Search the archive.

Daniel Zamora is a postdoctoral sociologist at the Université Libre de Bruxelles and Cambridge University. His book, Le Dernier Homme Et La Fin De La Révolution: Foucault après Mai 68 (The Last Man and the End of the Revolution: Foucault After May 68) coauthored with Mitchell Dean, was published in English by Verso in 2020; Can We Criticize Foucault?; ; [TMS—GK]

Debray writes in Modernes catacombes that Foucault, the rebel and subversive, has become an “official philosopher.” How do you understand this paradox? And how do you explain how Foucault can seduce so many of those in radical milieus who nevertheless affirm with force that they wish to put an end to the neoliberal era? It’s a very interesting question, but one I don’t have a satisfying answer to. I would, nevertheless, suggest that it’s in large part due to the structure of the academic field itself. You’d have to go back to Bourdieu and the precious works of Louis Pinto to better understand this evolution. It should never be forgotten that joining a “school,” or associating oneself with a certain theoretical perspective, means associating oneself to an intellectual field, where there is an important struggle for access to the dominant positions. Ultimately, calling oneself a Marxist in the France of the 1960s — when the academic field was in part dominated by self-identified Marxists — did not have the same meaning as it does to be a Marxist today. Concepts and canonical authors are obviously intellectual instruments, but they also correspond to various strategies for becoming part of the field and the struggles over it. Intellectual developments are then partly determined by relations of power within the field itself. Also, it seems to me that relations of power within the academic field have changed considerably since the end of the 1970s: after the decline of Marxism, Foucault occupied a central place. In reality, he offers a comfortable position that allows a certain degree of subversion to be introduced without detracting from the codes of the academy. Mobilizing Foucault is relatively valued, it often allows his defenders to get published in prestigious journals, to join wide intellectual networks, to publish books, etc. Very wide swaths of the intellectual world refer to Foucault in their work and have him saying everything and its opposite. You can be an adviser to the MEDEF and edit his lectures! [A reference to François Ewald, adviser to the main French business federation; see above.] I would say that he opens doors. And you can’t really say the same of Marx nowadays. This critique of the “margins” as the center of political combat could end up delighting all manner of counter-revolutionaries in France or Belgium. Aren’t you afraid of playing into their hands? I do think there exists a “conservative” critique of Foucault — and more broadly of what May ’68 represents in French social history. This critique is no longer marginal at all: you can find it among the thinkers of the conservative right like Eric Zemmour or within the National Front. It openly critiques the whole feminist, anti-racist and cultural legacy of May ’68 while having much less to say about the economic ravages of neoliberalism. It’s as if the problem were the political liberalism that came with the 1980s, and only by going back on these societal evolutions will we be able to “faire société.” One often hears this kind of thinking, according to which it was the destruction of family values or communitarian forms of the social bond that allowed the expansion of neoliberalism. There may be a grain of truth in these analyses, but they are totally deluded when they propose a return to more “traditional” ways of life. We’re heading towards a much more authoritarian kind of liberalism, with a return to family values, a return to a total fantasy of national culture, and the good old pre-globalized capitalism… As for the idea of “playing into their hands,” I don’t think it’s a problem. If there’s a problem with certain aspects of the legacy of May ’68, the role of the Left is not to close its eyes because the far right is saying it, but on the contrary, to render its own judgment, to formulate its own critique, so as not to totally lose the ideological battle. That is the task we need to get started on in order to reconstruct a left that is both radical and popular.

(David, Language, History, And Class Struggle in In defense of history)

It should be emphasized that this argument differs substantially from poststructuralist accounts that may in some respects look similar. Michel Foucault, for example, is noted for his view of language as a terrain of power and domination embodied in particular social institutions such as hospitals, asylums and prisons; and, especially in some of his later writings, Foucault appears to allow for the possibility of resistance to practices of domination. Foucault’s emphasis on power was an attempt to counter the political emptiness of theories that reduce social relations to their linguistic forms. In a vigorous riposte to Derrida, he charged deconstruction with locating everything in texts “so as not to put discursive practices back into the field of transformations in which they are carried out.”4 This position has considerable strengths. Yet, while disavowing enclosure within texts, Foucault himself fails to break out of discourse as a field closed in upon itself. Indeed, his whole concept of “power-knowledge” tends towards the position that power relations are constructed in and through practices of “knowing” humans. And these practices—of classifying, measuring, and surveying people—originate in the realm of discourse, where people are assigned to different categories of thought and description. It follows that “the development of humanity is a series of interpretations." Thus, for all his insights into social institutions and their practices of domination, Foucault reverts to his own version of discursive determinism; he is led by the logic of his position to conclude that inevitably “one remains within the dimension of discourse.”5(29)

Sui Generis protection. The history of sui generis approaches has been that as technology advances, they either quickly become obsolete (e.g., Semiconductor Chip Protection Act of 198419), or may raise uncertainties and impediments pertaining to copyright.20 ***[Footnote 20 Begins]*** The DMCA is now 25 years old and was based on at least a decade of comment, proposal, and treaty. Yet controversy swirls as to its interaction with copyright law, and even the Office has not yet felt to be in a position to comment, where both DMCA and copyright issues are in question, on whether a device owner enjoys ownership of embedded software. The online “safe harbor” provisions also remain controversial. (CTA is not suggesting the DMCA was a mistake – only that adding sui generis provisions can complicate, or even impede, already complicated copyright determinations.) ***[Footnote 20 Ends]*** Therefore CTA urges caution in such respects.

5.25The cluttering of the register with utility models that are probably not valid (or at the very least partially invalid) is posing a problem in some areas of technology. Not only does this add to the costs of checking that a product infringes no IP rights, but also can lull a holder of a utility model into a false sense of security. The certificate of registration—unlike a regular patent—can never be given the ‘benefit of the doubt’ for validity. Some utility model systems, for example in Croatia or Japan, allow or require the proprietor to request complete ‘substantive’ examination of the utility model. Other systems, such as Germany, only provide for a search to be carried out at the request of the applicant;34 complete examination is restricted to a revocation request35 or as a counterclaim in an infringement procedure. 5.26 The value to the company of the (unexamined) utility model is also difficult to ascertain if there is no examination. Venture investors are looking for valuable rights and if no patent examiner has looked at the intellectual property rights in detail, then their value can be uncertain. The EMB developers need the funding to continue with their product and their backers would want strong IP protection. The lack of examination can be overcome, at least partly, by having a patent expert review the rights, but this lacks the force of law, and its costs can easily be higher than the cost of patent prosecution.

