Artificial intelligence and the US-China geo-technological competition

PART 1

Artificial intelligence (AI) is the ability of machines to perform tasks that typically require human intelligence—for example, recognising patterns, learning from experience, drawing conclusions, making predictions or taking action—whether digitally or as the smart software behind autonomous physical systems (Reding and Eaton 2020, 14).

Artificial Intelligence in its current wave is primarily centred on Machine Learning, including Deep Learning.

​

One way of understanding Machine Learning is that it consists of automated statistical learning algorithms that are trained on large datasets, such that they become very effective at correctly recognising patterns and making predictions when encountering new real-world data.

​

And the data in question can be of any type: numbers, text, audio, images, video – if it can be digitised, it can be used.

So you heard in the first presentation more about the importance of data, so I won’t go into further detail on that issue.

But a general comment to understand why Machine Learning matters so much today: Since a few years, Machine Learning algorithms perform better than humans on a quite broad range of pattern recognition and prediction tasks.

Peering into the near future, with appropriate eyes and ears on robotic systems – meaning sensors – we’re looking at a world of intelligent connected devices – autonomous devices – capable of determining their own courses of action to solve particular objectives. This may occur with the devices acting alone, or in collaboration with other robotic systems, or in human-machine teams. And that may be in the cyber domain, or in the physical domains of Air, Land, Sea, or Space.

And so we’re looking at a fantastic ability to sort, to classify, to recognise patterns, to make predictions, in both cyber-space and in physical spaces. Over time, we’re looking at intelligent devices on factory floors, as well as on military battlefields.

Overall, the range of potential military applications is at least as vast as the range of tasks that require human cognition today, for example analysing and classifying visual data, organising logistics, operating support vehicles, or tracking and engaging hostile targets.

In short, AI is already doing many things, and there is lots more to come.

How can we conceptualise that from a social science perspective?

AI is what’s called a General Purpose Technology.

As defined by Lipsey, Carlaw, and Bekar (2005), a GPT is “a single generic technology, recognizable as such over its whole lifetime, that initially has much scope for improvement and eventually comes to be widely used, to have many uses, and to have many spillover effects.” Modern-era examples of GPTs identified by Lipsey, Carlaw, and Bekar (2005) include the steam engine, railway transport, the internal combustion engine, electrification, the airplane, the computer, and the Internet.

So, and this is important for this discussion, AI absolutely is a dual-use technology, but it is much broader than that, as it is also a General Purpose Technology.

To clarify, some examples of dual-use technologies that are not General Purpose: the jet engine, or for example radar, or for example laser technologies. All very important technologies with a fair range of both military and civilian uses. But none of these are transformative for the entire economy and for a wide scope of military activities in the way that a General Purpose Technology can be.

When a General Purpose Technology emerges, several things happen.

1/ it is very transformational. It changes jobs. Some jobs disappear. New ones appear. It changes ways of working. And it does so in many different industries and areas of civilian life. It also transforms military operations. So there’s a lot of adaptation and change involved. And all of those changes generate long-lasting positive economic productivity and economic growth effects. But with pressures for people and industries and states to adapt.

2/ it is nonetheless a relatively long process. A General Purpose Technology typically takes decades to fully mature and to unroll its full effects across societies. But there are phases during which a lot of excitement is generated across countries and across borders. Think back to electrification or the steam engine. Those were revolutionary. And there are multiple effects on public policies and also on foreign and security policies and defence policies. Leaders understand something important is happening.

3/ additionally, there is also a transformation process in military affairs. Just think for one moment about the impact of the steam engine on military affairs. And then think of the internal combustion engine. Or think of electrification. Or the adoption of computing – here I mean the 1950s, 1960s.

Obviously, at that point, Great Powers are going to start racing. Speed of technology adoption into the military sector becomes very important. It’s what we’re seeing now.

PART 2

In this second part I want to do a kind of short detour through aspect of economic thought.

And my main argument here is that the economic thinking and economic policy thinking that prevailed in the West in the last four decades has generally failed to account for intuitively obvious problems relating to state power and to competition between states, while also not giving a good account of the phenomenon of General Purpose Technologies.

The notion that a transaction is a win-win, is mutually beneficial, is central to how economic models are built. Economists focus almost all their time and energy on what economic agents do of their own free will. In that type of universe, nations are effectively modelled as sets of economic agents that have transactions with each other. In a model with more than one country, the agents may have transactions with agents in another country. In the simplest models, one renders all the agents of one country with what is called a representative agent, a kind of statistical average.

