Whether we admit it or not, the United States and China are leading a race to develop AI. This article looks into the claim that the United States’ democratic culture slows down AI development in comparison to China. How valid is this disadvantage thesis? The winner of the RAILS Student Paper Award 2020 presents his key findings.

An article by Cecil Yongo Abungu

In my paper “Democratic Culture and the Development of Artificial Intelligence in the United States and China”, I confront the argument that considered next to China’s, the United States’ democratic culture leaves it disadvantaged in the development of artificial intelligence insofar as it stands in the way of decisive and coordinated action. This argument has been made by highly regarded AI analysts like Kai-Fu Cheng, whose 2018 book AI Superpowers made a huge splash among many influential people. More concerningly, my research shows that it is an argument increasingly shared by influential policymakers in the United States. The disadvantage thesis (as I christen it) is further advanced by big tech company executives, who continue to testify in congressional hearing after hearing that any serious regulation of their activities might give China the edge in AI development. Is the fear justified? Is there reason to be concerned that the United States’ democratic culture disadvantages the United States’ AI development?

My paper argues that the apparent disadvantage is at worst inexistent and at best negligible. I focus on the aspect of democratic culture that has to do with internal and external attitudes towards the determination of big-picture social demands, which I delve into it based on the claim that it has an unignorably significant influence on how the law is used around the technical factors (including computing power and collection of data) and ecosystem factors (for example the level of personnel training and skill present, scale of research and innovation and ease of tool adoption) which play a significant role in AI development in both the United States and China. Within the same framework, I consider how the understandings of individualism in both jurisdictions influence their democratic culture because of the impact that has in discourses surrounding privacy, data collection and the design of law relating to them—which directly impact the technical factors at play—as well as state action and inaction with regard to ecosystem factors.

My argument is that when examined critically, the democratic culture in the United States does not in fact cause it any concerning disadvantage to China in AI development, and is instead responsible for spurring innovation and motivating the more pronounced focus on safety and ethics in the United States vis-à-vis China. On first observation, it may seem that both slow the pace of AI development in the United States in comparison to China. Indeed, stripped to its core, the disadvantage thesis is that in comparison to China’s, democratic culture in the US encourages the engagement of the people and interest groups in a manner that spurs robust consideration of the blindsides of AI but which on the flipside leaves them struggling to match China’s pace of implementing AI policy decisions. I argue that this unexplained focus on pace betrays an incredibly myopic understanding of what ought to be encompassed in AI development.   

It may also seem that the American understanding of individualism not only influences its democratic culture itself but has in the past few years controlled the discourse over limiting personal data access and protection of privacy rights in a way that Americans would prefer—but once again seemingly gives China an edge in data as a technical factor. My research shows that the evidence in fact controverts the claim that an American culture that promotes more democratic ideals in designing a legal framework to govern AI slows down action in AI development.

I further claim that while it is true that American understanding of individualism and its pervasive influence on the democratic culture prevents its privacy framework from allowing the collection of AI training datasets at the scale which China enjoys, there is a great deal of nuance that renders this only a qualified advantage for China, and possibly a blessing in disguise for the United States. First, I argue that amount of training datasets required depends on the function that a particular AI tool is designed to perform, and some AI tools require a reduced scale of data. Second, I argue that the advantage which China enjoys with regard to data is not universal because, as Alexander Loukissas has argued in his book, locality of data is just as important. Third, I argue that the inability to access the same vast amounts of data as China has in fact led to more innovation in the United States—and we already have some evidence of this in for example the promise offered by research that aims to create more synthetic datasets.

In conclusion, I claim the United States has no reason to sacrifice its democratic culture at the altar of AI development.

Published under licence CC BY-NC-ND. 

Cecil Yongo Abungu Written by:

Cecil Yongo Abungu is a Research Affiliate with the Legal Priorities Project, working on projects around longtermism and AI risk. Cecil has an LL.B. degree from Strathmore Law School in Nairobi and an LL.M. degree from Harvard Law School. He has published work in among others the African Journal of International and Comparative Law and the Vienna Journal on International Constitutional Law. He also has upcoming work in the Harvard National Security Journal. Apart from his research experience, he has taught a variety of classes to undergraduate students at Strathmore Law School in Nairobi.