Several countries have adopted Sui-generis laws for protecting traditional knowledge from patenting. These include South Africa, Peru, Panama, Costa-Rica, Portugal, India and Thailand. A number of countries have decided to develop national legislation to protect traditional knowledge. In addition inter governmental and non-governmental organizations have produced model legislation for interested countries to adopt. In June 2000 Panama has passed a Suigeneris law by name “Panama’s special system for registering the collective rights of indigenous people’s for the protection and defense of their cultural identity and traditional knowledge and setting out other provisions”.8 According to World Intellectual Property Organization (WIPO) the Sui-generis system of Panama actually constitutes the first comprehensive system of protection of traditional knowledge ever adopted in the world.9 The object of this law is to protect the collective intellectual property rights and the traditional knowledge of the indigenous peoples over their creations. Such creations include “invention, models, patterns and designs, innovations contained in images, figures, symbols, diagrams and petroglyphs”.10 Also included are the cultural elements of their history, music art and traditional artistic expression capable of commercial use”.11 Indian Parliament has passed Sui-generis laws such as, “The Indian Biological Diversity Act 2002” and “Plant Verity Protection and Farmer’s Rights Act 2001”. Policy objectives of “The Indian Biological Diversity Act 2002 are-“ to provide for conservation of biological diversity., sustainable use of its components and fair and equitable sharing of the benefits arising out of the use of biological resources and knowledge”12 Objectives of “Plant Variety Protection and Farmers Rights Act 2001” are to provide for strong, proactive farmers rights, registration of farmers varieties, Provisions for benefit sharing through a national gene fund and the requirement for prior informed consent for use in breeding”.13 Conclusion: As Sui-generis law may be a combination of intellectual Property law, customary law, equitable benefit sharing provisions, provisions for the rights of farmers and breeders, provisions of prior informed consent, provisions of contractual agreements and provisions of disclosure of origin of biological/genetic resources it is the best system of law for the protection of traditional knowledge. Sui-generis law better protects the rights and interests of traditional knowledge holders. Suigeneris law intends for the wider application of the knowledge, innovations and practices of indigenous and local communities with the involvement of holders of such knowledge, innovations and practices

There shall be adequate protection available against misappropriation of TK, the reason being that it affects the livelihoods of the TK practitioners. A sui generis legislation shall also ensure opportunity to any person even outside the custodian community for enrichment of TK. However, financial benefits from such exploitation of TK should flow down and reach all the stakeholders. This can only be ensured if knowledge societies are registered and its membership is governed systematically. In order to create such a system, it is to be ensured that TK holders/communities register a "knowledge society" according to the laws of the country. The TK holders/communities shall be recognized as custodians of the said TK only after that. After registering the "knowledge society", they can apply to the appropriate Government (deemed owner of TK) for recognition as custodians. It is the responsibility of the respective state governments to encourage individual practitioners of tK or communities to form knowledge societies. When the custodians are officially recognized, they get the power to manage the TK according to the statute. They are also empowered to market their TK as they deem appropriate, ensuring that any commercial value in it would benefit them first. Government authorities should empower the TKcommunities to obtain Patents on the innovations made by them on the TK and to negotiate with the potential customers. It is also necessary to educate them that no patents could be obtained, within India or abroad, on any TK per se or aggregation thereof, obtained from any region or place, irrespective of whether it is in the custody of a community/ individual or in public domain

(TK – Traditional Knowledge)

GIs are also important intellectual property rights employed to protect intangible cultural heritage. GIs are signs given different terms in national and international norms. They play a similar role to appellations of origin (AOs), in that both GIs and AOs highlight a qualitative link between the relevant product and its place of origin. The link with the place of origin is stronger for AOs, since the quality or characteristics of the product must result exclusively from their geographical origin: raw materials should be sourced in the place of origin, as should the processing of the product. In the case of GIs, a single criterion attributable to geographical origin is sufficient: the production of the raw materials and the development or processing of a GI product do not necessarily have to take place entirely in the defined geographical area. For the sake of simplicity, GIs and AOs will be referred to collectively as GIs, unless indicated differently. The legal instruments and the protection available for GIs vary considerably from one country to another. A sui generis system providing for the registration of GIs exists in some states, e.g., Kenya, Nigeria, South Africa, Egypt, Mexico, Peru, Cuba, India, Pakistan, Mauritius, and Sri Lanka. The most advanced of these sui generis systems is that of the EU, where GIs (as well as Protected Designations of Origin (hereinafter PDOs)38 and Traditional Speciality Guaranteed (hereinafter TSGs))39 are regulated by the EU Regulation No 1151/2012 (QS Regulation).40 The registration of EU GIs gives rise to a unitary intellectual property right (overriding any national rules on domestic GIs) covering the entire EU territory (Article 118(2) TFEU).41 The exhaustive nature of the QS Regulation distinguishes it from other EU Regulations on unitary EU intellectual property rights, e.g., the EUTM Regulation, the Regulation on Community Designs,42 and the Regulation on Community Plant Variety Rights.43 In fact, according to these Regulations, the interested parties may choose to register an EU trade mark according to the EU Regulation or a national trade mark based on the domestic legislation(s) in force in the member state(s) of registration.44 In contrast, interested parties can only register their GIs in the EU, based on the QS Regulation, and can no longer register a national GI in a single member state.

Current law is at best unclear. Judicial precedent is muddled and legislation has frequently been interpreted beyond its explicit meaning and guiding principles. This state of affairs is simply the product of the logical desire to fit a new human endeavour into the same legal framework that has governed novel endeavours from time immemorial. However, as the law has developed, its policy objective has become obscured by the push and pull of opposing lobbies in the debate. Demonstrating the resulting ambiguity of the current regime will involve looking in detail at the current state of the law. Observation of every facet of reasoning in this complex area of law is beyond the scope of this paper, however, and so with a view to giving an impression of that complexity and the bewilderment it causes, the most salient problems will be focused upon. It is not intended to examine trade secrecy law, given the more material and contentious issues surrounding the laws of copyright and patent. Suffice it to say, trade secrets, which prevent disclosure of works such as formulas and processes, are generally considered to be something of a dead duck in terms of software. This is due to the historical shift of emphasis in the industry from the practice of custom-designing and installing computers, where software secrets tended to be useful on only very few machines, to the sale of off-the-shelf computer programs, which operate on a wide range of hardware and can be running on thousands if not millions of general-purpose computers simultaneously.7 Considering whether the replacement of the current regime with a sui generis system is the best way forward will involve not just the (clearly important) objective of clarifying the law, but also the consideration of certain important real world factors. Given the omnipresence of software in our lives and its importance in the commercial world, such extra-legal factors will reveal some important considerations for the future of the legal protection of software. These will be considered further below.

