A Coverage of Technical Barriers to Trade and Industry 4.0: Examples of Industry 4.0-Type Technical Barriers to Trade Notifications

Recent notifications of draft regulations to the TBT Committee show that Industry 4.0-related regulations are increasing in number and variety, and that members, by submitting such notifications, consider them as falling within the scope of the TBT Agreement. Novel technologies are being regulated for a variety of policy reasons and in different ways. Here are a few examples.

B The Regulatory Challenge

Faced with the mindboggling rapidity and complexity of these transformations, governments struggle to react to them in a timely and coordinated fashion. Globally, this may generate a plethora of different regulations and technical standards across countries. This is a problem – in particular, when differences are unnecessary and ill founded – from the perspective not only of its trade impacts but also of consumers, industry and society in general.

TBT disciplines are unique in the way they put into practice the overarching goal of balancing the right to regulate and the avoidance of unnecessary technical barriers.Footnote ³⁸ They are intended to impact members’ entire regulatory lifecycle preventively, reactively and self-correctively. Industry requires regulations that address market failures but do not stifle innovation and competitiveness.Footnote ³⁹ The public requires regulation that fosters trust and confidence in products’ quality, performance and safety; and trading partners require regulations that are nondiscriminatory, not protectionist and not more trade restrictive than is necessary.

This characteristic of balancing between needs makes TBT particularly well suited to addressing regulatory challenges in these times of rapid and radical technological and societal changes.

C Transparency on Nascent Regulation

TBT obligations start early in the regulatory process. The TBT Agreement becomes relevant once the decision to regulate is taken by a member. When this happens, the Agreement subjects members to a series of disciplines aimed at preserving members’ policy space while ensuring that the regulation is the least trade-restrictive possible.

D Data and Dataflows: Is There a Technical Barriers to Trade Angle?

Despite competitive tensions, international cooperation is needed if Industry 4.0 is to succeed. AI needs to be able to access vast amounts of data. Volume matters because machine learning needs to be able to incorporate into future predictions as many past outcomes as possible.Footnote ⁴⁵ Much of this data is obtained from both national and cross-border Internet activity and digital platforms. Businesses and governments are also important sources of data. The McKinsey Global Institute estimates that in 2014 global data flows were more valuable than trade in goods,Footnote ⁴⁶ and PricewaterhouseCoopers predicts that by 2030, AI could raise global GDP by over $15 trillion.Footnote ⁴⁷

Discussions on e-commerce at the WTO are illustrative of the challenges faced on the global governance of data and the implications for Industry 4.0. At the risk of missing out important nuances, views on the question of data appear to fall into three camps.

There are those who are generally opposed to data localization requirements and restrictions on cross-border data flows, unless such measures would fall under limited exceptions. There are others who, while in general opposing these restrictions, wish to reserve the right to adopt appropriate safeguards to protect personal data and privacy, including rules on the cross-border transfer of personal data. Finally, some appear to favor wide latitude to exercise cyberspace sovereignty in pursuit of public policy objectives.Footnote ⁴⁸

While the TBT Agreement’s scope is on trade in goods and not on cross-border data flows per se, there are some important data-related considerations. With the IoT, products have embedded sensors that collect, transmit and exchange information in real time over a network regardless of their location. Since such sensors and the accompanying source code could be said to be a characteristic of the product, any technical regulation on source codes would probably be within the scope of the TBT Agreement.

Members in the TBT Committee have discussed around sixteen “specific trade concerns” related to a range of “cybersecurity measures”, covering ITC products and network equipment, vehicles, civil aviation, banking and insurance, amongst other sectors. Concerned members voiced issues about requirements that could discriminate against foreign technology and equipment, lack of clarity of the measures, inconsistency with international standards and best practices and the need for duplicative in-country testing of imported products.Footnote ⁴⁹

The challenge is: how do we square the right to take measures necessary for the protection of essential security interests and not create unnecessary obstacles to international trade? Here, there are important principles and obligations in the TBT Agreement that are worth recalling, especially in terms of promoting good regulatory practices and regulatory cooperation, and the use of international standards.

E Role of International Standards in Support of Regulatory Alignment

International standards are critical for achieving the full potential of Industry 4.0, while avoiding unnecessary barriers to trade in these technologies.Footnote ⁵⁰ Alignment of regulations for connected devices to international standards will facilitate trade by providing a common benchmark, enhancing competition and lowering prices for consumers.

The TBT Agreement is a driving force for harmonization and coordination at a global scale, through its provisions that strongly promote alignment of national regulations to international standards. The pivotal role of international standards to the attainment of the TBT Agreement’s principles and objectives is already reflected in various parts of its preamble. There, the Agreement “encourage[s] the development of international standards”, listing “important contributions” these documents can make to “improving efficiency of production and facilitating the conduct of international trade” and promoting “the transfer of technology from developed to developing countries”.Footnote ⁵¹

F Improving Coherence

Coherence in standards development is a major challenge for Industry 4.0. Entire batches of standards are being developed to underpin AI, the IoT, blockchain and autonomous systems (e.g. cars, trucks, trains, drones), to name just a few.Footnote ⁶³ These standards will all need to “talk” to each other, and interoperability will be critical to ensure performance, privacy, safety, etc. Imagine one autonomous vehicle trying to avoid an accident with another autonomous vehicle. If the other vehicle is following a different standard, the vehicles will not be able to communicate, and this could inadvertently provoke a crash.Footnote ⁶⁴ So, as this process unfolds, standards will be essential to keep the “parts” interoperable and contributing to the “whole”.

If, for example, two international bodies decide, independently and without talking to each other, to develop differing standards for addressing issues related to the safety of autonomous vehicles, they may well end up adopting two significantly different – or worse, conflicting – international standards addressing the same issue.  Trade will be very difficult, if not impossible, between countries that have not used the same international standard as a basis for their regulations.

The TBT Committee “six principles” highlight the importance of coherence,Footnote ⁶⁵ in order to avoid duplication or overlap between the work of international standardizing bodies. Cooperation and coordination are essential. A lack of coherence is also a barrier to participation by developing countries in international standardization as their scarce resources cannot cover participation in duplicative processes.

Lock-in and path dependencies in standards for one technology can also quickly lead to fragmentation and duplication in standards for other technologies. This translates to higher trade costs and impediments to innovation. Forward-looking cooperation between standards development organizations and between regulators can help chart a path toward convergence.

G Good Regulatory Practice

Good regulatory practice (GRP) describes best practices and procedures developed by governments and organizations to improve the quality of regulation.Footnote ⁶⁶ It must therefore also be an indispensable component of the Industry 4.0 regulatory process. A key feature of the fourth industrial revolution are technologies that straddle multiple sectors, jurisdictions and institutions. Regulation, on the other hand, tends to be organized along traditional sectoral lines. As new regulations are formulated, or existing ones redesigned, the impact on trade could be considerable. GRP provides governments with a toolkit of approaches and processes that can help them identify and address the trade impacts of their regulatory action.Footnote ⁶⁷

Examples of GRP include internal coordination (whole-of-government approach), transparency and public consultations, and regulatory impact assessment. Much work has been done in this area, both at the WTO and elsewhere, including in the context of the Asia-Pacific Economic Cooperation (APEC), the Organisation for Economic Co-operation and Development (OECD) and the World Bank.Footnote ⁶⁸ Application of GRP can help ensure the design of high-quality, cost-effective regulations that are consistent with the goal of open trade. Moreover, the wider dissemination of GRP can contribute to the establishment of a common, predictable framework for regulatory intervention, thereby facilitating international regulatory cooperation and harmonization.Footnote ⁶⁹

The TBT Committee has recognized that “GRP can contribute to the improved and effective implementation of the substantive obligations under the TBT Agreement”.Footnote ⁷⁰ GRP discussions in the TBT Committee have emphasized the transparency and accountability of regulatory processes.Footnote ⁷¹ Strengthening transparency and accountability can help avoid unnecessarily trade-restrictive regulatory outcomes. Other areas of GRP considered by the TBT Committee include analysis and review of regulatory alternatives (including the option not to regulate) and the design of regulations, including the advantages of simple, responsive and flexible regulations. Members have also stressed that GRP is an important element of capacity-building initiativesFootnote ⁷² and that preparing GRP guidelines could be particularly helpful for developing countries.Footnote ⁷³

The TBT Agreement’s provisions on transparency and discussions on GRP are closely linked. For instance, “early notice”, notification, comments, publication and entry into force are all processes that should lead to better regulation. Incorporating the transparency processes of the TBT Agreement into the regulatory lifecycle of a specific measure is a powerful means of fostering GRP, which has transparency and consultation as one of its fundamental components.Footnote ⁷⁴ At the same time, greater utilization of GRP could help contribute to avoiding unnecessary and unintentional regulatory barriers to trade.

