The Economics of Artificial Intelligence



Ajay Agrawal, Joshua Gans, and Avi Goldfarb | National Bureau of Economic Research

The last 200 years have produced a remarkable list of major innovations, not the least of which is artificial intelligence (AI). Like other major innovations, AI will likely raise average incomes and improve well- being, but it may also disrupt labor markets, raise inequality, and drive noninclusive growth. Yet, even to the extent that progress has been made in understanding the impact of AI, we remain largely uninformed about its international dimensions.

This is to our great loss. A number of countries are currently negotiating international agreements that will constrain the ability of sovereign governments to regulate AI, such as the North American Trade Agreement (NAFTA) and the Trans- Pacific Partnership (TPP)- 11. Likewise, governments around the world are freely spending public funds on new AI clusters designed to shift international comparative advantage toward their favored regions, including the Vector Institute in Toronto and the Tsinghua- Baidu deep- learning lab around Beijing. The international dimensions of AI innovations and policies have not always been well thought out.

This work begins the conversation. China has become the focal point for much of the international discussion. The US narrative has it that Chinese protection has reduced the ability of dynamic US firms such as Google and Amazon to penetrate Chinese markets. This protection has allowed China to develop signifi cant commercial AI capabilities, as evidenced by companies such as Baidu (a search engine like Google), Alibaba (an e-commerce web portal like Amazon), and Ten-cent (the developer of WeChat, which can be seen as combining the functions of Skype, Facebook, and Apple Pay).

While no Chinese AI- intensive company has household recognition outside of China, everyone agrees that this will not last. Further, a host of behind- the- border regulatory asymmetries will help Chinese firms to penetrate Canadian and US markets.Even the Pentagon is worried. Chinese guided- missile systems are sufficiently sophisticated that they may disrupt how we think of modern warfare; large and expensive military assets such as aircraft carriers are becoming overly vulnerable to smart weapons.

This may do more than transform the massive defense industry; these AI developments may radically shift the global balance of power.As international economists, we are used to hype and are typically dis-missive of it. Despite AI’s short life —AI’s rapid insinuation into our daily economic and social activities forces us to evaluate the international implications of AI and propose best- policy responses. Current policy responses often rest on a US narrative of a zero- sum game in which either the United States or China will win.

Is this the right premise for examining AI impacts and for developing AI policies? Further, calls for immediate action by prominent experts such as Bill Gates, Stephen Hawking, and Elon Musk will likely encourage governments to loosen their pocketbooks, but will government subsidies be effective in promoting broad- based prosperity or will subsidies become yet another form of ineffective corporate welfare? What specific policies are likely to tip the balance away from ineffective corporate handouts?Using comparative advantage theory, trade economists have thought long and hard about the right mix of policies for successfully promoting industry.

Many of our theories imply a laissez-faire free- trade approach. However, since the early 1980s our theories have shown that certain types of government interventions may be successful,. These theories emphasize the role of scale and the role of knowledge creation and diffusion. Unfortunately, the precise policy prescriptions produced by these theories are very sensitive to the form of scale and the form of knowledge creation/ diffusion. And competition can play an important role too.

We therefore start in section 19.2 by identifying the key features of AI technology in regard to scale and knowledge.

To date there are no models that feature the particular scale and knowledge characteristics that are empirically relevant for AI. In section 19.3 we use these features (a) to off er some suggestions for what an appropriate model might look like, and (b) to draw implications for policy.

This leads to high- level thinking about policy. For example, it provides a foundation for assessing recent proposals put forward by AI researcher Geoff Hinton and others on the potential benefit of public investments in AI.

However, these models are not sufficiently fine-grained to directly capture existing regulatory issues that “go behind the border” such as privacy policy, data localization, technology standards, and industrial regulation. In section 19.4 we therefore review the many behind- the- border policies that already impact AI and discuss their implications for comparative advantage and the design of trade agreements. We begin with a factual overview of the international dimensions of AI.


Read the full report here