Artificial Intelligence (AI) is a broad term used to describe a multi-aspect discipline in computer science that includes technologies that replicate functions ordinarily attributed to human intelligence. Unlike ordinary software, AI possesses capabilities for learning; when an AI system is exposed to more datasets, the results achieved are generally improved. 1 In one’s daily life, a person may interact with numerous systems and devices that make use of AI, including those that rely on deep neural networks, natural language processing, image data classification schemes, and AI-tailored hardware. For example, navigational systems may use AI to find the best route to a destination; chatbots may service customers by answering or providing further information to questions through the use of AI-based natural language processing; online retailers may use AI-based analytics to inform a customer’s preferences and likely next purchases; and self-driving vehicles may use AI together with a combination of sensors, cameras, and radar systems to travel without a human operator. AI is everywhere, and it is a key driver in the Fourth Industrial Revolution, a technological revolution that will fundamentally alter the way people live, work, and relate to one another.2 In a study by PWC, AI is projected to add $15.7 trillion dollars to the global economy by 2030, boosting US GDP by 14.5%.3 Emerging AI technologies are predicted to transform the global economy and have significant implications for America as a global innovation leader. Increased investment in AI-based innovations requires careful assessment of the intellectual property protections afforded to such innovations. Certain aspects of AI may be protectable by traditional IP protection rights, such as patents, copyrights, and trade secrets. For example, Patents may protect elements of AI inventions, e.g., the structure of the database on which the AI will be trained and will act, the process of training the AI model using one or more datasets, the AI algorithm as applied to a technical problem, the results or insights of the AI invention through an automated process, the policies/weights to be applied to the data that affects those results or insights; Copyright protection may apply to the expression or implementation of an AI algorithm or process, such as, in the form of software, the selection of data as embodied in a copyrightable compilation, the output provided by an AI-based system, such as the artwork, the musical production, or journalistic articles; and Trade secrets may protect secret AI-based information not commonly known or available to the public, that derives economic value from not being generally known, such as AI training data sets. One important aspect of AI is the model itself, i.e., the formulas, processes, and methods for handling data in an AI-based system. This paper explores how aspects of an AI model may be protectable by patents, copyrights, or trade secrets. It also explores how traditional IP-protection mechanisms may be unavailable or inadequate to provide adequate legal protection for the AI model itself, despite the fact that such AI models may warrant legal protection. Additionally, this paper considers the creation of a sui generis right for trained AI models. The term “sui generis” means “of [its] own kind; in a class by itself; unique.”4 A sui generis right is created legislatively to protect a very specific type of work that may not be adequately protected by traditional forms of IP. Examples of current sui generis IP protections include integrated circuit layouts, ship hull designs, fashion designs, databases (EU), and plant varieties. In considering the creation of a sui generis right for AI models, this paper addresses numerous topics should such a protection right be created, including, inter alia, policy considerations, the subject matter covered, rights conferred, registration requirements, fair use, marking, the duration of such a right, and litigation issues. While the issues evaluated herein are done so with a USjurisdictional focus, some concepts presented may be applicable elsewhere. II. ARTIFICIAL INTELLIGENCE, MACHINE LEARNING, AND AI MODELS Artificial intelligence (“AI”) is an umbrella term for a class of different computer technologies which focus on replicating one or more functions attributed to human intelligence. One of the biggest fields today are algorithms that replicate human learning. Machine learning (“ML”) is focused on providing computers with the ability to learn without being explicitly programmed. Typically, with ML, a large data set is digested to produce a trained model that can, in deployment, receive the same types of inputs and output a decision, classification, etc. Typically, the AI step is the training; the trained model is the result. After the AI step, a trained model is often thought of as a black box. In truth, a set of coefficients for use in one or more mathematical functions is created. These coefficients need to be used wherever the trained model is to be deployed. How an AI model trains data is exemplified as follows: (1) a data set labeled input and output examples are fed to an algorithm; (2) then, the algorithm runs on the data with its parameters adjusted until it reaches a satisfactory level of accuracy; (3) from this analysis, the algorithm creates a function that can predict future outputs. Suppose, for example, the AI model is given pictures of cats that are labeled as “cats.” The model is then trained on the labeled data of “cats” until it can recognize the patterns in the images of cats. As a result, the model would be able to predict if later images are showing “cats” or “not cats” by responding to the previously recognized patterns.5 A lot of effort goes into the “AI step,” including, data acquisition, data curation, data labeling, model architecture selection, training algorithm selection and configuration, the training itself, validation, performance evaluation, tweaking, and iteration. These efforts result in a collection of coefficients; with the coefficients in hand, the above efforts can henceforth be skipped. After putting in this hard work, one must consider, what are the best strategies to protect the fruit of the such efforts, even when only textbook (non-inventive) methods and architectures are used? Copyrights Copyright, is a form of protection in creative works that grants an owner of that copyright the exclusive rights to reproduce, display, perform, and create derivative works of such creative works.6 To qualify for copyright protection, a work created must be original to the author and possess some degree of creativity.7 1. Machine-Created Works are Not Copyrightable Turning to whether the works created by machines are copyrightable, the USPTO and other intellectual property offices throughout the world have provided guidelines that only works created by humans can be protected by copyright.8 According to the Compendium of U.S. Copyright Office Practices, “[c]opyright law only protects works of authorship that are created by human beings. Works made through purely mechanical processes or with an automated selection and arrangement are not eligible for copyright protection. The U.S. Copyright Office will refuse to register a claim in a work that is created through the operation of a machine or process without any human interaction, even if the design is randomly generated.”9 In a recent federal copyright case, known as the monkey selfie dispute, the court of Appeals of the Ninth Circuit held that an animal is not entitled to a copyright.10 Furthermore, since copyright protects the expression of ideas and not the ideas itself, the software code underlying the AI models and training datasets, used to train the AI models, may be protected as a literary expression under copyright laws.11 Copyright protection, however, is unlikely to be available for the functionality of AI models in automating a decision process, the underlying algorithms, and output of the AI models. 12 It is hard to argue a modicum of creativity in a collection of numbers generated by a machine. 2. Machine-Created Works Must be Creative The U.S. Copyright Act requires the work to be creative, meaning the work must be the product of a creative spark in order to be copyrightable. Works lacking creativity are not afforded copyright protection.13 The U.S. Supreme Court has shed light on this creativity requirement and ruled that compilation works, such as a database must contain a level of creativity in order to be copyrightable. Accordingly, just listing names, phone numbers, and addresses lack creativity.14 Under Section 313.2 of the Compendium of the U.S. Copyright Office, works produced by a machine or mere mechanical process that operates randomly or automatically without any creative input or intervention from a human author will not be afforded copyright protection.15 Machinegenerated AI models replicate a decisions process that enables automation and understanding. These AI models are trained using data and human expert input to automate the decision process.16 Merely automating a process by an expert would most likely be considered to lack creativity, and therefore, would not be afforded copyright protection. 3. Enforcement and Infringement Considerations The AI field, namely AI system, is a new and constantly evolving technology, with minimal clarity on the working and processes of the AI system, outside of the AI developer community.17 Such uncertainty has created issues in regards to establishing and proving infringement, as well as enforcement issues on the legitimacy of using third-party copyright-protected content to train AI models with varying laws in different jurisdictions. 18 For example, current UK copyright laws and the EU Copyright Directive (to be implemented in the EU Member States before June 2021) permits “text and data analysis for non-commercial research.”19 However, Section 107 of the U.S. Copyright Act provides a statutory framework for the “Fair Use Doctrine” that permits the use of the third-party content to train AI systems, if the use is sufficiently transformative and does not compete with the original works.20 The Fair Use Doctrine, therefore, creates an exception to infringement in the U.S. that is not applicable in other jurisdictions. Based on the above described limitations of copyright protection, the extension of such traditional copyright protection to machine-created works, namely AI models, is unlikely at best. However, due to the large investments in AI technology, there are legitimate business concerns associated with the IP protection of AI models.21 Such concerns have served as a catalyst to exploring alternate considerations, such as sui generis rights, or even other forms of traditional IP protection (i.e., patents) to provide IP protection to AI models. B. Patents Given that AI models fundamentally rely on mathematical constructs, the Supreme Court’s decision of Alice Corp. v. CLS Bank Int’l, 573 U.S. 208 (2014), regarding subject matter eligibly pursuant to 35 U.S.C. § 101, can create unique challenges for AI related inventions. In particular, inventions claiming AI models are at risk of being found “directed to” an abstract idea without “significantly more.” Since Alice, however, the Federal Circuit has repeatedly found software-related inventions, which AI related inventions pertain, as patent eligible where such software-related inventions demonstrate an “improvement” to the functionality of a computer.22 For example, most recently, in Uniloc USA v. LG Elec. USA, the Federal Circuit provided that "[o]ur precedent is clear that software can make patent-eligible improvements to computer technology, and related claims are eligible as long as they are directed to non-abstract improvements to the functionality of a computer or network platform itself.” No. 19-1835, slip op. at 10 (Fed. Cir. Apr. 30, 2020).23 The Federal Circuit has specifically recognized the importance of AI and the danger of applying Alice in a way to impede innovation: “the danger of getting the answers to [the] questions [raised by Alice] wrong is greatest for some of today's most important inventions in computing, medical diagnostics, artificial intelligence, the Internet of Things, and robotics, among other things.” Smart Systems Innovations, LLC v. Chicago Transit Authority, 873 F.3d 1364, 1378 (Fed. Cir. 2017). In addition, the USPTO has signaled that AI inventions are patentable. For example, in the U.S. Patent Office (USPTO)’s 2019 Revised Patent Subject Matter Eligibility Guidance (January 7, 2019), the UPSTO provides an example AI-based method for training a neural network for facial detection, that is patent eligible under Alice. 24 In general, in view of Alice, when considering AI inventions for patent protection, a patentee should focus on the unique technical features that may be identified in an AI-invention workflow and related components. These may include, for example: • The pre-processing of training data (e.g., preparing unique datasets for input into a particular AI algorithm); • The training process (e.g., improvements or adjustments to a machine-learning or neural network algorithm); • The application of trained models (e.g., to control machines or to provide unique results); and • The hardware that executes a trained AI model, including any improvements to the hardware or its deployment in a given technical field. In addition, under Alice, an important test is whether the claims of a computer or software related invention recite a specific improvement to “the functioning of a computer itself” or “any other technology or technical field.” See Uniloc USA, No. 19-1835, slip op. at 5-10. With an AI-related invention, this can be described and claimed where an AI model improves the underlying functionality of a system, e.g., an autonomous vehicle, where the AI model enhances, e.g., the safety, fuel efficiency, or power usage of the vehicle. On the other hand, while the Federal Circuit has generally recognized AI as an important technology (see Smart Systems, 873 F.3d at 1378), it has yet to address the eligibility of an AI related invention. Given this, it is possible that an AI related invention, especially one with claims having features typically found ineligible (e.g., claims directed to fundamental economic practices) may be found ineligible despite incorporating AI aspects. Accordingly, in such circumstances, other legal protections, such as those found herein, may need to be considered. C. Trade Secrets Although patenting is an option, there are also challenges to patent protection on AI models as described above. As an alternative, trade secret can protect information where the owner “has taken reasonable measures to keep such information secret” and the information “derives independent economic value, actual or potential, from not being generally known” to other persons. See 18 U.S.C. § 1839(3) (Federal Defend Trade Secrets Act, definition of “trade secret”). Protection is available at the federal and state level. A benefit to using trade secret to protect AI models is the relatively low cost. Unlike patents, trade secrets involve no registration and have immediate effect. Furthermore, trade secrets do not require compliance with formalities or public disclosure. If AI models are likely to be continuously revised or retrained, then this low-cost benefit can make trade secrets an appealing option. On the other hand, trade secrets may be an inefficient means of protection in software development. For example, if the model is deployed on a product, reverse engineering becomes possible. A trade secret is protected as long as it remains a secret. Even if reverse engineering is prohibited by contract, the value of one’s crown jewels may be significantly diminished if it is already exposed publicly. The algorithm can be moved to a server to better protect against reverse engineering, but that may not always be practicable. Additionally, collaborative scientific research is intrinsic to the development of new technology, leading to parties engaging in joint ventures, subcontracting relationships, and/or licensing opportunities.25 While parties can take steps toprotect their trade secrets, such as through nondisclosure agreements and restricted access, difficulties in maintaining confidentiality continue to persist.26 Some industries also make trade secret protection impracticable because solutions need to be proven or disclosed before acceptance. For instance, a greater demand for AI systems to be transparent to other technology producers or consumers in the general public will make it difficult to keep valuable AI models a trade secret. D. Open Contribution – Journals, Open Source, Etc. Open contribution, specifically open source contribution or journals, is a form of free contribution to an AI model or other aspect of an open source software.27 In open contribution, particularly for open source software, members of the public may work collaboratively by providing comments, or edits to various aspects of a project.28 The owner of the open source software, or AI model, may review and determine whether to accept or reject the edits, as well as whether to modify the project based on the commented provided. In general, open contribution of software reduces development time, community supported development and code review, and platform adoption. To maintain IP integrity, various forms of IP protection may be implemented, such as customized or standard open source licensing agreements, or a hybrid approach in which traditional IP protection (e.g., copyrights and patents) are still sought in parallel to seeking open contribution.29 When traditional IP protection, namely patents and/or copyrights, are sought with open contribution, traditional IP protection should be sought prior to open contribution and should be broad enough to include any substantive modifications implemented on the AI model.30 Otherwise, the owner of the AI model may be precluded from seeking any patent or copyright protection for an AI model. On the other hand, by implementing a non-traditional approach for IP protection, open contribution is more economically affordable than the more traditional copyrights or patent methods of IP protection. The cost of filing a non-provisional patent application far extends the nominal, if not free, cost of seeking open source contribution from a community of developers, as well as working with a larger pool of talented information technology professionals to improve the AI models at no additional cost, since there is zero cost associated with open contribution.31 Therefore, open contribution may lead to an abandonment or significant reduction in copyright registrations and patent filings for AI models. Furthermore, in instances, where previously sought IP protection no longer protect the final AI model, or broader IP protection is sought in light of the constantly evolving nature of AI industry, open source licenses may be sought to protect the IP rights associated with the AI model, instead of seeking traditional IP protection. Depending on the modifications to the AI model, namely the software code, functionality, output and algorithms, prior traditional IP protection may no longer be adequate to protect the final AI model and its associated features after open contribution has been completed. The open source license may be crafted to provide IP protection to the final AI model, such as AI output, functionality, software