H Regulatory Dialogue and Cooperation

Besides harmonization via international standard-setting processes, the TBT Agreement provides members with “sign-posts” to engage in other forms of cooperation through various mechanisms, such as “equivalence” or “arrangements for conformity assessment”.Footnote ⁷⁵ These are additional mechanisms for encouraging the reduction of regulatory diversity and associated trade costsFootnote ⁷⁶ – key challenges of Industry 4.0.

Equivalence refers to an arrangement in which members recognize that, although each other’s product specifications and rules are different, they achieve the same result. When this happens, they can decide to accept the rules of the trading partner as “equivalent”.Footnote ⁷⁷

This facilitates trade by allowing firms to produce according to domestic requirements and still directly access foreign markets without having to meet another set of requirements. It does not require regulations to be changed from the preferred domestic policy. The way that members choose to cooperate through equivalence agreements may vary depending on trade flows, their respective levels of protection, the costs of demonstrating achievement of the appropriate level of protection in light of risk and the possibility of unilateral or mutual recognition of equivalence.Footnote ⁷⁸

The TBT Agreement sets out a framework for members to reach “equivalence agreements” on their TBT-related technical regulations.Footnote ⁷⁹ The Agreement does this not by requiring but by encouraging members to “give positive consideration to accepting as equivalent technical regulations of other Members, even if these regulations differ from their own, provided they are satisfied that these regulations adequately fulfil the objectives of their own regulations”.Footnote ⁸⁰

I Facilitating Acceptance of Conformity Assessment Results

Divergent systems that trading partners use to verify conformity to applicable standards and regulations can create impediments to trade.Footnote ⁸¹ Around half of specific trade concerns raised at the TBT Committee over the last ten years have been on CAP. Duplication of testing and certification (due to nonrecognition of results) is a frequent source of trade friction. It is likely that conformity assessment will become even more complex with Industry 4.0.Footnote ⁸² Developing countries face particular challenges in the digital age, as systems and laboratories are not always available or as effective as they could be.Footnote ⁸³

The TBT Agreement, of course, requires members to ensure that their CAPs, among other obligations, do not create unnecessary obstacles to international trade.Footnote ⁸⁴ But the way to achieve this result is far from obvious.

Governments greatly prize their freedom to regulate as a sovereign right and they have a responsibility to do so. No government wants to be told by another government (let alone an international organization) how to protect their security interests, consumers or the environment, or what tests or certificates to accept and from whom, or what results to follow. They may be unwilling to accept a test report or certificate from a foreign body that they do not know or may not trust.

Moreover, limited development of National Quality Infrastructure (e.g. standardization, metrology, accreditation, conformity assessment procedures), supporting laws and policy frameworks in some economies may limit options available to regulators when choosing their conformity assessment procedures.Footnote ⁸⁵

This can make it difficult to strike the balance between, on the one hand, the strictness of the procedures put in place to assess conformity to a given regulation’s objectives, and, on the other, the risks that “nonconformity” with those objectives would create.Footnote ⁸⁶ This challenge is not unique to Industry 4.0-related products, but achieving the appropriate balance may be particularly difficult for new technologies for which the risks are still not fully understood. Here, the TBT Agreement encourages the use of several tools to build trust and confidence.

These include recognition of conformity assessment results and the use of international and regional systems for conformity assessment. The TBT Agreement encourages members to recognize the results of CAP of other members, recognizing that this needs to be built upon cooperation between them, including in respect of the adequate and enduring technical competence of the relevant conformity assessment bodies in the exporting member.Footnote ⁸⁷ The Agreement provides a basis for cooperation, stating that prior consultations might be needed to arrive at a mutually satisfactory understanding on recognition.Footnote ⁸⁸

The TBT Agreement mentions accreditation as one means to build trust and confidence in the technical competence of foreign bodies providing CAP results.Footnote ⁸⁹ It also encourages members to allow foreign conformity assessment bodies to participate in their national conformity assessment procedures on a national treatment and most-favored-nation basis.Footnote ⁹⁰ The potential of accreditation to lower the costs of conformity assessment by eliminating the need for duplicative tests and certifications is not reached automatically: regulators have to be willing to rely on accreditation within their conformity assessment schemes.

Further, the TBT Agreement encourages members to enter into negotiations to conclude agreements on mutual recognition (MRAs) of the results of each other’s conformity assessment procedures.Footnote ⁹¹ MRAs are one regulatory cooperation mechanism that allows parties to recognize specific results (e.g. test reports or certificates) in specific sectors.Footnote ⁹² While little is known about the actual implementation and functioning of MRAs, they can require significant time and costs to negotiate and maintain.Footnote ⁹³ MRAs have been most successful in specific sectors like electrical and electronic products (e.g. the Association of Southeast Asian Nations MRA).

The TBT Agreement requires members, wherever practicable, to formulate and adopt, as well as participate as members of, international systems for conformity assessment.Footnote ⁹⁴ This can help to strengthen regional and international regulatory cooperation between members in the area of CAP.Footnote ⁹⁵

These types of international and regional systems have grown in importance since the entry into force of the TBT Agreement in 1995. In the TBT Committee, members have discussed a range of systems based on arrangements between accreditation and conformity assessment bodies, including those operated by the International Laboratory Accreditation Cooperation, the International Accreditation Forum and the International Electrotechnical Commission (e.g. the IECEE CB schemeFootnote ⁹⁶), as well as other organizations. In the TBT Committee, members have discussed how to increase regulators’ reliance on these types of systems to facilitate trade.Footnote ⁹⁷

A Data-Driven Business Models

The automotive industry is in transition where business model changes are concerned.Footnote ²⁵ “Traditional” automotive manufacturers are now transforming into a new mobility ecosystem – from a mass-produced goods-sale business into highly customized data-based service suppliers.Footnote ²⁶ In order to address changing consumer demand, the automotive industry is becoming less and less industrial and, simultaneously, increasingly intent on services, especially the operation and maintenance of vehicles.Footnote ²⁷ Moving toward data-driven business models, AI further shatters the boundaries between the physical and the remaining components of the CAV.Footnote ²⁸

At the same time, another significant transformation in the automotive industry is the use of CAVs to supply Mobility as a Service (MaaS) solutions. Vehicle manufacturers will play an important role in MaaS services.Footnote ²⁹ On the one hand, the transportation environment will be increasingly dominated by car-sharing, ride-hailing, and related services. In this context, being driven by CAVs will represent the essential nature of future transportation, and automotive firms will supply car-sharing and ad hoc use applications to drive up utilization rates. On the other hand, revenues generated by MaaS may become the core source of shareholder value creation when traditional car sales decline.Footnote ³⁰ To summarize, the data-driven economy enabled by CAVs will displace vehicle ownership with MaaS, leading to a new transport system landscape.Footnote ³¹

In light of these changes, business models for CAVs will become increasingly complex. Considering the trends toward MaaS, the typical business model will be based on a function that combines tangible vehicles and intangible services. Automotive firms will offer services to adjust existing functionalities and update software to increase the automated capabilities of a vehicle. Because the economic value of CAVs relies on their use value, throughout their operational life, tangible vehicles and intangible services must be combined to jointly fulfill customers’ needs.Footnote ³² In this respect, it is fair to say that the higher the level of automated driving systems, the more service-oriented the automotive industry. Despite this reality, most CAVs, especially levels 1–4 of driving automation, fall in between traditional “goods” and “services.”

B Integrated Technical Features

CAV systems are highly integrated, both internally and externally. In terms of internal integration, a CAV may contain several driving automation features that have individual, narrow-use specifications, but which together may provide advanced automated driving functions. Technically speaking, a combination of automation features means that an aggregate level of safety is necessary for CAVs. In other words, part-based standards, for example steering wheels, no longer make any sense.Footnote ³³ CAVs should be regulated at the system level to ensure the overall safety of the entire driving system. Only an aggregate “measure of safety” can cope with the challenges of “cumulative” automation features and therefore adequately protect CAV safety.Footnote ³⁴ The “target” of safety regulation, therefore, should be gradually shifted from “auto parts” (goods) to integrated CAVs (overlapping boundaries of goods and services).