Better models take account of effects by sector, typically including institutional sectors – so we’re talking about the state, the financial sector, the non-financial corporate sector, and households. In those models, the state in one country will generally be assumed to be looking for an optimal policy. At this stage, many interesting research questions could be posed. But the most common choice is to assume that the state seeks to maximise the welfare of the country under equilibrium conditions. And that typically leads to the conclusion that lower import barriers are better than higher ones, and far better than autarky. That result is strong, theoretically and empirically.

Models typically find that a small economy will find it best to have no trade barriers at all, whereas a large economy may find it preferable to have net positive barriers on average, though they should be relatively low. The next question concerns which sectors benefit from higher import competition. Here, it’s long been known by economists that less competitive industries, including their workers, are losers from trade liberalisation. On the other hand, liberalisation is beneficial in the aggregate because all consumers benefit from cheaper goods and services, and those gains outweigh losses from the uncompetitive industries. What economists would then recommend – this was 20 years ago – was that it is up to the state to compensate the losers of globalisation, using proceeds from taxation, and that would still be a superior equilibrium than having a protectionist trade policy.

Now these general findings, which I saw promoted by most economists throughout my early career, tend to skip over many issues that have now become obviously much more salient.

1/ There is no clear understanding regarding the sources of long-term economic growth. In many economic models, technological change is assumed to occur randomly. And there is a high degree of faith in the market being able to generate new technology. It is often assumed by economists that if the state tries to pick winners, it will do a poor job, whereas the disciplining force of the market would be better. But that is all short-term thinking and is applicable basically in a world of fixed technologies, or in a world in which technological change is exogenous, and effectively random.

The concept of General Purpose Technology, which is comparatively recent, led to a niche strand of literature which is rather multi-disciplinary. With the rise of Artificial Intelligence, everyone has heard about the concept of General Purpose Technology. This wasn’t always so. But now that we all realise there is such a thing as General Purpose Technologies, we all want to know how they come about, and how best to ride the wave when they are here, also accounting for different stages of maturity.

2/ There is no understanding of exploitative strategies, for example regarding the role of industrial espionage, or more simply theft of intellectual property. Economists often do not model theft or espionage. The traditional stance is to call for secure property rights. Once every country has sensible Western-style institutions, which they should all rationally want otherwise they’ll be stealing from each other and doing each other down, then we can all trade together, and again we’re back to a win-win, all nations benefit from trade.

What has typically been assumed away is the possibility that an entire state would be deliberately oriented towards economic espionage as part of a deliberate strategy, which is what the United States is accusing China of doing.

3/ There was, for a very long time, no explicit assumption about what might happen if China, at some point reaching a very large GDP, would decide to devote a large share of it to military spending, let alone to technologically advanced military equipment. If the starting point, again implicit, is the Peace Dividend of the 1990s and everyone is talking about win-win cooperation and market competition, then why worry about future military power. Of course, some people did, but in a typical Western European context at least, those voices were barely audible. In Washington DC in recent years, there is a completely different degree of concern at being overtaken.

4/ The economics literature has long had a bias against state intervention, including for example the debate about picking winners. If one goes back to economic literature from the 1980s, 1990s, and early 2000s, this is anecdotal on my part, I should stress, by and large empirical studies that find that state intervention is inefficient get a lot more citations than those that find cases of successful state intervention. There is evidence that state institutions can learn to identify promising small businesses and help them grow. I’m not suggesting the state should substitute the market in general – we all know central planning doesn’t work. But the other extreme, to assume that state funds to promote innovation based on rewarding certain companies over others are always a bad idea, that isn’t true either. First because competitive mechanisms can be simulated. For example, with innovation prizes and a jury of experts who don’t have conflicts of interest. Second because, within certain limits, it is possible for a state-sponsored programme to make relatively good choices if it leverages high quality expertise. Within limits of course.

5/ My fifth and final point concerns investment time horizons and risk taking. It is in fact an old result, well established in the 1950s already by Kenneth Arrow, one of the most prominent economists of the 20^th century, that there is a positive welfare case for the state to fund basic research, because basic research is too risky and too long-term an investment for private investors to undertake with sufficient volume. One could easily extend that reasoning to financing PhD scholars for example, rather than charging them tuition fees.

Relatedly, the 1990s saw a boom in venture capital during the dot com period. This had an odd effect on policy beliefs. It was tempting to forget the previous decades of slow development of the Internet, impossible without DARPA funding for example, and to just celebrate the market. On the one hand, there is no question that Europeans missed the boat, and still miss the boat, when it comes to venture capital. But venture capital addresses a relatively narrow window of opportunity. It is, when done well by skilful investors, a relatively low risk bet. Typically, the technology is mature, the business concept is sound, the market is there, the people have the right skills, they just need a boost to scale up and tap into market size effects.

