International Trade and Artificial Intelligence: Is trade policy ready for Chat GPT?

As artificial intelligence continues to reshape industries and transform the global economy, trade policy must evolve to keep pace and ensure equitable growth for all nations.” This is how Chat GPT, the viral artificial intelligence (AI) chatbot sensation that has dominated the media since late 2022, suggests starting an article about AI and trade.

Already back in 2018, McKinsey predicted AI would contribute USD 13 trillion to the global economy by 2030. Today, with news outlets increasingly reporting on AI and major tech giants such as Google, Microsoft, and Baidu all rushing to release their own chatbots, the policy relevance of AI for international trade is a contemporary reality.

The policy relevance of AI for international trade is a contemporary reality.

At the intersection of trade policy and AI, the traditional view is that policy must get out of technology’s way. AI in this conception brings unbridled productivity if only trade barriers can be stemmed as the technology spreads from developed economies to the Global South. Under this view, also shared by the Organisation for Economic Co-operation and Development, policy recommendations include further liberalization of information and communication technology goods trade; lowering barriers to digital services trade; facilitating mode 4 delivery of services (presence of natural persons); and harmonizing data flow regulation. These policy recommendations would doubtlessly help the proliferation of AI technology in international trade, but they do not tell the whole story.

Policy Gaps in International Trade Regulation

Beyond merely providing an additional impetus to liberalize international trade, AI presents several policy gaps that current rules cannot address.

The rise of AI brings a familiar problem in the form of the goods–services distinction, which has proven difficult in the case of digital goods, for example. Although World Trade Organization (WTO) case law has shed some light on how to determine if a product is a good or a service, the WTO’s work program on electronic commerce—which was meant to settle the question more conclusively—is still being negotiated after 25 years. As AI is incorporated into more goods (think self-driving cars and AI robotics), it will become increasingly important to establish universal rules to determine whether General Agreement on Tariffs and Trade or General Agreement on Trade in Services (GATS) commitments dictate. The current patchwork that has emerged in the form of free trade agreements creates a fragmented landscape that is likely to boost the cost of trade.

AI presents several policy gaps that current rules cannot address.

Where AI can clearly be considered a service, other issues emerge. GATS market access commitments for certain professions including accounting, legal services, or medical services are often tied to certification requirements or legal personhood. This poses problems for AI systems such as legal tool Harvey or even Chat GPT (which recently passed the bar exam). Can such systems be said to have received education and training such that they are covered by GATS commitments? Similarly, where GATS commitments are based on legal personhood, does this exclude AI systems? Would this also be the case in a situation where a form of electronic or digital personhood is adopted for AI systems, as has been suggested in the European context? 

Another GATS-related problem concerns the four modes of delivery: cross border, consumption abroad, commercial presence, and the presence of natural persons. These modes are ill-adapted to products that have AI embedded in them, like self-driving cars, smartphones, or medical devices. Such products create a market access issue for services trade, which, according to the GATS, is not supposed to be subject to tariffs. However, services that are embedded into goods are liable to tariffication because the value of the services is included in the cost of the final product for which a tariff is levied. Software bought online and delivered through mode 1 is not tariffed, for instance, yet the same software, when installed on an imported computer, will effectively be tariffed, given that the customs value of the computer includes the value of the software. This has led some to call for the addition of a 5th mode of services delivery, meant to capture the services content embodied in goods exports. In 2009, the European Union’s estimated mode 5 services exports amounted to EUR 300 billion. A 2017 study, meanwhile, suggested that multilateral liberalization of mode 5 trade could increase world trade by EUR 500 billion. Though some attempts have been made to include mode 5 facilitation in bilateral agreements, this has not been attempted at the WTO.

The TRIPS agreement does not define how to deal with AI-generated works, and individual members have taken different approaches in their domestic legislation.

Lastly comes the issue of intellectual property. AI is producing graphics, poetry, and even music—all of which are the legal purview of intellectual property rights (IPR). In the context of international trade, it is the WTO’s Agreement on Trade-Related Aspects of Intellectual Property Rights (TRIPS) that sets minimum standards for the protection and enforcement of IPR. Unfortunately, the TRIPS agreement does not define how to deal with AI-generated works, and individual members have taken different approaches in their domestic legislation, ranging from full protection of AI-generated works to a requirement of human creativity that effectively leaves such works unprotected. This patchwork is likely to become increasingly unsatisfactory as the share of intellectual property—both copyright and patents—generated by AI and traded across borders continues to rise. 

A Job for the WTO?

As the only organization with a near-global mandate to regulate trade, the WTO seems like the logical place to forge agreements to deal with these policy gaps. The goods–services distinction, GATS professional certification issues, mode 5 services delivery, and IPR issues all situate themselves comfortably within one or several WTO committees and working groups. So, too, do the more traditional issues of ICT tariffs, services liberalization, and data regulation. Legislating at the WTO prevents international fragmentation and provides a rules-based system that stands in opposition to the might-makes-right approach that preceded the organization’s creation.

The suitability of the WTO must be considered in light of certain limitations, however. Chief among them is the pacing problem, a well-known phenomenon wherein technological innovation progresses too quickly for regulation to keep up. This issue is not unique to the WTO; it also exists in domestic legislative contexts. Even so, the fact that the WTO’s Joint Statement Initiative on e-commerce is finalizing texts on electronic signatures and spam in a world where AI, cryptocurrencies, non-fungible tokens, and blockchain technologies remain virtually unregulated illustrates the risk of relying exclusively on the WTO to legislate.

The WTO seems like the logical place to forge agreements to deal with these policy gaps.

Secondly, the lack of universality regarding normative approaches to AI regulation also inhibits a multilateral solution. Indeed, AI regulation is intrinsically linked to questions of privacy, morality, and property that differ across cultures. Even markets with similar levels of development, such as the European Union and the United States, have taken radically different approaches to privacy and free speech (think the General Data Protection Regulation). This is all without considering the divergent national interest with regard to trade liberalization. Rich countries, which dominate the AI market, want to liberalize trade as much as possible. The calculus for poorer countries, where tariffs remain an important source of government budgets, looks starkly different. 

These limitations notwithstanding, it appears unadvisable to let AI regulation develop exclusively in national or regional contexts. The policy issues detailed above require solutions, and multilateral solutions are preferable from a trade standpoint. 

It appears unadvisable to let AI regulation develop exclusively in national or regional contexts.

In the absence of hard law provisions, which have largely remained elusive in the WTO, the organization can still play an active role in this area. Soft law, informal rules, and norms can still be introduced in the WTO—as they have been for years—without relying on hard law. This would allow states to experiment with AI regulation in line with their economic, social, and moral objectives. Soft law approaches have proven successful in other international policy-making areas, and in the context of the WTO, could act either as a sort of greasing mechanism for the growing number of AI-impacting free trade agreements or, more ambitiously, as a halfway house, building toward hard law commitments in the future.

Timing is of the essence. With policy already lagging behind technology and the use of AI set to increase drastically in the real economy, some sort of international coordination is necessary. At a time when WTO reform is front and centre in the minds of many trade diplomats, an innovative approach to tackle the policy issues raised by AI could be an illustrative case in point of the organization’s continued value.

Whether or not states come together to regulate AI, it is clear that this technology is set to transform industries and trade patterns. Much like trade generally, AI will generate great economic growth but also painful labour market disruptions. Faced with the risk of worsening the digital divide by remaining inactive, policy-makers should make sure whatever solutions they arrive at contribute to creating a more sustainable and inclusive global economy.

Pascal Krummenacher is a trade policy professional and a former project officer at the World Trade Organization.