In terms of external integration, vehicles are rapidly transforming into “connected devices.” Under the technological trends of V2X, a CAV is merely one component of the entire transport ecosystem. Depending on the level of automation, a CAV may be able to communicate with its occupants, other vehicles, road users, the surrounding transportation physical infrastructure, and all other Internet-based devices and applications. Indeed, in the extreme, in the age of AI, the world we are living in can be described as a convergence of all “IoT devices.” CAVs will always be data-driven – by digitally connecting to one another and their surroundings. By interacting with the external physical and digital environment through V2X communications, CAVs are literally “components” of a holistic transport landscape.Footnote ³⁵ This complex transport ecosystem requires a regulatory framework that considers security convergence, namely the combination of physical security and cybersecurity.

Despite this reality, security risks are particularly complex for CAVs, because they operate across both the physical and the digital world.Footnote ³⁶ Compared to conventional vehicles, risks to a CAV involve threats related to the integrated environment. When communicating with other vehicles and infrastructure, CAVs become a channel for attack and an opportunity for hackers. Hackers can target the CAV itself, the servers supporting it, or the external systems that communicate with the CAV. It is also technologically possible for attackers to seize control of an entire fleet of CAVs by breaching the infrastructure.Footnote ³⁷ In this context, the internal and external integration of CAVs may lead to complex security concerns. There is a need not only for “device” security but also for entire ecosystem security, with a strategic approach to threats. Regulators must ensure that CAVs are safe, both mechanically and in terms of protection from cyber attacks.Footnote ³⁸

C Technical Barriers to Trade Agreement or General Agreement on Trade in Services?

Considering their business models and technical features, CAV-related safety standards may disrupt the scope of coverage of the TBT Agreement. Are CAVs goods or services?Footnote ³⁹ Classification determines whether and to what extent the TBT Agreement rules are applicable. When “goods” and “services” converge as a package in the CAV market, the same is true for relevant safety standards.Footnote ⁴⁰ The integrated CAV system, under which services are embedded with the physical body of the CAVs, is disrupting the traditional boundaries of trade regimes in terms of standards.

It should be noted that WTO case law generally supports the existence of a “boundary” between trade in goods and trade in services.Footnote ⁴¹ The Appellate Body has repeatedly stressed that whether a specific measure is scrutinized under the General Agreement on Tariffs and Trade (GATT), the General Agreement on Trade in Services (GATS), or both is a matter that can only be determined on a case-by-case basis.Footnote ⁴² Here, the Singapore Standard Council, for example, has issued a Technical Reference related to an enhanced cybersecurity framework for CAVs.Footnote ⁴³ The Technical Reference, among other things, requires CAV developers to provide comprehensive documentation for a security-by-design review, to conduct cybersecurity assessment, and to comply with “cybersecurity principles” throughout the full lifecycle of CAVs, including design, development, operations, maintenance, and decommissioning.Footnote ⁴⁴ The purpose of this Technical Reference is to provide rules to govern the cybersecurity assessment framework of CAVs on public roads. Toward that end, drawing from best practices in the industry, the Technical Reference provides standards. These standards apply to “a cyber-physical vehicle system,” which includes embedded control systems and “a coupling between the computational elements and physical elements.”Footnote ⁴⁵ In this particular instance, is the standard under Singapore’s Technical Reference for Autonomous Vehicles a measure of goods or services?Footnote ⁴⁶

Figure 6.1 demonstrates the evolution of CAV standards.Footnote ⁴⁷ At the core of the issue is whether the regulations under the emergence of a new ecosystem of mobility should be subject to the TBT Agreement. Based on the factors delineated earlier regarding the transformation of the auto industry, the MaaS, the cumulative automation features of integrated CAVs, and the complex security concerns involved, the classification or reclassification of CAVs – in the context of the SAE’s six levels (levels 0–5) – will prove an interesting case. At one extreme, level 0 (no automation, human driver), representing conventional, “personally owned” vehicles,Footnote ⁴⁸ should be classified as goods and thus fall within the lower-left quadrant. At the other extreme, level 5 (full automation, “passengers only”),Footnote ⁴⁹ representing shared CAVs, should be classified as services and thus fall within the upper-right quadrant. The remaining levels of CAVs (levels 1–4) comprise both goods and services under this four-quadrant analytical framework. During this time of transitional human–vehicle cooperation, CAV measures govern “trade in goods” but also affect “trade in services.”

Figure 6.1 (Re)classification of connected and autonomous vehicle standards

As shown in Figure 6.1, as CAVs evolve from level 0 to level 5, more and more domestic regulations will be subject to GATS. That said, CAV standards are arguably a “technical standard” within the meaning of Article VI:4.Footnote ⁵⁰ The requirements in GATS Article VI:5, including “reasonable” and “not more burdensome than necessary,” should be further explored in relation to high-level CAV standards.Footnote ⁵¹ However, the GATS contains too few trade rules to handle level 5 CAVs.Footnote ⁵² More importantly, in the interim, regarding levels 1–4, how should CAV standards be reclassified? The obligations under the TBT are far more substantial than those under the GATS. The danger of legal unpredictability may be imminent. To conclude, automation systems are “disruptive” in the way that they challenge existing governance frameworks and disrupt the boundaries of the TBT Agreement.

A Industry-Driven Standardization Process

One noteworthy angle in the ongoing process of CAV standardization is the industry-led approach. Such a regulatory scheme emphasizes market incentives rather than top-down regulation.Footnote ⁵³ Government agencies consider themselves to be in partnership with developers in pursuit of the safe and rapid deployment of CAVs. One outstanding example is the subtitle of the US National Highway Traffic Safety Administration (NHTSA) guidelines: “Accelerating the Next Revolution in Roadway Safety.”Footnote ⁵⁴ By stressing its role as a facilitator in the exchange of information among CAV stakeholders, the US government does not play an exclusively dominant role in shaping the standards of automation systems.

This privatization of governance is attributable, in part, to governments’ lack of requisite technical expertise, as well as the flexibility necessary to address ever more complex regulatory tasks.Footnote ⁵⁵ One of the potential legal hurdles that might slow the deployment of CAVs is bureaucracy, which often moves much more slowly than technological changes. Relevant agencies are aware that CAV development is taking place in a remarkably complex and dynamic technological environment, and that governments are currently in no position to provide mandatory performance standards for these emerging technologies.Footnote ⁵⁶ Indeed, the development of CAV standards requires timely action. The scheme under which the private sector leads the standardization process has proven to be a more effective approach.

Procedurally and practically, the concept of “co-governance” has increasingly been advocated.Footnote ⁵⁷ The CAV ecosystem comprises a variety of interconnected stakeholders, including the automotive industry and software businesses.Footnote ⁵⁸ Governments must govern alongside private and civic sectors in a more inclusive, collaborative, and dynamic manner to drive cross-industry discussion. Through a relatively inclusive and transparent process,Footnote ⁵⁹ governmental agencies participate on an equal footing with stakeholders.Footnote ⁶⁰ In a world in which technological development is firmly in the control of industry, “regulated” parties should be treated as committed partners.Footnote ⁶¹ The collaborative approach may prove perfectly sensible.

B Voluntary Standards

An empirical survey reveals that in most jurisdictions, CAV standards go by many names, including “guidance,” “guidelines,” “recommendations,” “informal standards,” “best practices,” and “codes of conduct.”Footnote ⁶² These informal regulatory mechanisms, which in general are created to be “morally” or “politically” binding, can be considered to fall under the broad umbrella of “soft law.”Footnote ⁶³ This demonstrates how CAVs have been “regulated” in the loosest sense of the term.

Empirical studies also demonstrate that more and more jurisdictions aim to minimize mandatory governmental regulation and favor voluntary, industry-led, nonbinding standards to enhance CAV safety.Footnote ⁶⁴ In the UK, for example, the Automated and Electric Vehicles Act 2018 received Royal Assent on 19 July 2018.Footnote ⁶⁵ The UK’s Law Commission completed its first round of consultation in February 2019, which focused on regulatory frameworks, namely how to ensure safety when using CAVs, and how current road rules should be adapted for AI.Footnote ⁶⁶ The primary considerations identified in the consultation paper identify areas in which there may be ambiguity in the law, as well as potentially necessary reforms. Stakeholders have actively responded to the key questions raised in the consultation paper, including, among other items, how to allocate civil and criminal responsibility when control is shared between the automated driving system and a human user, impacts on other road users, and protection from risks.Footnote ⁶⁷ More importantly, the government has already published a “Code of Practice” to provide “guidance” for CAV testing. Although, by the very nature of the code, compliance is voluntary, it nevertheless sets out “principles” and details “recommendations” that the government believes should be followed to minimize potential risks and maintain safety.Footnote ⁶⁸