code, algorithms, that were excluded from the traditional IP protection of copyrights and/or patents.32 Additionally, the AI industry is constantly evolving and changing in complexities, challenges, and functionality. In light of the endless series of unknown future advancements or changes in AI technology, there exists minimal understanding and clarity on the processes that enable AI systems to work, outside of the AI developer communities. Since open contribution is largely dependent on the community of AI developers, an owner may seek open contribution. However, due to the ongoing nature of open contribution in which the code, AI output, or other aspects of the AI model may significantly change during the course of open contribution. IP protection in the form of open source licenses, rather than traditional IP protection, may be utilized as a preferred method to protect IP rights associated with the AI model. The open source license to grant broader IP protection for the AI model. Some popular standard open source licenses that are widely used, or have strong communities, include Apache License 2.0, MIT license, FreeBSD license, and Common Development and Distribution License. In addition, new open source datasets have developed to provide a large volume of contracts, including open source licenses, that may be useful for granting IP protection for the AI model, for example, Atticus Project. Many of these open source datasets include permissible license terms and free to the public.33 Some of the source licenses may be customized may include adding a custom patent license to an existing open source license, or crafting a termination clause that includes specific grounds for terminating the open source license related to the specific concerns or circumstances connected to the open source community.34 Lastly, other non-traditional forms of IP protection, such as sui generis, have been compared to open contribution as a vehicle to obtain broad IP protection to AI models. However, unlike sui generis, open contribution is considered to require less legislative involvement and possibly less economic investment, depending on the sui generis protection process enacted by the legislature.35 With open contribution, no legislative involvement is necessary to craft special legislation for IP protection associated with AI technology, namely AI models. But rather, open contribution merely requires free contribution from one or more communities of developers. Additionally, the openness provided by open contribution seems to promote competition and socially beneficial for the common good.36 In conclusion, with open contribution, the IP protection for AI models may largely depend on the developers or open source community complying with the terms of the open source license, and/or the licensor (owner of the AI model) seeking traditional IP protection, as necessary, in parallel with or prior to engaging in open contribution.37 E. Existing Sui Generis Rights In their present form, sui generis rights are generally a targeted form of protection than traditional forms of intellectual property. However, the rights are also conferred through less rigorous procedures than are often required. Sui generis rights differ in that they are not restricted to products of creativity or particular forms of expression and are instead rewarded to protect the investment and labor required to produce certain works or products. 38 The intent of the protection is to limit unauthorized extraction and re-utilization of the works in order to protect the underlying investment necessary to produce or compile the works. Examples of sui generis IP protections have included integrated circuit layouts, ship hull designs, fashion designs, databases (EU), and plant varieties. We review two examples of sui generis protections with respect to the topic of trained model rights – masked works and databases. 1. Mask Works Sui generis protection on masked works cover images that represent a pattern for the layers of a semiconductor chip. Under the Semiconductor Chip Protection Act of 1984, the associated circuit layout becomes legally protected upon registration similar to Copyright registration and is processed by the US Copyright office. If successfully registered, the owner/applicant receives exclusive rights to reproduce the mask itself, to import or distribute chips incorporating the mask, or induce others to perform the preceding. The exclusion prohibits only copying, not independent creation and allows for reverse engineering for instructive purposes and subsequent incorporation into new original works; the rights lasts 10 years. In contrast with the database rights described in the following section, mask works require some level of originality. Accordingly, while the protection of mask works is specialized to suit a specific category of invention, the qualifications for protection maintain aspects of originality or novelty associated with conventional IP rights. More specifically, a mask work must not only be original in terms of independent creation, but also must not be “commonplace, or familiar in the semiconductor industry, or variations of such designs, combined in a way that, considered as a whole, is not original.”39 In contrast, other forms of sui generis rights withdraw the requirement for originality entirely, and instead recognize the value and investment associated with a work as opposed to the creative aspects. 2. Databases (EU) Recognized in a small number of jurisdictions such as the European Union and United Kingdom, sui generis protection on databases covers collections of data arranged in a systematic way and individually accessible by electronic means. The developer or owner must “show that there has been qualitatively and/or quantitatively a substantial investment in either the obtaining, verification or presentation of the contents.”40 Once shown, the right is created automatically and does not require registration.41 The protection affords the owner the right to prevent “extraction and/or re-utilization of the whole or of a substantial part … of the contents of that database.”42 These rights last 15 years, and no registration requirement is mentioned in European Directive. Advocates for database rights posit that the legal uncertainty surrounding database and related information systems creates legal uncertainty and shifts away from dissemination toward trade secrets, thus stifling innovation.43 The critics of database rights often align with proponents of open source coding who argue that that the rights create an “unnecessary obstruction for the access to and reuse of information....”44 Additionally, critics argue that the alleged obstruction is not warranted because the value has not been shown to incentivize the database market in the EU. These arguments suggest that innovation would be better served by allowing databases without creative elements to be substantially reused as reference materials for future innovation and collaboration. It is similarly suggested that property rights are not necessary to promote the use and development of databases. Critics also warn that the term of such rights may be perpetual if “substantial investment” in the database occurs with later public disclosure. IV. BUILDING THE SUI GENERIS RIGHT FOR TRAINED AI MODELS A. Policy Considerations As discussed above, there are challenges to protecting AI models using existing IP protections. Given the inapplicability or inadequacy of these protections, a mechanism may be needed to reward AI developers or owner’s rights to the “sweat of the brow.” As explained earlier, training AI models can often times involve numerous steps from data acquisition to the iterative tweaking of the model. AI models play a crucial role in, not just providing an outcome, but are constantly assessed to correct for error and adapt for evolving data. As society places more emphasis on eliminating bias in AI systems, it is not just the data itself that requires handling but also the way the model is applied to curating the data.45 A high cost to create with a low cost to misappropriate (copy) afterward could cheapen the value of AI models or deter the advancement of newer or more robust models. It can be helpful to reconsider a few limitations of existing IP. As mentioned above, patents are limited to constructions which are novel and non-obvious. Patents are also presently entangled with the development of subject matter eligibility law, which is of considerable concern to the nature of AI models. Copyrights are not intended to protect the underlying idea, but rather the original expression of such an idea, and thus, they may miss the core mark with AI models. Trade secrets may be generally less preferred within the community.46 Yet, the potential shortcomings of the existing IP protections alone may not necessarily justify the creation of a new medium in sui generis rights. For example, many commentators predicted that patenting of software inventions would stifle independent creation. But those concerns may not have come to fruition in the digital age thus far.47 Additionally, the availability of anti-trust protections might be considered to balance the threat of over-constraining the marketplace. More specific to AI models is that the underlying premise that misappropriating another party’s AI model is simple or low cost. Indeed, implementation of a trained model for a similar, but nonidentical process, may still require considerable investment in implementation, including time, personnel, and supporting infrastructure (e.g., data storage, processing, etc.). In practice, it is unclear that such AI model misappropriation could so readily occur that would undermine the creator’s investment. Assuming misappropriation can practically occur, if adequate protection of AI models requires extensive IP protections across multiple realms (patent, copyright, and trade secret) simultaneously and with some remaining doubt about their adequacy; at what point does this patchwork approach suggest overall inadequacy? Moreover, existing IP can be zero-sum, such that establishing one right can deteriorate or eliminate the other, for example, in comparing the disclosure requirement of patents with the secrecy requirement of trade secrets. Although not necessarily determinative, the potential for unintended impacts of sui generis rights on existing IP protections should not be ignored. A balancing of factors may better reveal the incentives and disincentives for such sui generisrights. For example, a significant disproportion in the investments of different parties may be informative. Hypothetically, if an originating party of an AI model attributed a significantly greater level of investment to develop the original AI model than another party as a potential misappropriator of the original AI model, then a comparative framework exists to consider the basis of the sui generis right, and the associated policy both for and against such a right. For the time being, it seems that the concerns which merit consideration in the creation of sui generis rights for AI models stem from the particularly high investment costs for development. Accordingly, comparative demonstration of the low barriers for misappropriation may favor such a new right. For example, significantly low resource requirements, low cost, few personnel requirements with low level of technical expertise may indicate disparate investment levels sufficient to justify such a new right. Consideration of investment disparity might likewise provide a benchmark for either or both the establishment of rights and/or enforcement of established rights. Of course, one counterargument to sui generis rights in general is that the existing law is itself quite flexible. Often, in the evolution of humanity, an old rule (whether statutory or judicial) is applied in a modified way to address shortcomings in facing new instances. For example, before the computer age, the U.S. considered adopting sui generis rights for computer programs.48 However, Congress ultimately adopted a framework clearly including computer programs to be copyrightable as literary works under the Copyright Act of 1976.49 Among other justifications, advisors cautioned against the economic impact of excluding computer programs from copyright protections wherein the programs themselves are “expensive to make and cheap to copy,” echoing the intangible inequity of the issue.50 Yet, “[i]t has proven difficult to apply traditional copyright doctrine derived from cases involving novels, dramatic plays, and fabric designs to protect nonliteral aspects of computer programs, which has caused courts to develop special tests for software copyright infringement.”51 Commentators have observed U.S. courts to essentially create a sui generis sub-form of copyright protection to fit computer programs into copyright law.52 In one practical instance of data protection itself, U.S. case law has denied compensation to sports leagues for data collected from their events.53 Historically, raw data has not been protected under copyright law and third-parties cannot be prevented from collecting event data from live or recorded broadcasts, which could then be commercially exploited.54 However, as entities contract with data companies to collect data from their events, the problem of data reutilization evolves.55 In the United Kingdom, the landmark decision of Football Dataco Ltd v. Sportradar GmbH found that a data company’s sui generis database right can be violated by third-party publication if the data is real-time, processed sports data rather than merely fixed and easily obtainable characteristics because it requires “considerable skill, effort, discretion, and/or intellectual input.”56 In comparison, the sui generis database protections afforded in Dataco would appear to have required significant manipulation of traditional copyright protections to distinguish real-time data from other so called "fixture lists" by measures other than the effort invested. Consequently, the level of investment remains a guidepost whether in rethinking existing law or analyzing a sui generis form of protection. Nevertheless, new forms of IP face intense obstacles while patchwork approaches can be costly and require time to develop case law, if at all. With the evolution of technology and business practice in general, IP systems are incentivized to periodically evaluate the role of sui generis rights and to determine whether a new approach or a revitalized old approach is appropriate. This is true even if the outcome merely yields consideration of the issues in existing protections, further defining and evolving the nature of the IP system. B. Subject Matter Covered The scope and qualifications required to qualify for protection may be broadly considered in relation to content, investment, and originality. Concerning content, the scope of the subject matter afforded protection may be broadly considered. The subject matter can include any mathematical terms embodied as computer data for use in a mathematical function derived automatically by an ML method, as well as any mathematical function or algorithm embodied as computer instructions incorporating such mathematical terms. Accordingly, the considerations for the protection of trained models may, in some ways, be similar to those afforded to software under copyright law. However, as proponents for intellectual property protection for software have argued, copyright protection “only prevents another from copying the particular expression used by the copyright holder in conveying the underlying idea; the underlying idea itself remains unprotected and free for public use.”57 The question then becomes whether the protection of a particular form of expression is adequate to protect works generated by ML and AI. For example, should the expression of a computer-generated model or algorithm be the focus of protection or the underlying operation and resulting capability? The apparent deficiencies in copyright protection lead to one of the more contentious issues - originality. To avoid this issue, one option is to remove it entirely as in the European model for database protection. Another option is to require originality similar to mask works, which must not only be original in terms of independent creation, but also must not be “designs that are staple, commonplace, or familiar in the semiconductor industry, or variations of such designs, combined in a way that, considered as a whole, is not original.”58 However, variations in mask works may be visibly discernable and supported by a list of distinguishing components or aspects. Such variations that are typically discernible may not be effective if considered in relation to trained models because the differences in trained models may not even be readily distinguishable to their developers. This lack of understanding is related to the primary benefit of trained models - the developers only enabled the system to create the model and do not program the solution as in conventional software development. As discussed previously, originality in copyright protection requires that the work possess “at least some minimal degree of creativity.”59 As such, an inability to argue a modicum of creativity in AI models makes it nearly impossible to determine and distinguish originality, making an originality consideration similar to mask works unworkable. Another qualification for trained model rights may be related to eligibility. If intended to avoid the complexities and challenges that prevent patent protection from effectively covering trained models (e.g., inventorship and subject matter), the eligibility requirements for trained models may be relaxed. However, the subject matter covered may at least need to be distinguished from software to avoid abuse. One possibility would be distinguishing trained models from software by requiring applicants to demonstrate that a model was derived automatically from an ML method or the product of a computer. Such a requirement could be enforced through a requirement of documentation. The documentation requirements may provide a framework to distinguish trained models from software written or generated by human influence. Such a documentation can also provide some level of explainability to AI.60 Even so, distinguishing the extent of human intervention necessary to preclude rights, or alternatively the level of autonomy of a computer, may be challenging. Subject matter qualifications may also extend to investment similar to sui generis database protection (a “significant investment” model). That is, developers of trained models may need to demonstrate a substantial investment in either training or enabling the generation of a trained model. Such a requirement may be utilized as an alternative to a significant demonstration of originality. For example, rights could be afforded to developers who can demonstrate independent creation, which may be substantiated by documenting investment and labor related to supplying training data and/or equipment to process the trained data and optimize the coefficients of a model. In this way developers would have to be able to show that there was, qualitatively and/or quantitatively, a substantial investment in either the obtaining or verifying the model or its coefficients. A qualitative showing could include the size of the investment in training the model. Efforts in procuring and curating data to train the model could be factored into the process of training.

