Comment on “The Impact of Foundational AI on International Trade, Services and Supply Chains in Asia”

Meltzer (2024) does an admirable job of mapping recent technological developments in AI to the current institutions of the international trading system. This is a useful taxonomy.

I raise two concerns. First, AI's net effects on employment and output are inextricably linked to the pace of AI's adoption by industry. Meltzer notes that recent leaps in the capabilities of AI models have been accompanied by exponential increases for the “AI compute” needed to train these models. However, it is still very uncertain whether the cost of AI compute in the twenty-first century is going to decline over time at a rate resembling that of general-purpose computing in the twentieth century, and, therefore, whether twenty-first century AI comes into widespread industrial use quickly.

In the twentieth century, “Moore's Law” was code for a cadence of technological innovation in semiconductors that produced 20% to 30% annual declines in (quality-adjusted) computer hardware cost, and impressive declines in the energy needed for computing, lasting decades. Continuing price declines for computing created economic incentives to use computing in all kinds of new applications, across all sectors of the economy, substituting for labor, other forms of capital, and raw materials, and increasing productivity and living standards. (Jorgenson, 2001).

By the second decade of the twenty-first century, however, while some Moore's Law-style miniaturization of semiconductors continued at a slower pace, quality-adjusted semiconductor prices were no longer falling at earlier double-digit rates. (Flamm, 2017, 2021; Sawyer & So, 2018) Slackening technological improvement in chip manufacturing means that some new engine is needed to drive AI compute costs lower. Absent steady declines in AI compute costs, a rapid uptake of AI across broad sectors of the economy is unlikely to materialize.

Second, national security is tightly linked to AI, and, therefore, likely to drive national policies and investments relevant to cutting edge AI technology. This was the case when military investments first kickstarted the computer and semiconductor industries seventy-five years ago (Flamm, 1987). Militaries the world over are currently investigating incorporation of AI into autonomous military weapons systems platforms, and into security and disinformation systems that conduct information warfare operations in cyberspace. Use of the most advanced available semiconductor manufacturing technology is required to produce specialized AI compute capability in its fastest, densest (lowest weight and smallest size), and least energy intensive form factor, which translates into greater potential capability for a “smart” weapons system or an edge-connected network device, for any given weight, energy, and size budget.

As a result, responding to rising global political tensions and following the US lead, EU members, Japan, Korea, and Taiwan have significantly tightened national security export restrictions on leading edge semiconductor manufacturing technology used to fabricate leading edge AI compute chips, and the chips themselves. In a similar vein, 5G network gear produced by Chinese producers (Huawei, ZTE) is banned or being removed from telecom networks in the US, Britain, Australia, and several EU countries.

Global rules on trade and investment in tech sectors are currently being rewritten. Political, regulatory, and economic policies intended to encourage “onshoring” of high-tech supply chains are being put into place. Balkanization of the global economy into regional trading blocs, resulting in costly trade diversions, and truncation of cost-reducing scale economies, is the likely economic outcome of this ongoing “deglobalization”.

Meltzer's vision of increasing cooperation on AI through existing international framework agreements ignores these current realities. We have entered a historical period in which international economic policy is increasingly used as an instrument of national power (Hirschman, 1945). One can only hope that we can find a strategy to reduce these political tensions, reap the full potential economic returns from new technologies in the largest possible global market, and cooperate with others in applying those innovative technologies to solve pressing global problems—like climate change.