On the other side of the Atlantic, the US NHTSA and the US Department of Transportation (DOT) jointly issued the Federal Automated Vehicles Policy in September 2016.Footnote ⁶⁹ The Policy was designed to “set forth a proactive approach to providing safety assurance and facilitating innovation.”Footnote ⁷⁰ The NHTSA issued “Automated Driving Systems: A Vision for Safety 2.0” (Guidance 2.0) in September 2017,Footnote ⁷¹ which, based on the comments of key stakeholders (e.g., the automotive industry) as they considered “best practices,” offers a “flexible, nonregulatory approach” to CAV safety.Footnote ⁷² In October 2018, the DOT released “Preparing for the Future of Transportation: Automated Vehicles 3.0” (Guidance 3.0), which again serves as “voluntary guidance” and “is intended to be flexible.”Footnote ⁷³ As advanced by some commentators, the NHTSA’s CAV guidelines are indeed “an exuberant celebration of volunteerism.”Footnote ⁷⁴

In Asia, one striking case is China’s CAV roadmap. Following the “New Generation Artificial Intelligence Development Plan,” issued by the State Council in 2019, the Ministry of Industry and Information Technology announced a Three-Year Plan for Promoting Development of a New Generation of the Artificial Intelligence Industry (MIIT), implemented to detail action plans for the development of driver assistance systems, vehicle intelligence algorithms, and automotive smart chips.Footnote ⁷⁵ Further, the MIIT has published “communication guidelines,” which cover ninety-two standards, with a focus on CAV technology.Footnote ⁷⁶ The government also announced that more than thirty key standards that are critical to autonomous driving systems will be introduced in the coming years.Footnote ⁷⁷ China’s approach to CAV development is systematic. Following China’s top-down standardization process, the government continues to take primary responsibility in terms of standardization development. However, in the case of CAVs, the government has been working closely with the CAV industry, including the China Industry Innovation Alliance for the Intelligent and Connected Vehicles, to ensure the “relevance and flexibility” of standards.Footnote ⁷⁸ The alliance, together with other industry associations, has been commissioned by the MIIT to develop a common set of “protocols” for CAVs.Footnote ⁷⁹

To summarize, most countries maintain CAV safety policies while emphasizing the voluntary nature of standards and safety assessments.Footnote ⁸⁰ Governments tend to refrain from mandating CAV-specific design features and performance standards. In addition, relevant authorities are inclined to offer nonregulatory approaches to CAV safety. CAV developers may “consider” the guidance as they develop, test, and deploy CAVs on public roadways. They are also “encouraged” to submit a “safety self-assessment,” describing their treatment of each guideline.Footnote ⁸¹ This chapter is not the place to provide a detailed analysis of whether existing approaches can meet CAV safety needs. However, a few general comments bear emphasis. To some extent, the voluntary approach seems realistic. Standards of a soft law nature offer advantages over traditional command-and-control regulation because they provide greater flexibility and adaptability and lower compliance and administrative costs, directly address industry-specific and consumer issues, and adapt to the rapidly changing political landscape.Footnote ⁸²

C Technical Barriers to Trade Agreement: In or Out?

The proliferation of “soft” safety standards in the CAV industry reveals a pattern of self-regulation. In 2018, the SAE formed a committee of stakeholders, took the lead, and published the CAV safety “principles.”Footnote ⁸³ In 2019, industry leaders across the CAV technologies also published “Safety First for Automated Driving,” a nonbinding framework for the development, testing, and validation of safe CAVs.Footnote ⁸⁴ The distinguishing feature of these standards is their legally nonbinding nature, which strongly correlates to the self-regulation of nonstate actors. Indeed, they were literally drawn up by the private entity that is to be regulated.Footnote ⁸⁵

This ongoing shift to voluntary co-governance raises an important question: Under what conditions can the complaining party invoke the dispute settlement system against nonbinding CAV standards that have developed by the private entities?Footnote ⁸⁶ In Figure 6.2, the vertical axis represents the relative level of governmental involvement in the standardization process, while the horizontal axis represents the relative degree of the binding effects of the standards. At the crux of the matter is this: the emerging CAV standards have implications for the boundaries of the TBT Agreement. First, the fact that the CAV standardization process may lack sufficient governmental involvement raises the question of whether the TBT Agreement will apply. Second, these CAV standards may fall outside the definition of “technical regulation” in the TBT Agreement. As clearly argued by Lim in Chapter 5 of this book, the first issue is more important than the second, because although the TBT disciplines on technical regulations are relatively deeper, voluntary standards are still subject to Article 4 and Annex 3 of the Agreement.Footnote ⁸⁷ Nevertheless, as shown in Figure 6.2, the transformative nature of disruptive innovation renders the interpretation and application of TBT rules challenging.

Figure 6.2 Co-governance of connected and autonomous vehicle standards

To illustrate, the WTO provides a dispute settlement mechanism whereby a member considers that its benefits under the covered agreements are being impaired by “measures” taken by another member.Footnote ⁸⁸ In this context, any act or omission “attributable” to a WTO member can serve as a “measure” of that member for purposes of dispute settlement proceedings.Footnote ⁸⁹ “A complainant must clearly establish that the alleged measure is attributable to the responding Member.”Footnote ⁹⁰ The decisive factor here is “the degree of governmental involvement.”Footnote ⁹¹ WTO case law indicates that “private actions” may fall within the scope of the TBT Agreement if the support provided by a government is sufficient to become a “governmental act.”Footnote ⁹² Considering the approaches taken by WTO panels and the Appellate Body in past disputes,Footnote ⁹³ the conduct of the private body may come under WTO disciplines if there is a demonstrable link between the government and such “conduct.”Footnote ⁹⁴ To summarize, a nexus must exist between the responding member and the “measure” – regardless of whether it is an act or an omission.Footnote ⁹⁵

At the core of the question is whether the standards published by a nongovernmental body at the request of the government, or with some degree of government support, can be viewed as a measure of a member government.Footnote ⁹⁶ Is there an adequate connection between a private entity’s self-regulation and government action? To what extent should a tribunal impose the responsibility of WTO members with regard to industry standards? The emergence of the self-regulating, collaborative approach for CAV standard-setting therefore leads to the question of how the TBT Agreement should respond to the trend of public–private convergence in the standardization process.Footnote ⁹⁷ As shown in Figure 6.2, the CAV case study demonstrates that the co-governance model, in which governments “more or less” “work together” with stakeholders of the CAV ecosystem, arguably falls into the middle-lower portion of the “Governmental Involvement” axis.

Furthermore, even if sufficient governmental involvement can be found in the CAV standard-setting process, the next question is whether these “standards” can constitute “technical regulation” within the meaning of the TBT Agreement, therefore allowing a set of deeper trade disciplines to apply.Footnote ⁹⁸ In past WTO cases, the panels and the Appellate Body elaborated on the term “mandatory,” indicating that it includes “binding as well as obligatory, compulsory, not discretionary,” or “required by law or mandate; compulsory.”Footnote ⁹⁹ However, as clarified in several disputes, “the mandatory character of the measure may result from a combined effect of various measures or actions attributable to the Member in question.”Footnote ¹⁰⁰ In other words, a “voluntary and non-mandatory” measure may become “mandatory” as a result of “some other governmental action” or “some other action attributable to the Member concerned.”Footnote ¹⁰¹ As shown in Figure 6.2, the industry-driven, voluntary standards of CAVs may be determined to fall outside the definition of “technical standards” under the TBT Agreement because of its private and soft nature, unless compliance with the standards becomes de facto mandatory because of “some other action attributable to the Member concerned.”

That said, CAV standardization is indeed an interesting case study in the determination of “de facto mandatory.”Footnote ¹⁰² In terms of administrative actions, CAV safety guidance, although industry-led and nonbinding, may become a core requirement for “duty of safety” and may also have important evidentiary value in regulatory investigations. The UK’s CAV consultation paper, for example, indicates that “there is a rebuttable presumption that a product is safe if it conforms to voluntary standards published by the Commission.”Footnote ¹⁰³ In terms of judicial litigation, depending on the level of automation, courts need legal standards to determine liability when CAVs crash. More specifically, for levels 1–4, in which humans and CAVs are codrivers, determining “cause” during the transition to a driverless future will become increasingly complex in lawsuits related to negligence or product liability. Did the CAV properly alert the human driver?Footnote ¹⁰⁴ Should the CAV have been designed to automatically reduce vehicle speed on a snowy road? Or should the system prevent driving altogether?Footnote ¹⁰⁵ In a negligence or product liability lawsuit involving CAVs, the key issues will be whether the design of the CAV is defective in nature. This will inevitably lead to an inquiry into the “standard of care” that is imposed on CAVs, or the definition of a “design defect” for CAVs.Footnote ¹⁰⁶ How do manufacturers address “reasonable care” when designing the automated driving system? What safety standards should apply to CAVs?Footnote ¹⁰⁷

In practice, the industry-led voluntary standards provide a baseline for judges in the evaluation of appropriate levels and evidence of CAV safety prior to deployment,Footnote ¹⁰⁸ which may become a strong incentive for CAV manufacturers to comply with “guidance,” “best practice,” or “codes of conduct,” thereby demonstrating a commitment to meeting expert safety expectations.Footnote ¹⁰⁹ More specifically, compliance with industry standards can provide convincing evidence regarding whether there is a defect.Footnote ¹¹⁰ CAV manufacturers may be able to partially mitigate the legal risk by demonstrating conformity to industry safety standards. At the same time, judges may resort to industry standards when evaluating the duty of safety in tort cases.Footnote ¹¹¹ In brief, self-regulation is being afforded legal status through complementary evidence.Footnote ¹¹² As a result, private soft law is no longer merely a self-imposed corporate obligation. It can also constitute a source of law in court proceedings.