If the technology is relatively immature, then a lot more state funding is necessary. If the technology is rather more mature but quite costly, then only large corporations can afford the corporate R&D programmes and facilities to move forward.

PART 3

So, taking these points about economic policy together, we can see now perhaps more clearly what policy components would make sense given that the reality is not simple win-win trade with nations that have no particular power ambition and that play by the rules. It is a competition for national performance, more than for corporate performance. It is partly about GDP and productivity, but it is also about military power in the future. And it is a race, with antagonistic policies, with deliberate violations of intellectual property rights, industrial espionage, and traditional espionage.

So that there is no doubt about how the United States sees this, let me read out a revealing quote from a US official that expresses the spirit of the American awakening to the challenge. This is from Christopher Wray, who is the Director of the FBI and who already held that post in the latter part of the Trump Administration.

“The Chinese government is fighting a generational fight to surpass our country in economic and technological leadership.” “They have shown that they’re willing to steal their way up the economic ladder at our expense”.

The policy question then becomes: how do states win technology races, or at least stay ahead or not fall back, within such races?

In short, the United States wants to:

1/ get better at what it is doing at home, and

2/ reduce the ability of foreign powers to have access to what it is doing.

This splits naturally between two dimensions: Domestic Innovation Systems, and the External Dimension. For each dimension, I will now list 9 essential areas of work.

For domestic innovation systems:

R&D subsidies; Tertiary STEM education; Research universities; Private venture capital; Government venture capital; Government procurement; Innovation networks; Innovation clusters; Domestic industrial capacity.

For the external dimension:

IP protection (as contested); Standardization (as contested); Trade openness with non-rivals;

Trade & investment restrictions on rivals; Export controls; Foreign investment screening; Counterintelligence; Espionage-related sanctions; Espionage vulnerability mitigation.

​Two examples, briefly, concerning the domestic dimension. Remember that earlier I talked about deviations from classical liberal economic policy.

The first example is In-Q-Tel. In-Q-Tel was originally set up as the state venture-capital arm of the Central Intelligence Agency. To illustrate the influence of the In-Q-Tel example, one may note that both its current Chief Executive Officer, Chris Darby, and one of its former Chief Executive Officers, Gilman Louie, served among the 15 commissioners of the National Security Commission on Artificial Intelligence. That Commission was the leading US government sector effort, in the 2019 to 2021 period, to generate policy recommendations to advance US interests in the field of AI.

With In-Q-Tel, the idea is to learn from private-sector practices in the area of venture capital investment and repurpose them for state needs and more patient time horizons. When a company is selected, the idea is that it should pursue product development strategies aimed at serving both civilian markets and government needs. In this way, rather than effectively taking over a commercial company and limiting its growth potential to future government contracts alone, the government body encourages an intermediate trajectory made up of mixed revenue streams, in the hope that this will generate greater returns to scale and higher efficiency thanks to the disciplining effect of private-sector competition. Conversely, the advantage of this approach as compared to not intervening at all is that the commercial company will integrate current and likely future government needs into its product and business-development strategy, rather than ignoring them and finding itself, at a later date, unable to supply the government sector according to the latter’s requirements.

A second example is from the US Department of Defense and it’s called the Trusted Capital Marketplace. The problem statement that the Trusted Capital Marketplace tries to address may be expressed as follows: Certain technology companies that are not part of the traditional defence industry may be developing dual-use products that are of potential interest to the defence sector while having limited awareness of national security concerns. This may make them more vulnerable targets for both licit and illicit attempts to acquire their technologies on the part of foreign state actors. At the same time, their business development needs may lead them to seek investment from any potential source, thus exposing them to potential risks.

With the Trusted Capital Marketplace, the idea is to filter both innovators and investors, in terms of some minimum degree of national security, and then bring them together within a restricted, but trusted, marketplace where demand and supply for capital investments can meet. Note that this is not the same as asking all actors to have a security clearance. But it can and should be an antechamber for those technology companies that might, later on, get supply contracts for the DOD, and then they would undergo traditional security vetting procedures.

A third example, which is of a more general nature, concerns the revival of discussions on industrial policy. This is a topic that was almost taboo in the company of distinguished liberal economists not so long ago. And there were good reasons for it. There were policy failures, which are well-known from the Development Economics literature. For example, import substitution policies to support nascent industries in developing countries, back in the 1960s, 1970s, were not especially successful according to many economists. But industrial policy is also about investment, about priorities, about a degree of planning and reliable state funding. There are stories of failures, but also success stories. And then there are cases where only state investment could have made a particular project happen. The Apollo Programme has led to a special expression, which is “Moonshot Technology” – the idea that you will invest in something very uncertain, very costly, not mature, but with exceptional returns if it works out. Within US discussions there is a lot of talk that is more oriented towards private investors, trying to promote high-risk, long-term investments. Moonshot is the riskiest category, although it is not a firm category. Another expression is Patient Capital – as opposed to the normal, “impatient” investment timelines of Venture Capital investors.