The establishment of a sui generis system for the protection of AI-generated works underscores a proactive and pragmatic approach to accommodate the rapid pace of technological evolution and shifting creative paradigms. An integral aspect of such a system would involve regular and systematic reviews, a practice that mirrors existing frameworks and has proven effective in maintaining legal relevance and addressing emerging challenges. As a starting point for discussion, a 5-year review period seems reasonable. The concept of periodic reviews is not uncharted territory within the legal landscape. A precedent is evident in the Copyright Office's periodic reviews conducted under Section 512. This process assesses the effectiveness of the Digital Millennium Copyright Act (DMCA) in addressing online Electronic copy available at: https://ssrn.com/abstract=4557004 infringement issues. The periodic review model acknowledges the fluidity of the digital age and serves as a valuable reference for crafting similar review mechanisms within the sui generis framework. In navigating this uncharted terrain, the sui generis approach proposed here emerges as a beacon of adaptability. It acknowledges AI as a dynamic co-creator while upholding the rights and intentions of human participants. By fostering a balanced ecosystem, this approach empowers creators to leverage AI's capabilities, while ensuring that the outputs remain true to their vision. As legal scholars and policymakers grapple with the intricacies of protecting AI-generated creations, the sui generis approach emerges as a formidable solution. Its adaptation allows for the preservation of the inherent value of both AI and human creative input yet recognizes and addresses the inherent imbalance between humans and AI in the creative process. By embracing this nuanced approach, society can fully embrace the artistic possibilities of AI while safeguarding the foundations of creativity, innovation, and the public domain. In a world where human imagination and artificial intelligence converge, the sui generis system offers the promise of harmony amidst complexity, embodying the true essence of creative evolution.