To conclude, compliance with CAV standards may become de facto mandatory, which somehow blurs the line between mandatary/voluntary standards. The key question, however, remains: Is the TBT Agreement capable of addressing de facto mandatory “self-regulation”? To what extent should a WTO tribunal assume the responsibility of members with regard to CAV safety standards that are prepared and published by a private entity? How should the TBT Agreement respond to the trend of private standardization processes in which the government involvement per se is often minimum, if not nominal? Evidently, the development of disruptive innovation inherently involves changes in governance frameworks and calls for new governance approaches that break the boundaries of existing trade disciplines.

A Patents and Trade Secrets

At first glance, obtaining patent protection for an AI-generated invention appears straightforward; such an invention would be patentable if it meets the definition as set out by Article 27.1 of the Trade-Related Aspects of Intellectual Property Rights (TRIPS) Agreement and domestic legislation:

Thus, in order to qualify for protection, an invention must meet the following conditions:

• novelty – meaning it is not part of the state of the art (as defined by the relevant patent law);

• inventiveness (or non-obviousness); and

• utility (or industrial applicability) – which means, among other things, that it is capable of actually working in the real world (e.g. an invisibility cloak, similar to what one might see in Star Trek or Harry Potter, would not meet this test).

These conditions distinguish an invention from a mere discovery of, say, a naturally occurring phenomenon or equation.Footnote ¹³ Thus, discovering that a wave coming from the earth’s core that interferes with satellite broadcasts is useful, but not patentable, whereas inventing a device which filters the interference would be a patentable invention.

Inventiveness (and novelty) are judged from the standard of a person skilled in the art (i.e. the “skilled person”) standard.Footnote ¹⁴ The skilled person is a legal fiction of a person who knows everything about a particular field yet is not creative or imaginative. While the standard of the skilled person differs across jurisdictions,Footnote ¹⁵ Laddie, J. discussed the nature of the skilled (but non-inventive) person in the art in the case of Pfizer Ltd.’s patent:

AI potentially challenges the standard of inventiveness in three ways. First, many AI systems work by sifting through vast amounts of data to find patterns, which some liken to the process of discovery, which is problematic as it is generally accepted that utilizing something that already exists in nature is a “discovery”, and therefore not patentable.Footnote ¹⁷ Second, as an AI has far greater capacity to analyse vast amounts of data trained with specific data from designated fields of art than any human, such an AI not only will more easily find inventions obvious but, if taken to its logical extreme, it could not only become a skilled person but might also find all inventions obvious.Footnote ¹⁸ Third, as AI is increasingly employed in research and development (R&D), there is the potential to raise the standard of inventiveness of developers and those skilled in the art. What these potential issues mean is that in order to accommodate AI, there may have to be a re-think of the inventiveness standard – this will be difficult and lead to many unintended consequences. For example, raising the bar for inventiveness would ensure information and discoveries are not locked away but may make it harder for ordinary human inventors to acquire a patent.Footnote ¹⁹

In addition to the three standards of patentability, a patent application must also provide for sufficient disclosure of the invention to enable a skilled person to reproduce the invention.Footnote ²⁰ This is in keeping with the patent system’s other goal to promote social and technical advancement and increase public welfare through the disclosure of inventions to the public for the benefit of the public at large. AI-related issues pose a significant challenge to the system in this regard – simply stated, what does one need to disclose when dealing with an AI-generated invention? The answer is not as simple as one might think given that the outcomes of an AI invention might change depending on the input data and/or the algorithms.Footnote ²¹ Additional questions involve unique complexities and will lead to divergences of opinion, such as: must algorithms used by the AI be disclosed as part of a patent application by an AI? Should data used by the AI also be disclosed as well, and if so, how much data? Does the human expertise used to select and curate the data and train the algorithm be disclosed, and if so, to what extent? Requiring strict standards which demand inventors to supply greater detail and precise information may mean inventors simply bypass the patent system in favour of maintaining the invention as a trade secret. This in turn may not be to society’s benefit. In short, the difficulties with enablement/disclosure should not be underestimated given the potential size of the data sets involved.Footnote ²²

Finally, the Trade-Related Aspects of Intellectual Property Rights (TRIPS) Agreement allows (but does not require) members to create exceptions to patentability, such as inventions “necessary to protect ordre public or morality, including to protect human, animal or plant life or health or to avoid serious prejudice to the environment”,Footnote ²³ and diagnostic, therapeutic and surgical methods for the treatment of humans or animals as well as plants and animals other than micro-organisms, and essentially biological processes for the production of plants or animals other than non-biological and microbiological processes.Footnote ²⁴ While these lines may appear straightforward, the reality is not quite so simple and distinctions can be blurred.Footnote ²⁵

Compounding the issue is the fact that exceptions from patentability are defined at the regional and national level. In regard to AI, this raises issues involving software and business methods. These forms of invention are excluded from patentability in some nations, meaning that software or computer program-related inventions may be patentable in one jurisdiction but not in another.Footnote ²⁶ The result of disunity could be more forum shopping whereby organizations, inventors or creators actively seek jurisdictions whose laws are more favourably disposed to protecting their inventions or creations, in this case created or generated by AI. Moreover, this connects to the issue of broader societal benefits resulting from inventions; if AI applications or algorithms are excluded from patentability, developers and organizations would essentially have no choice but to keep such AI as a trade secret, thereby undermining the goal of the patent system to disseminate technical knowledge.

To date, applications for patent protection when the inventor is named as an AI system have been rejected. For instance, the UK Intellectual Property Office (UKIPO), European Patent Office (EPO) and US Patent and Trademark Office (USPTO) all recently denied applications for patents that named an AI, called DABUS, as the inventor in December 2019, January 2020 and April 2020, respectively. These applications, for a warning light and food container, were made on behalf of Stephen Thaler, CEO of Imagination Engines, by the Artificial Inventor Project.Footnote ²⁷ In rejecting the application, the UKIPO hearing officer decided that the UK Patents Act 1977 requires an invention by a natural person.Footnote ²⁸ The EPO likewise rejected the applications on procedural grounds as an application for a European patent must designate an inventor and “state the family name, given names and full address of the inventor”. In so holding, the EPO found this requirement was consistent with a “clear legislative understanding that the inventor is a natural person” and consistent with EPO and national decisions.Footnote ²⁹ Similarly, the USPTO held that the US patent statutes preclude interpretation of “inventor” to cover machines because “the plain reading” of words such as “whoever”, “himself”, “herself” and “individual”, as well as the requirement that an inventor executes an oath, is as a “person”. The USPTO also cited US case law which holds that inventors cannot be states or corporations.Footnote ³⁰

This leaves trade secrets as the more likely avenue for protection of AI inventions. Trade secrets protect information that is secret, of commercial value, imparted in a situation of confidentiality and subject to reasonable efforts to protect its secrecy. In some ways, trade secrets better suit the companies which develop AI as such inventions do not require registration and can last indefinitely, provided they are kept secret.Footnote ³¹ Moreover, given the fast pace of development and difficulty in reverse engineering AI systems, companies are in fact increasingly relying on trade secrets to protect investments and developments in AI. AI companies also rely on trade secrets to protect their valuable algorithms given their inability to acquire patent protection for algorithms and reluctance to disclose the algorithm in a patent application,Footnote ³² and because copyright law protects expression and not the underlying idea behind an algorithm, effectively rendering copyright unsuitable for protecting the functional aspects of algorithms.Footnote ³³

While one cannot register a trade secret – doing so would itself alert others to its presence and provide for public disclosure – there are systems in place to prove the existence of a trade secret without disclosing the secret. An example of this is WIPO Proof, which provides tamper-proof evidence of the existence of a trade secret by providing a data- and time-stamped digital fingerprint of a digital file containing the trade secret and a repository of these fingerprints (which WIPO refers to as tokens).Footnote ³⁴

B Copyright

The TRIPS Agreement (Article 9.2) states that copyright attaches to original works which are “expressions and not to ideas, procedures, methods of operation or mathematical concepts as such.” In the context of software, this makes clear that expressions are protected, whereas the underlying ideas in the software or computer programs – that is, the AI algorithms and other processes – would not be eligible for protection.