The United States tries to combine a great diversity and dynamism in its private markets and private investors, together with a large Federal Budget. One could say that one of America’s favourite activities is finding new ways of combining public sector and private sector efforts. But when comparing US efforts with European ones, besides the inventiveness of US forms of private-public partnerships, the other big difference is the ability of the US Federal Government, when it wants to, and when Congress is behind it, to create new structures fast, to scale them up, fast, to staff them up, fast. Americans have an understanding of the importance and benefits of scale. Expressions like “scaling up” or “ability to scale” are rather more common in US discussions than in European ones.

I will focus now on some examples regarding the External Dimension.

1/ trade and investment restrictions on rivals

An example here is Executive Order 14032 of 3 June 2021 by the President of the United States –banning US financial investments into Chinese companies involved in either the military-industrial complex of China, or that are developing surveillance technology to facilitate repression or serious human rights abuses.

You may recall that Huawei, among others, was listed. But this is much broader than just 5G, or surveillance. We’re talking about shipbuilding, aerospace, missile technologies – and surveillance too.

2/ foreign investment screening

Screening of Foreign Direct Investment, that is, of strategic investments in corporations in our countries, is an area of policy that has developed further in recent years.

In the United States, strengthened legislation came in 2018 – that’s the Foreign Investment Risk Review Modernization Act of 2018 (FIRRMA), which modernized and strengthened the key US structure in this case, which is called CFIUS, The Committee on Foreign Investment in the United States.

In the European Union, we have new EU law, namely Regulation 2019/452 of 19 March 2019 “establishing a framework for the screening of foreign direct investments into the Union”.

3/ Protection of Intellectual Property: Trade Secrets

Under both US and EU law, a trade secret is information that is not generally known or discoverable by others, is maintained in secrecy by its owner (meaning a company), and it gives its owner a competitive advantage because it is secret.

Trade Secrets is one of two main approaches for companies to protect innovative ideas – the other main approach is patents, which relies on the opposite approach, namely publishing the information, but giving a legal monopoly to the patent owner to allow the patent owner to gather income from royalty fees.

Trade Secrets is an area that underwent legislative changes, in both the US and the EU, in 2016, and partly for the same reasons. In both the US and EU context, the combination of the rise of digital technologies and of the rise of China were clearly referenced reasons for strengthening legislation.

CONCLUSIONS

What we are going through is a race focused on who gets ahead, between two very large states, with respect to a General Purpose Technology.

The US approaches this in a manner that is quite different from classical liberal economics – although the influence of the spirit of the 1990s can be felt.

Certainly, banning external investments, restricting inward investments from abroad, and tightening export controls, are all restrictive measures that go against the liberalism of the 1990s. If one takes a much longer term view, some of the new measures have similarities with measures that existed during the Cold War with respect to the Soviet Union in terms of blocking technology transfers with strict export controls.

One thing the United States is not doing, which is perhaps surprising, is that it isn’t yet switching to large new Federal budget authorisations for R&D, for basic research. During the Cold War, the US devoted a substantially larger share of GDP to Defence R&D specifically – and one of the greatest success stories of US federal spending during the Cold War is probably DARPA. DARPA remains a crown jewel in many respects, but it is not as central to current discussions as it could be – presumably because there is so much focus on Machine Learning and Deep Learning, which are quite mature. But more money into DARPA would make a lot of sense, in my view.

What we do see is a lot of dynamism in creating new public-private mechanisms to try to tap into the rapidly moving technology that is AI – to stimulate investments, and to accelerate technology adoption by the Federal Government, and by the Department of Defense.

Overall, what this means from a social science perspective, especially for the field of economics, is that most of the literature that was produced about innovation between 1990 and 2015, approximately, is not fully usable. Almost the entire body of literature assumed a benign international security environment. It also assumed that international trade and investment were forces for good, because they ensured additional competitive forces. Of course, it must be noted that most of that literature used data only from OECD countries.

This leaves us with a mixed policy equilibrium. Economic liberalism among Allied nations is good. Economic liberalism in non-strategic sectors is good. But economic liberalism in strategic sectors with rival powers, that’s on the way out. This sounds very intuitive, and it is not completely new. It was also the norm during the Cold War. We seem to be moving partly back to that.

Edward Hunter Christie, Senior Research Fellow, FIIA