When delegates arrive in Geneva today for the long delayed World Trade Organization (WTO) summit, they will find an institution in the middle of an existential crisis. For 18 long months, the WTO has been debating a moderate proposal from South Africa and India which would have allowed countries to temporarily override the property rights of pharmaceutical corporations so they could produce patented Covid-19 vaccines. Opposition by Britain, Switzerland and other European countries has kept it from progressing. Even a global pandemic, it seems, isn’t sufficient cause to prompt a temporary rethink of the WTO’s pro-business approach. Just as bad, the WTO cannot agree a common approach to the food crisis fast spreading across the globe, or the invasion of one of its members by another, or, most serious of all, the climate catastrophe facing humanity. All it can do is fall back on the mantra of more free trade. Unable to break with a “market knows best” ideology which is actively exacerbating the world’s problems, the WTO is now a failed institution. It’s time to bury it. The WTO’s crisis is part and parcel of the greater crisis of free-market globalisation as a whole. It was formed in the mid-1990s, the high point of free-market capitalism, when the answer to every problem was more markets, more private sector, less government red tape. There was, we were told, simply no alternative. Summing up the feeling a few years later, Tony Blair told the Labour party conference, “I hear people say we have to stop and debate globalisation. You might as well debate whether autumn should follow summer.” But it hadn’t always been that way. Before the WTO, there had been a much looser set of international trade rules created at the end of the second world war. This system was also based on the idea that trade was a good thing and high tariffs were usually a bad idea. But to the extent that free-trade policies didn’t achieve those goals, countries had a fair bit of freedom to ignore them. There was no enforcement mechanism, and plenty of room for developing countries, in particular, to design policies they thought worked best for their own development. The WTO changed all that. Radical free trade became an end in itself. WTO treaties laid the basis for a set of hard global economic rules, embedded in international law. At their core, they removed power from the state to interfere with the supposed rights of big business and big finance, undermining the ability of governments to protect their farmers and infant industries, and to regulate big finance and big business. Unlike its predecessor, the WTO incorporated a dispute system with real teeth which made the whole system enforceable. We can see the results all around us. Far from creating a global version of a farmers’ market, it has encouraged the growth of agricultural monopolies practising intensive farming, which dominate our food system at huge cost to the environment. It has encouraged manufacturers to “offshore” production to wherever labour is cheapest and regulations are lowest, creating political resentment and fuelling rightwing populism in the west It has created staggeringly complex and vulnerable supply chains, best symbolised by an oversized container ship becoming stuck in the Suez canal and bringing global trade to a halt. And it has allowed a handful of multinational corporations to dictate the world’s ability to produce the vaccines needed to end a global pandemic, because they owned the – often publicly funded – research behind those medicines. None of this should surprise us. The great theorist of capitalism, Karl Polanyi, warned us 75 years ago that trying to turn the whole world into a gigantic marketplace would end in the “demolition of society”. The fascism of the 1930s was the nightmare reaction that Polanyi lived through, but we see our own version of that social breakdown today. This system cannot cope well with crisis. But crises are now a defining feature of our era. Which is why many governments, including Joe Biden’s administration in the US, are moving gradually away from the WTO’s free-trade system. It’s a process initiated, perhaps unwittingly, by Donald Trump. Seeing which way the wind was blowing, Biden has continued in this direction – with less bullying and bluster – eschewing free trade deals and standing up to the corporate monopolies that globalisation has created. This is to be welcomed. But we’ll need to go further and faster if we don’t want to replace a global free-market system with one of nationalist competition in which the strongest player wins. We need a fundamentally different global trade system, which helps rather than hinders government action in protecting their people and the planet. The WTO cannot play that role. It is not reformable, as the Covid waiver debacle proves. Developing states have always been especially constrained by this system. If we want change, these countries will have to work together to begin creating change on the ground: building up their own industries, supporting their small farmers, regulating and taxing big business and big finance, and using the proceeds to build up public services to remove people’s basic needs from the market. There are hopeful signs – from South Africa’s creation of “open source” medical research to Indian farmers successfully refusing to be further liberalised. The globalisation project cut short the developing world’s experiment with economic decolonisation. It’s time to return to it. Any attempt by the US and Europe to punish these countries will have to be met with resistance, there and here. From this point, we can begin rebuilding an international architecture for actual fair and free trade. None of the problems we face will be solved by the WTO’s “market knows best” proposals, which will only accelerate social and environmental breakdown. It would be no loss to the vast majority of the world if this WTO summit was the last. It’s long past time for something else.