Originality (which is different from the novelty standard in patent law which requires that the invention not be part of the prior art) – the aspect of a created or invented work that makes it new or novel, and thereby distinguishes it from reproductions, clones, forgeries or derivative worksFootnote ³⁵ – remains a basic prerequisite in copyright law.Footnote ³⁶ The question in regards to our inquiry is whether an AI-generated work can be regarded as original. The Berne Convention references “authors”,Footnote ³⁷ which may not be determinative but perhaps lends itself to the conclusion that there must be a human involved in the process. Indeed, courts in several countries have interpreted originality as requiring a fairly significant degree of human ingenuity. For instance, in the famous case of Naruto v. Slater, the 9th Circuit Court in the USA held that “[t]o qualify as a work of ‘authorship’ a work must be created by a human being”.Footnote ³⁸ In that case, it was not enough for a photographer to place cameras in strategic locations and tempt the animals to pick up the camera and take pictures. Likewise, in Acohs Pty Ltd v. Ucorp Pty Ltd the Full Federal Court of Australia found that data sheets created by a computer program (a simple data-collecting mechanism) were not subject to copyright because there was not a sufficiently involved human author.Footnote ³⁹

These judgments indicate that copyright over computer programs and software will generally vest with whoever created the source code of that software. Likewise, content generated by “AI-like” software which performs functions based on programmed rules but without exhibiting true intelligence or originality, for example a “smart-home” device that can dim lights or check the weather forecast on command, would likely remain the copyright of the author of the program’s code or the person making the input. The same would likely apply for programs used as part of an artistic or technical process but which are ultimately controlled by human choices.

That being the case, while the creator of the AI program would retain copyright over original source code, that individual may have no rights to original work created by the software that they did not envision or program. Thus, for instance, while the source code of an AI program designed to create original music or generate business recommendations would be subject to copyright, the decisions and work generated by that AI may not be copyrightable if there is not a sufficient level of human input. The more distant the human involvement from the ultimate original work (as the AI continues to evolve), the less likely it would be that copyright would attach to the individual.

Whereas the US Copyright Office and others apply a “human authorship policy” that prohibits copyright protection of works that are not generated by a human author,Footnote ⁴⁰ not all jurisdictions concur with this interpretation. For instance, UK law acknowledges the possibility that works could be “computer-generated”Footnote ⁴¹ and provides that the author of a computer-generated work is deemed to be the person “by whom the arrangements necessary for the creation of the work are undertaken”.Footnote ⁴² Interestingly, China may also be heading towards protection for AI as a court decision in Guangzhou in January 2020 awarded RMB1500 in damages for infringing a financial article written by Tencent’s robot Dreamwriter without authorization: “the article’s form of expression conforms to the requirements of written work and the content showed the selection, analysis and judgment of relevant stock market information and data … the article’s structure was reasonable, the logic was clear and it had a certain originality”.Footnote ⁴³ What remains unclear in jurisdictions which hold that AI-generated work can enjoy copyright protection is, among other things, whether such protection extends to other related copyright rights such as sound recordings, broadcasts, performances or adaptations. This is an important question, but as of yet undecided and untested.

Yet another interesting question is whether copyright law ought to be used to regulate deep fakes – the generation of simulated likenesses of persons and their attributes, such as their appearance or voice.Footnote ⁴⁴ Deep fakes raise complicated copyright questions such as whether deep fakes created by information that may be copyright protected should benefit from copyright, and if they should, to whom the copyright in the deep fake should belong; and whether the person whose likenesses and performances are used in the deep fake ought to receive compensation, and if so, how this could be done.

More fundamentally, other questions involving the term of protection and liability of the copyright owner will also need to be addressed. In terms of the former, many copyright laws provide specific periods of time during which the work and the rights arising thereof are legally protected that are usually determined in reference to the lifetime of the work’s author, and exceptionally the work’s first publication or transmission. The life of the author cannot be used when AI is the author, given the theoretically indefinite lifespan of the system,Footnote ⁴⁵ but consensus has not yet emerged on the appropriate length of protection. In regards to liability, unlike an original work written by a person, some AI systems store their information in a form that cannot easily be read by humans or reverse engineered. Given this, it may be impossible to discover why a system made a particular decision or produced a particular output. In such cases, liability will likely attach to the person or entity that controls or directs the actions of the AI. This is difficult, however, and may not always be apparent where one party has created the AI and another has decided what data to put into it or what questions to ask it. In the interim, the practical reality is that business entities will need to ensure that there are contractual indemnities in place for any actions of the AI that infringe copyright work.Footnote ⁴⁶

C Intellectual Property in the Data

A thorough discussion of AI and IP cannot ignore the important issue of data, as there may be IP in the data and there certainly is IP in the systems that manage and handle data. Developers rely on vast troves of data in the initial training of AI systems as well as for personalization, product improvement or localization (i.e. adapting AI systems to work in a variety of different local conditions). Considerable resources must be spent finding suitable training data, correcting training errors or ensuring the data has not been corrupted (for example, by a cyberattack).

Yet IP protections for data are limited save for some sui generis legislation and the limited protection offered by copyright law for databases as collections.Footnote ⁴⁷ In the USA, for example, databases may be protected by copyright law not as such but as compilations which are defined as a “collection and assembling of preexisting materials or of data that are selected in such a way that the resulting work as a whole constitutes an original work of authorship”.Footnote ⁴⁸ Such protection is of limited value, however, as the US Supreme Court held that a compilation of facts is copyrightable only if the selection or arrangement “possesses at least some minimal degree of creativity”.Footnote ⁴⁹ Pre-existing materials or data included in the database therefore may be protected by copyright, or may be unprotectable facts or ideas.Footnote ⁵⁰ In contrast, Europe grants copyright protection to databases which, as such, by reason of the selection or arrangement of their contents, constitute the “author’s own intellectual creation”. However, additional sui generis protection afforded under the Database DirectiveFootnote ⁵¹ is granted to reward the substantial investment of the database maker in creating the database and prevent free-riding on somebody else’s investment in creating the database, and exists in parallel to the copyright protection on the structure of the database.Footnote ⁵²

That there is weak IP protection for data and no system of property rights raises numerous questions regarding the equity of current setups among AI companies that take freely provided data from individuals, then use this data to create products that those same individuals are charged to use.Footnote ⁵³ This situation is analogous to the one lesser-developed countries experienced decades ago when they complained that developed countries had appropriated their traditional knowledge (TK)Footnote ⁵⁴ without adequate compensation, thereby exacerbating the wealth gap between developed and developing countries.Footnote ⁵⁵ TK does not enjoy IP protection, though sui generis legislation in some countries does grant protection. But unlike the international north–south divide that characterized the TK debates decades ago, the current debate on remuneration to data providers is both international and intra-national (though the challenge now could be characterized as a struggle between the tech giants who have both the data and the technological infrastructure and those companies who do not possess such assets).

Granting property rights in the data raises a host of challenging issues, which are raised but not analysed here, including adequate compensation for use, restriction on use (and whether prevention of data transfer without compensation falls afoul of obligations undertaken in free trade agreements) and whether and the extent to which property rights in data hinder innovation. Another IP-related aspect to data worth considering is that there is IP in the form of know-how (or other trade secrets) in the use of data for AI. Because of the growing liability concerns and other issues associated with faulty AI systems,Footnote ⁵⁶ organizations employing AI systems will need to become more careful about the implementations of such systems, and will need to ensure the quality of the data used to train and update such systems to ensure that the data is appropriate for the task at hand, that it was not tampered with and that it is accurate (the last point is a problem given that the sources of data companies rely upon for, say, digital marketing may be suspect).Footnote ⁵⁷ Thus there is IP in the curation of data – to ensure the data has been validated, is appropriate for the purpose at hand and has not been tampered with – as well as IP in the creation of AI applications and deployment of AI technologies (that may be protected by IP rights or block their use by others’ IP rights). The importance of these IP-related data issues will only grow in relevance, yet current levels of protection are limited and it is uncertain whether greater levels of protection would lead to clearer outcomes or enhanced levels of innovation.

