The Diplomat author Mercy Kuo regularly engages subject-matter experts, policy practitioners, and strategic thinkers across the globe for their diverse insights into U.S. Asia policy. This conversation with Dr. William C. Hannas – professor at Georgetown University and lead analyst at the Center for Security and Emerging Technologies (CSET) – and Huey-Meei Chang, CSET’s senior China S&T specialist and co-editor with Dr. Hannas of “Chinese Power and Artificial Intelligence: Perspectives and Challenges” (Routledge 2023) – is the 488th in “The Trans-Pacific View Insight Series.”

Identify salient myths and misinformation about China’s artificial general intelligence. 

Myths and misinformation about China’s quest for artificial general intelligence are as common as the misinformation “generated” by today’s large generative models. We’ll address a few such misconceptions, the first being the Chinese term for “AGI” itself.

“Artificial general Intelligence” is understood loosely as “the ability of an intelligent agent to learn any intellectual task that a human can.” Setting aside complications such as differences in human cognitive abilities and the difficulty of defining “G” or general intelligence itself, the concept and its associated term are broadly understood.

What about the Chinese term? Ben Goertzel, an AGI proponent who spent years in China, is credited with popularizing the term “AGI” in the early 2000s. Meanwhile, his colleague Pei Wang proposed the term 通用人工智能 – literally “general artificial intelligence” – to reference the same concept. Goertzel acknowledged later that he preferred the word-order of Wang’s term but stuck with “AGI” because he felt the acronym “GAI” might be misconstrued.

Now a lot has been said about differences between Western and Chinese visions of AI. And there are major divergences in China’s approach to AI that we will get into. But we shouldn’t get hung up on psycholinguistic nuances. Our team has studied thousands of Chinese AI documents and finds that the two terms – Chinese and English – are used interchangeably. In fact, it’s common to see “AGI” in parentheses after the Chinese term. So, let’s put that issue to rest. Both sides are talking the same endgame.

Examine the myth that China’s AGI progress can (or should) be defined by Western frontier models. 

The next myth is more insidious, namely, the belief that we can gauge China’s progress toward AGI along a continuum defined by Western “frontier models.”

While many AI experts worldwide discount the likelihood of large language models reaching human-level intelligence, Western companies working toward AGI nonetheless regard them as the primary path and have sequestered enormous resources based on this belief that more of the same will lead to the Holy Grail.

China, for its part, does not see large models as the only way to the Buddha. “Brain-inspired” (类脑) AI is a recognized discipline in China with its own funding codes, government support, and a development scheme called “one body, two wings” (一体两翼) that uses commercial profits to fund exploration. We surveyed 850 Chinese AI scientists on attitudes toward brain-inspired AI and 84 percent of the responders stated BI-AI would have more impact than other competing approaches to AGI.

China also views whole-brain mapping as another path to AGI. Xi Jinping himself is on record endorsing “connectomics” (脑连接图谱) – he used the technical term – as foundational for building intelligent technology. In addition, China invests in brain-computer interfaces not only for therapeutics but as a path to cognitive enhancement, the goals being affect control and AGI.

Finally, there is China’s work in embodied (具身) or physical AI, regarded by top ranking Chinese scientists as a likely path to AGI. (This is the subject of a forthcoming CSET paper).

Given these developments, it’s fair to say that the DeepSeek surprise opened Western eyes but also closed them. While we now take Chinese AI seriously, we imagine that large statistical models – DeepSeek being one of them – are the only Chinese path to AGI, when there are several typologically different approaches seriously pursued there.

Analyze the myth about China’s AI applications development and ambitions of dominating the global AGI competition. 

This leads to a third myth making the rounds, namely, the notion that China is more concerned with developing AI applications than winning an AGI race, unlike the West, which is focused on the latter.

The putative lesson is we should focus more on apps, less on building a nationwide archipelago of data centers, to stay competitive – as if the two goals were mutually exclusive. China sees no such contradiction. Rather, it believes that investing in physical AI is another way to get to AGI. That is, China’s pursuit of applications-oriented AI is meant to advance the country economically while paving the way to an AI “first mover advantage“ (先发优势)。

And it doesn’t stop there. Credible Chinese scientists see AGI leading to ASI or “artificial superintelligence” – AI that surpasses human intelligence and can bootstrap itself to higher levels without human intervention. Alibaba’s CEO Eddie Wu Yongming, for example, went public recently evangelizing this hypothetical development, arguing that AGI is “not AI’s endpoint, but a new beginning.”  Other top Chinese figures have spoken similarly.

Elucidate the broader context of Chinese AGI in the China-U.S. artificial intelligence race. 

Recent op-eds comparing the United States’ and China’s AI programs fault the former for its narrow focus on AGI while praising China for its practical success in applying AI throughout the whole of society.

While accurate in itself – America’s focus on large models may not lead to AGI and China is outpacing the U.S. in AI infusion – the argument is a false dichotomy, as China has by no means de-emphasized its state-sponsored goal to achieve AGI. Instead, it is and has been pursuing both projects simultaneously.

While known for its practical uses of AI, China, as noted, invests in not one but multiple paths to AGI, including the “frontier” models favored in the West and other, wholly different approaches such as “embodiment” described above.

Chinese scientists are aware of the salutary effects of embodiment and, hence, see no contradiction between practical AI and AGI development. The one supports the other. So, if there is a lesson to learn from China, it’s not to take all eggs from one basket and put them into another but to adopt multiple complementary approaches. And then back them nationally.

Looking ahead, what should be the priority for the United States and its allies? 

Given the stakes, one might suppose that the U.S. and its allies are staffing a comprehensive “watchboard” to monitor what is happening in China AI and other advanced technologies, on the premise that good policy depends on good intelligence. Regrettably, no such mechanism exists, inside or outside the U.S. government, although it’s clear it should be a national priority.