In India, Prime Minister Narendra Modi is calling for economic self-reliance and raising trade barriers. China has adopted a dual circulation strategy which puts increased emphasis on domestic demand and relegates trade to a supporting role. The buzzword du jour within the EU is strategic autonomy, which recognizes the need to strengthen domestic production capabilities and reduce over-dependence on trade. And in the US, the Biden administration is pursuing a worker-centric trade policy and has maintained most of the trade restrictions put in place by the Trump administration. Skepticism about trade has been further intensified by the coronavirus pandemic. Countries and companies are now aggressively looking to reduce trade risk through more localized production and less trade. Given these sobering realities, one might have reasonably expected to see a sharp decline in trade. But in defiance of most projections, trade has held up surprisingly well. The trade collapse that never happened Given the fallout from the pandemic and rising protectionism, the WTO estimated that global trade could plummet by as much as 32% in 2020.[1] Many independent analysts felt the assessment was optimistic. Yet, the actual decline was roughly 5%.[2] Given the broader macroeconomic environment, in which overall demand was dampened by a 3.5% reduction in global GDP, trade remained buoyant.[3] Of course, acute specific circumstances produced eye-catching trade reductions for individual countries. German exports were down 9.3%, Philippine exports were down 16.3%, and Japanese exports slumped to their lowest level in 11 years.[4] While these apparent doomsday scenarios for trade dominated the headlines, they masked the real story. Despite the considerable headwinds, trade took a beating but kept on ticking. What explains the surprising resilience of trade? Trade barriers often redirect rather than thwart trade flows Bilateral tariff and non-tariff barriers have proliferated. These restrictions can sometimes be quite effective in accomplishing their primary objective: to diminish imports from the targeted country. But in many cases, their overall effect is to redirect rather than block trade. For instance, China has significantly increased trade restrictions on a variety of imports from Australia. Although this initially created painful disruptions, Australian exporters were able to adjust. This is a testament both to the resiliency of Australian companies, as well as the enduring appeal of trade. Today, Australian cotton exports that are no longer welcome in China have found receptive markets throughout ASEAN. Barley shipments that previously went to China are now being unloaded in Saudi Arabia.[5] On the other side of the bilateral trade equation, barley importers in China have not been left in the lurch by the loss of Australian supply. Argentina has stepped in to fill the gap. Argentine barley exports to China have grown so significantly, Argentine farmers now plan to expand capacity by 28% to meet the new demand.[6] And China’s ban on imported coal from Australia has not stemmed coal imports. US coal exports to China have risen to their highest levels since 2013 and are straining the capacity of US companies to keep up with demand.[7] The bottom line? Trade restrictions which close one door frequently open another. Trade on an overall basis continues to chug along. Two important caveats are worth making, however. Although trade continues to flow, it will frequently be less efficient. Importers are being compelled to fill their orders from suppliers they might not have selected absent the trade restrictions. And exporters do not always find new markets quickly and seamlessly. Australian wine exporters, for instance, are still struggling to replace sales to China. New trade patterns necessitated by barriers could also produce an increased balkanization in trade relationships and a fraying of truly global trade ties. Restrictions are now being put in place not merely for trade-related considerations. Geopolitical calculations, human rights issues, and environmental concerns are increasingly factored into trade policy. This could drive the formation of formal or informal trade blocs among partners who share common perspectives across the ideological, social, and geopolitical spectrum, while trade declines with less simpatico partners. At this point, however, it is entirely unclear if such a scenario will materialize to a meaningful extent. There are numerous examples of trading carrying on despite rising tension and differences in ideology. Trade battles rage on – but so does trade The US-China trade war has not abated under the Biden administration. In some respects, it has grown more complex. President Biden is casting the bilateral economic relationship within the framework of a broader ideological struggle between democracy and authoritarianism. Businesses have not, however, taken heed. With the exception of especially sensitive technology sectors where trade has essentially been prohibited, trade between the US and China has not missed a beat. China’s exports to the US have grown by double digits year on year every month this year.[8] Even in comparison with 2019 (before the pandemic kicked in), China’s exports are rising significantly. The punitive tariffs have essentially become a manageable business irritant rather than a brick wall. The US is not the only example of trade flourishing despite tense relations. The EU has imposed measures on China and been sharply critical of labor conditions in Xinjiang. It has even sidelined consideration of its Comprehensive Agreement on Investment (CAI) with China. But despite the actions and strong words, the EU’s imports from Xinjiang more than doubled in first half of 2020.[9] China has also shown that it can engage in a war of words on one hand, while carrying on trade with the other. Australia has faced a barrage of vitriolic denunciations from China over its positions on issues ranging from the South China Sea to the origins of Covid. China’s umbrage appears to be limited, however, by economic reality. Australian iron ore exports to China, whichcannot easily be sourced from elsewhere, were down by only 2% in 2020 from the record-breaking year of 2019. The imperative to trade remains compelling Despite a recalibration in attitudes toward trade, tempestuous relationships, and a rise in protectionism, reports of trade’s demise have been greatly overblown. Governments may bluster, and even enact restrictions, but companies and consumers carry on. The greatest efficiencies will rarely be found entirely within the border of any single country and the imperative of cross-border commerce will continue to be formidable.

