This post originally appeared in The Cipher Brief.
When we think of arms races, we are naturally drawn to the physical: missiles, satellites, fighter jets, submarines and aircraft carriers. While those assets are important for conventional warfare, there is an emerging technological arms race in artificial intelligence (AI) that will have profound implications for the balance of power in the 21st Century.
When it comes to AI, the arms race is a two-country competition, between the U.S. and China. Given the economic implications as well as the tactical consequences, the stakes are incredibly high for both countries. Each player has relative advantages, and the future remains to be written. But make no mistake: this is a two-horse race from here on in.
China has held AI aspirations for the better part of a decade, as evidenced by Tianhe-IA, which topped the list of the world’s supercomputers in 2010 – the first time the honor went to someone other than the U.S. or Japan. Still, there was a “Sputnik moment” for China in 2016, when Google’s DeepMind team used AlphaGo to soundly defeat both the South Korean and Chinese champions in the ancient board game of Go.
The response, in typically rapid fashion, was China’s manifesto in the form of a State Council report on July 8, 2017, entitled “A New Generation of Artificial Intelligence Development Plan.”
The plan’s stated goal: “to seize major strategic opportunities for the development of artificial intelligence, construct the first-mover advantage of artificial intelligence development in our country,” and lead the world in artificial intelligence in 13-and-a-half years.
Most informed observers think that China can achieve its goal. Let’s consider the forces at play:
Government investment vs. government intervention
For the governments of the Western world, AI is an abstract concept that is feared as much as embraced. The U.S. is typical in this regard, with significant attention paid to the downsides and risks, and little attention to the importance of this, the largest technology wave ever. The current administration’s understanding of AI, its component parts and its levers is poor. Congress is no better.
The defense and intelligence sectors have the thought leadership in the government and the greater sense of urgency, but it’s not clear if they have the unified support required to move the needle.
The opposite is true in China. AI is a critical component and has received the commensurate attention, investment and regulatory landscape to succeed.
That last part, the regulatory element, will be important. If the U.S. insists on adopting a regulatory framework that slows the adoption of AI in favor of traditional jobs and industries, then the U.S. will be at a material disadvantage. This is because the Chinese government is actively investing in AI startups from chips to software, but it is also actively creating environments – mini-Silicon Valleys, capital, infrastructure – conducive to the development of AI.
Handicapping the component parts of AI
As we start 2018, we do so with the recognition that it has never been easier to be an AI company. Dozens of companies on both sides of the Pacific brand themselves as such – even if they are only doing some linear regression. There is a growing sophistication as to the component parts of AI and how they all need to be present. Again, our experience is that this is better understood in the defense and intelligence communities than the government at large.
But there are some key concepts that are worth noting.
The first is the ability to automatically discover patterns and relationships without having to ask explicit questions of the data. This is known as unsupervised or semi-supervised learning and is reserved for sophisticated players. These techniques reveal the unknown unknowns as well as the known unknowns and unknown knowns. This is an area where the U.S., driven by the “Tech Five” (Google, Facebook, IBM, Microsoft and Amazon) and a select group of startups, has a demonstrated advantage.
A second key concept is the area of prediction. This is the most visible dimension of AI and, given every AI company does it, would suggest that the U.S. does not likely have a demonstrable advantage here. Having said that, the combination of unsupervised approaches with supervised approaches generates better outcomes, so in the problem of that combination, the U.S. likely holds some advantage today.
An often overlooked but critical component of AI is that of justification. This goes beyond transparency and gets at what the algorithm is doing on an atomic level and in a way that is easily understand by the user community. This is far more important in a Western democracy – where companies have to transparently explain their actions as governed by legal norms, such as not discriminating against a particular class of people with their technology. Transparency in AI is really a regulatory and policy requirement in western democracies. Performance has to take a back seat to “explainability.” If your algorithm is taking race or sex into consideration, that is generally illegal here — also justification becomes even more essential if the technology is involved in any type of kinetic action.
China, for instance, doesn’t have the same protections and so they don’t have the same need for “explainability.” Over time we expect China to see the utility in transparency, but for now the expectation is that they value performance over the ability to understand a model at the atomic level.
The next important concept is the ability to take the output of an AI system and operationalize it. China’s scale, regulatory landscape and interest in the area have given it an advantage here. This is quite important, as it enables ongoing learning. Operationalization and ongoing learning will do more to close the gaps than anything else, and China has the superior position here.
Market economy vs. market forces
This is a subtle but important point. China’s particular brand of capitalism affords it significant control over all its economic levers. China has committed itself to adjusting those levers on behalf of AI. This commitment will endure economic cycles as well as hype cycles. In the U.S., AI investment comes from the Tech Five, the venture community and, to a far smaller extent, the government. There are some who are projecting that the overblown nature of AI in the U.S. is due for a correction. If the market for AI goes through a period of contraction, particularly in the venture community, don’t expect China to blink. They will invest right through the correction, picking up strategic assets as needed and to the extent the Committee on Foreign Investment in the United States allows.
Data as the fuel
When people said data was the new oil, they were understating its importance. AI has an insatiable appetite for data. There is a reason the Tech Five are dominating the AI landscape in the U.S.: they produce a stunning amount of data. Still, those massive datasets were rivaled, even surpassed, by the data gathered by their Chinese colleagues like Baidu and Tencent.
Furthermore, the general absence of privacy laws in China make that data more accessible to more parties. The Tech Five guard that data zealously and view the U.S. government with considerable suspicion. If the Chinese government wants training data, they get it. If the U.S. government wants training data, especially if it’s not open source, they likely need either private-sector cooperation or a court order.
This is going to represent a major advantage for China.
The prevalence of theft
One thing we have learned in the past half-decade is that the concept of data security is just that – a concept. In reality, nothing is safe. While AI is an exceptionally open field with almost every major advancement getting published, as the stakes get higher, that dynamic may change. If it does, expect governments/companies to resort to stealing what they consider to be of value and/or competitive importance. Stealing expertise is hard, but stealing algorithms or code is easy. It is safe to say that China has a considerable advantage in this regard, developed over a decade of taking what they couldn’t access in the market. It will be interesting to see how the U.S. plays this game considering it is not usually the one with the technology deficit.
While the future of AI is highly uncertain, these component parts will play a major role in determining who becomes the leader in the space. As a long-time technologist and partner to the defense and intelligence community, I share the sense of urgency and concern held by my counterparts in government. The AI arms race is real, and to depend solely on the Tech Five or the venture community to underwrite the country’s commitment is to take a large risk, especially given the primary competition is committed to carefully managing the process.
This is not to suggest that the government play a greater role in the investment game (greater than what the nonprofit venture capital firm In-Q-Tel provides currently). Rather it is to suggest that AI, in all its component parts, from robotics and drones to cognitive systems, gets the level of attention that the challenge demands.
That means more leadership defining more experiments that generate more data. This is not an arms race that we can afford to lose.