D The Profound Question

IP refers to creations of the mindFootnote ⁵⁸ used in commerce, and the IP system is designed to encourage creation and invention and reward creators and inventors through IP rights. As per Stephen Thaler’s claim that an AI is capable of autonomously creating a patentable invention (DABUS) without any human intervention, the most fundamental question which requires resolution is whether AI’s lack of corporeal existence is sufficient to deny it IP rights. This conundrum forces us to confront a related question – is the IP system, which until now has been intimately associated with the human creative and inventive spirit with its respect and reward for the encouragement of human innovation and creativity – more about advancing creative and technological progress or upholding human rights?

Dr Francis Gurry, immediate past Director General of WIPO, noted that “intellectual property is key to economic development”,Footnote ⁵⁹ and studies have shown that IP rights benefit developing as well as developed nations.Footnote ⁶⁰ There have always been developmental and commercial aspects to IP, even with the fundamental objective of the patent system being to encourage investment of human and financial resources and risk taking in generating inventions that may positively contribute to the welfare of a society, promote creations, distinguish the origin of goods and services and prevent consumer confusion.Footnote ⁶¹ Even national constitutions promote IP; for example, the American Constitution’s Patent and Copyright Clause states that “[The Congress shall have power] to promote the progress of science and useful arts, by securing for limited times to authors and inventors the exclusive right to their respective writings and discoveries”;Footnote ⁶² AI could undoubtedly increase the pace of innovation and creation.

There is potential benefit to the signalling value of protecting the IP of AI-generated creations and inventions, including that jurisdictions which permit the registration of IP rights to an AI for its inventions or creations would be seen as pro-innovation and perhaps attract more development, investment and employment opportunities. Likewise, a lack of IP protection for AI-generated inventions might discourage companies from investing in AI technologies and prevent breakthroughs in important areas like drug discovery.Footnote ⁶³

What is certain is that a lack of protection will lead to greater use of trade secrets, which could serve to further retard innovation and knowledge dissemination. While limiting the use of trade secrecy could potentially mitigate this potential problem, attempts to do so could violate obligations undertaken in free trade agreementsFootnote ⁶⁴ and would definitely meet fierce resistance by tech giants and other AI companies dependent on the protection of confidential algorithms and other information for business pursuits. Moreover, countries considering weakening trade secrecy laws would do well to remember India’s past attempt to emasculate trade secrecy – when India attempted to force Coca-Cola to release its secret recipe under its Foreign Exchange Regulation Act of 1973, Coca-Cola refused and simply left the country.Footnote ⁶⁵

But the question remains unanswered whether prioritizing innovation and creation over people is fair or wise. As AI continues to increase in sophistication, society may be unwilling to sacrifice individual rights at the altar of innovation. Industry promises to protect the rights of marginalized groups and individuals, but such promises often ultimately ring hollow as history has shown that such self-regulation can be woefully inadequate at protecting people, particularly those in marginalized communities who are frequently targeted by manipulation campaigns.Footnote ⁶⁶ Furthermore, in times of global economic crisis the priority may be coping with the significant socio-economic challenges brought about by the COVID-19 health crisis.Footnote ⁶⁷

In sum, there are countless legal, technical and policy arguments for and against ownership of IP by an AI in the areas of patents, copyrights, design rights and trade secrets, as well as questions regarding property rights in the data or whether to establish a sui generis system for original content, and posing one question in one area can generate many others elsewhere. At one level, there is no practical need to let an AI become an IP holder. After all, one could simply name a human in an application the way some companies designate their chief engineer in patent applications, even though the actual inventors were other employees; or it has been suggested that the system could treat AI as we would treat a pet, arguing that pets have intelligence and a certain level of autonomy but not legal personhood. Similarly, the AI operator legitimately controls, confines and possesses the AI during conception and thus ownership of the AI invention should be held by the AI operator, their employer (work-of-hire) or successor.Footnote ⁶⁸ While the latter approach allows for easy identification of the origin of the invention and a true entity entitled to the exclusive right, it does not suffer from problems of wrongful credibility (i.e. truthfully showing the involvement of an AI, and avoiding divisive discussions of legal personhoodFootnote ⁶⁹). This model, however, may not work or may seem unfair where human intervention is minimal.

The final challenge to address is how the question of IP rights affects business investment. Some have claimed that the USA’s more permissive software patenting regime than Europe is a primary reason why more software development took place in America.Footnote ⁷⁰ This may be overblown, as the USA did not see an outflow of investment, innovation or talent following the effective raising of standards after the US Supreme Court’s decision in Alice Corp. v. CLS Bank International, where the court avoided giving a clear definition of the expression “software patent” and held that “merely requiring generic computer implementation fails to transform [an] abstract idea into a patent-eligible invention”.Footnote ⁷¹ Similarly, the 2018 report on the impact of the Database DirectiveFootnote ⁷² made no mention of any great new flows of technological investment into the European Union as a result of the Directive.

A Stakeholders in Conventional Trade

B Stakeholders in Digital Trade

The TREC Agreement that is under negotiation would provide for the rights and obligations of contracting parties similar to those digital trade rules under recent RTAs, which serve as important references. For example, Article 19.3.1 of the USMCA prohibits contracting parties from imposing customs duties on digital trade, and Article 19.4.1 of the USMCA requires contracting parties to accord no less favourable treatment to a digital product created by another party or by a person of another party.Footnote ¹⁷ In addition, some digital trade rules under the USMCA, such as Articles 19.5.1 and 19.7.2, require the contracting parties to adopt or maintain certain laws on digital trade within their territories.Footnote ¹⁸ Under Article 8.74 of the JEEPA, contracting parties are required to ensure that all the measures of general application affecting electronic commerce are administered in a reasonable, objective, and impartial manner.

However, many rules under the proposed TREC Agreement would also concern the rights and obligations of private entities, at least indirectly. First, the protection of the interests of consumers would be a central element in the TREC Agreement, as the existing regional trade rules on digital trade suggest. For example, Article 19.2.1 of the USMCA recognizes the importance of a framework to promote consumer confidence in digital trade. More specifically, Article 19.7.1 of the USMCA explicitly emphasizes the importance of adopting and maintaining transparent and effective measures to protect consumers from fraudulent or deceptive commercial activities.Footnote ¹⁹ Similarly, Article 19.8.1 of the USMCA recognizes the economic and social benefits of protecting the personal information of users of digital trade, as well as the contribution this makes to enhancing consumer confidence in digital trade.Footnote ²⁰

Second, the conduct of enterprises and other private entities would be indirectly subject to rules under the TREC Agreement through the domestic laws of the contracting parties, since such conduct could undermine the interests of consumers protected under the TREC Agreement. For example, Article 19.7.2 of the USMCA requires contracting parties to adopt or maintain consumer protection laws to proscribe fraudulent and deceptive commercial activities that cause harm or potential harm to consumers engaged in online commercial activities.Footnote ²¹ Similarly, Article 19.8.2 of the USMCA requires contracting parties to adopt or maintain a legal framework that provides for the protection of personal information of the users of digital trade.Footnote ²² Most notably, Article 19.13 of the USMCA requires contracting parties to not only adopt or maintain measures regarding certain unsolicited commercial electronic communications sent to an electronic mail address, but also to provide recourse in its law against suppliers of unsolicited commercial electronic communications that do not comply with any measure adopted or maintained pursuant to this obligation.Footnote ²³ Thus, consumers would be able to seek recourse against an enterprise in domestic procedures by claiming that its conduct violates the relevant domestic law incorporating rules under the TREC Agreement.

Third, enterprises and other private entities would also be beneficiaries whose interests must be protected under the TREC Agreement. For example, provisions such as Article 19.5 of the USMCA, concerning the domestic electronic transactions framework, and Article 19.6 of the USMCA, concerning electronic authentication and electronic signatures, are intended to facilitate business activities in digital trade.Footnote ²⁴ In addition, provisions prohibiting the contracting parties from requiring localization of computing facilities,Footnote ²⁵ or the transfer of, or access to, source codes,Footnote ²⁶ are inserted to address one of the most urgent concerns of enterprises.

These features of the rules under the proposed TREC Agreement would characterize the nature of disputes arising under the Agreement in two ways. First, while disputes may arise between contracting parties under certain circumstances, a greater number of disputes would likely arise between a consumer and an enterprise, or between an enterprise and the government. For example, a consumer may claim that certain practices of an enterprise inappropriately use his or her personal information. Alternatively, an enterprise may claim that a regulation of the government unduly restricts its business activities in digital trade. Second, disputes would more often arise under domestic law rather than directly under the TREC Agreement. The TREC Agreement would presume that many of its rules are to be incorporated into the domestic laws of the contracting parties. It would be reasonable for consumers and enterprises to refer first to a relevant domestic law to determine if their benefits are legally protected under said domestic law. These characteristics must be considered when constructing dispute settlement mechanisms for digital trade disputes.

