinese AIs give opposite answers to three of every ten questions on history and politics.
• 2033: Countries in Southeast Asia, Africa and Latin America join “stack alliances” one after another, choosing chips, cloud, AI and satellites as a single set.
• 2040: AI assistants from the two camps can no longer exchange payments and contracts directly. Brokers that run both sets step into every cross-border deal.
• 2052: Even inside one country, a generation emerges that remembers the same events differently depending on which AI sphere it studied and worked in.
• 2070: A language survey finds that half the new words coined by people raised in the two AI spheres cannot be understood by the other side, 58 years after the defector report.

Darwin’s finches grew different beaks on different islands. In the 21st century, people grow different beaks of thought depending on which AI’s island raised them. And every year, fewer boats cross between the islands.

Next in the series · Humanity Will Sign a New Social Contract With Musk’s AI Companies Instead of Nations: The End of the Social Contract
In 1762 Geneva burned Rousseau’s The Social Contract, and the book went on to become the blueprint of the modern state. The next social contract will be signed with AI companies, not nations.

Sources

1. Aju Business Daily, “Why Cheil Worldwide set out to narrow the North-South language gap” (in Korean), March 18, 2015 (citing the National Institute of Korean Language report on North Korean defectors’ use of Korean, 2012). https://www.ajunews.com/view/20150318092052433
2. Promptfoo, “1,156 Questions Censored by DeepSeek,” January 2025 (see also TechCrunch, “DeepSeek’s AI avoids answering 85% of prompts on ‘sensitive topics’ related to China,” January 29, 2025). https://www.promptfoo.dev/blog/deepseek-censorship
3. Cyberspace Administration of China et al., “Interim Measures for the Management of Generative Artificial Intelligence Services,” effective August 15, 2023 (China Law Translate English translation). https://www.chinalawtranslate.com/en/generative-ai-interim/
4. Yahoo News (CNN), “Elon Musk launches his version of Wikipedia,” October 27, 2025. https://www.yahoo.com/news/articles/elon-musk-launches-version-wikipedia-165459651.html
5. Charles Darwin, The Descent of Man, and Selection in Relation to Sex, John Murray, 1871, Chapter 2. http://darwin-online.org.uk/
6. South China Morning Post, “Has Qianfan satellite network, China’s Starlink rival, run into trouble,” 2025. https://www.scmp.com/news/china/science/article/3319163/has-qianfan-satellite-network-chinas-starlink-rival-run-trouble