A Non-Trade Values in Conventional Trade

B Non-Trade Values in Digital Trade

The TREC Agreement under negotiation would provide exceptions similar to Article XX of the GATT and Article XIV of the GATS, with a view towards protecting non-economic interests. In fact, digital trade rules under recently concluded RTAs provide GATT Article XX-type exceptions. For example, Article 19.11 of the USMCA provides that while no party shall prohibit or restrict the cross-border transfer of information, a party is not prevented from adopting or maintaining a measure that is inconsistent with the obligation but “necessary to achieve a legitimate public policy objective, provided that the measure is not applied in a manner which would constitute a means of arbitrary or unjustifiable discrimination or a disguised restriction on trade; and does not impose restrictions on transfers of information greater than are necessary to achieve the objective.”Footnote ³⁵ Disputes arising out of these exceptions may be settled in a manner similar to the settlement of disputes involving Article XX of the GATT, through the weighing and balancing of the right of a contracting party to invoke an exception and the substantive rights of the other contracting parties protected by the proposed TREC Agreement.

However, digital trade rules under RTAs suggest that balance between trade benefits and non-trade values would also need to be sought in different circumstances under the proposed TREC Agreement. First, a measure justified as an exception under the TREC Agreement would impact the interests of a specific private entity rather than the interests of another party. For example, Article 19.16.2 of the USMCA provides that a regulatory body or judicial authority of a party is not precluded from requiring a person of another party to preserve and make available a source code of software or an algorithm expressed in that source code to the regulatory body under certain circumstances, while Article 19.16.1 generally prohibits the parties from requiring the transfer of, or access to, such source code or algorithm as a condition of the import, distribution, sale, or use of that software.Footnote ³⁶ Specific circumstances under which disclosure of a source code is required would be provided in the domestic law of each contracting party. Disputes involving this exception would require the weighing and balancing of the public policy objectives of a contracting party invoking the exception against the economic interests of a private person who is required to make available its source code, which may not be properly undertaken through the WTO’s inter-governmental dispute settlement procedures.

Second, in some cases, the proposed TREC Agreement would provide obligations, rather than exceptions, to take certain measures to achieve legitimate non-economic objectives. For example, Article 19.7.2 of the USMCA requires contracting parties to adopt or maintain consumer protection laws to proscribe fraudulent and deceptive commercial activities that cause harm or potential harm to consumers engaged in online commercial activities.Footnote ³⁷ Similarly, Article 19.8.2 of the USMCA requires contracting parties to adopt or maintain a legal framework that provides for the protection of personal information of the users of digital trade.Footnote ³⁸ Additionally, Article 19.13 of the USMCA requires each party to adopt or maintain measures to limit unsolicited commercial electronic communications.Footnote ³⁹ These provisions require, rather than allow, contracting parties to restrict digital trade to achieve legitimate non-trade objectives.Footnote ⁴⁰ It is questionable at best to assume that disputes involving such obligations can be properly regarded as “trade” disputes, and that non-trade policies can be properly examined by panels and the Appellate Body.

A Conventional Trade Disputes

B Digital Trade Disputes

The previous subsections have pointed out that digital trade rules in the proposed TREC Agreement will likely differ from conventional trade rules in the WTO agreements in terms of the diversity of stakeholders and the nature of the balance between trade and non-trade values. These differences suggest that disputes arising from the TREC Agreement may take unique forms when compared with conventional trade disputes. More specifically, digital trade disputes can take six different forms, depending on the nature of the involved parties.

First, disputes may arise between contracting parties to the proposed TREC Agreement. Some provisions of the TREC Agreement may provide for the rights and obligations of the contracting parties, and violations of these provisions may trigger disputes between parties. For example, if a contracting party imposes a customs duty on digital trade in violation of the TREC Agreement, another contracting party whose digital product is subject to the duty may bring a dispute against the party imposing the duty. Alternatively, a contracting party may claim that a restriction to cross-border flows of personal data imposed by another contracting party is a violation of the Agreement, while the latter party may claim that the restriction is justified as an exception to achieve its legitimate public policy objective.Footnote ⁴²

Second, digital trade disputes may also be brought by a business enterprise of a contracting party against another contracting party. Some of the provisions under the proposed TREC Agreement may require contracting parties to protect the interests of enterprises engaged in digital trade. If a contracting party fails to take appropriate measures to do so, it may face a claim by an enterprise, arguing that the contracting party has violated requirements under the TREC Agreement. For example, an enterprise of a contracting party may claim that it is forced to transfer its source code to the government of another contracting party, contrary to obligations under the TREC Agreement. This type of dispute may also arise between an enterprise of a contracting party and its own government.

Third, the government of a contracting party may claim that an enterprise of another contracting party has engaged in unfair digital trade practices. For example, a contracting party may consider that an enterprise from another contracting party is abusing the personal data of its consumers and is therefore violating the obligations of its domestic law, incorporating rules of the proposed TREC Agreement. It could handle the matter pursuant to its own domestic law, but it may also seek to consult with the government of the other contracting party on the matter.

Fourth, digital trade disputes may be disputed between enterprises of different contracting parties if the conduct of an enterprise of a contracting party undermines the digital trade activities of the enterprises of another contracting party. Although many of these disputes between enterprises are commercial in nature, they may involve issues related to the interpretation and application of the proposed TREC Agreement.

Fifth, digital trade disputes may also be brought by a consumer of a contracting party against its own government. As discussed earlier, the proposed TREC Agreement requires contracting parties to protect consumer interests, such as privacy, through domestic laws and regulations. A consumer may claim that his or her government’s failure to do so constitutes a violation of the TREC Agreement.

Sixth, and finally, digital trade disputes may be brought by a consumer of a contracting party against an enterprise of another contracting party when the former considers that the conduct of the latter violates its interests, as indirectly protected under the TREC Agreement. In such cases, the consumer may seek to obtain remedy from the enterprise.

What would be the appropriate form of dispute settlement for such digital trade disputes arising under the TREC Agreement? The first category of disputes is similar to conventional trade disputes and could therefore be dealt with under general trade dispute settlement procedures. The TREC Agreement should provide that the rules and procedures under the DSU shall apply to disputes arising under the Agreement. Nonetheless, some special or additional rules would be needed in order to allow the contracting parties some flexibility in the implementation of the Agreement in light of the novel and evolving nature of digital trade. For example, both developing and developed parties should be given grace periods, during which a contracting party would refrain from using the dispute settlement mechanism. The use of enforcement measures, such as suspension of concessions, should be restricted.

The second category of disputes is similar to certain disputes under the Anti-Dumping Agreement. As in the case of disputes under the Anti-Dumping Agreement, this type of dispute would be best dealt with through the domestic procedures of a contracting party taking measures at issue. To effectively address this category of disputes, the TREC Agreement should require the contracting parties to establish and maintain domestic procedures that are accessible to enterprises. This type of dispute would be principally reviewed under domestic laws and regulations that have incorporated the rules under the TREC Agreement. The TREC Agreement could explicitly require domestic tribunals to apply domestic laws and regulations, in accordance with the TREC Agreement.

The third category of disputes shares some features with certain disputes under the GATS, in that the disputes are triggered by the conduct of private entities. It would be useful for the TREC Agreement to provide consultation procedures, by which the government of a contracting party can request consultations with the government of another contracting party regarding the enterprises of the latter party. The complaining party may also seek to apply its domestic law to a foreign enterprise allegedly engaged in trade-restrictive practices. A cooperative mechanism would be desirable to avoid the excessive extraterritorial application of domestic law.

The fourth category of disputes may be better dealt with outside the framework of the TREC Agreement in light of its commercial nature. Existing judicial and non-judicial procedures employed to handle commercial disputes could also be used to address this category of disputes.

It is essential that the proposed TREC Agreement would be capable of properly settling the fifth and sixth categories of disputes, given the importance of the protection of consumer interests. Principally, these types of disputes should be dealt with through domestic procedures because they are easily accessible to consumers. Domestic procedures are also desirable because domestic courts and tribunals are better suited, when compared with international mechanisms, to make decisions regarding how to weigh and balance the different interests of consumers and enterprises within the jurisdiction of a contracting party. To ensure that domestic procedures function as an effective dispute settlement mechanism for consumers, the TREC Agreement should require contracting parties to not only establish and maintain domestic procedures that are accessible to consumers, but also ensure that domestic laws and regulations are applied in accordance with the TREC Agreement.