It’s been a rocky few weeks for U.S. trade partners. On June 1, the United States imposed trade barriers on steel and aluminum imports from the E.U., Canada and Mexico, after the exemptions that the U.S. granted earlier ran out. The U.S. justified these measures through a rarely used national security exception built into World Trade Organization (WTO) rules. These longtime allies then countered by targeting politically sensitive U.S. exports, from bourbon and Harley-Davidsons to peanut butter. Here’s the unexpected twist: The U.S. measures may have been unwarranted, but they are likely legal. The countermeasures by U.S. trade partners, on the other hand, are undoubtedly justified, but they are most likely illegal. The U.S. just broke a big trade taboo By turning to the national security exception, the U.S. has broken an abiding taboo in trade relations — for the 70 years the WTO and its predecessor regime, the General Agreement on Tariffs and Trade, have been in place, member countries have invoked the national security exception just six times. It has been formally used as a defense in legal proceedings only once. And the WTO has never ruled on its usage by a member state. [Trump has announced massive aluminum and steel tariffs. Here are 5 things you need to know.] In 1947, GATT negotiators included a national security exception in the form of Article XXI as a last-resort option in the event that a country’s trade commitments threatened its national security. Countries have always understood that security interests come first. None other than Adam Smith liked to remind everyone that “defence … is of much more importance than opulence.” But the result is that Article XXI is what John Jackson, the great scholar of the GATT, called a “catch-all clause” that is “so broad, self-judging, and ambiguous that it obviously can be abused.” For 70 years, countries avoided the security exception The world’s trading regime avoided this abuse for decades because countries agreed that the security exception was too risky to call upon, precisely because it was so unconstrained. That just changed. By invoking this exception in peacetime, using it as the reason to impose new tariffs on U.S. allies, and referencing reasons unrelated to security — like slow progress in the NAFTA renegotiations — the U.S. has completely broken with its own practice. [What is NAFTA, and what would happen to U.S. trade without it?] Past U.S. administrations steered clear of using the exception, even when it might have made sense to do so. Here’s an example: In 1984, when Nicaragua first challenged the legality of U.S. sanctions on the Sandinista regime, the U.S. opted to be found in violation of GATT rules, rather than set a dangerous precedent by invoking the security exception. This is a new and dangerous precedent Until now, countries understood that to open that door could be the end of the global trade regime. As U.S. House Democrats put it in a 2006 letter to the U.S. Trade Representative: “If the U.S. […] for any reason that it deems ‘necessary to its essential security interests’ can invoke a self-defining ‘essential security’ exception, what is to prevent other countries from using this exception to block U.S. exports or other U.S. rights such as enforcement of intellectual property rights without ample justification?” The answer is, not much — and that’s why the U.S. move creates such a dangerous precedent. The first to exploit that precedent may be Russia, which has invoked the same national security exception in an ongoing WTO dispute against Ukraine. Remarkably, in its own submission in that dispute, the U.S. took Russia’s side and stuck to its position that the national security exception is “self-judging,” “non-justiciable,” and “unreviewable” by any WTO panel, effectively giving Russia a pass. So which measures are legal, and which are not? Canadian Foreign Affairs Minister Chrystia Freeland called the U.S. move illegal, as did French President Emmanuel Macron. It’s impossible to know for certain how a WTO panel might handle this question, precisely because no panel has ever ruled on the issue. But the drafting history of the security exception actually offers the U.S. position a lot of support. In this view, the clause really is “self-judging”: a country’s security interest is whatever that country says it is. From a practical standpoint, it is also unlikely that a WTO panel would be bold enough to challenge the U.S. definition of its own security interests, especially at a time when the U.S. is exerting increasing pressure on the WTO’s dispute settlement mechanism by blocking the appointment of WTO judges. What about U.S. trade partners’ reactions? The E.U., Canada and Mexico have announced immediate retaliation. Here’s the wrinkle: Retaliation in the WTO is only legal if formally authorized by the WTO, after countries engage in the dispute settlement process, which includes arbitration over compliance and the amount of retaliation. Countries can retaliate legally only if a party is in continued violation after a ruling. This process has never taken less than two years. [Does Trump want a trade war? What you need to know about Smoot-Hawley tariffs and the 1930s.] In skipping this process, the E.U. has attempted to justify its moves by declaring that the U.S. tariffs are actually a safeguard measure, a type of trade remedy designed to help countries deal with unforeseen import surges — this means countermeasures could be allowed. Although legally ingenious, the E.U. stance is unlikely to hold water: The U.S. has explicitly invoked the national security exception — Section 232 of U.S. trade law. Safeguards fall under Section 201. All of which means the retaliation by U.S. trade partners is most likely illegal. U.S. trading partners are taking the law into their own hands. What’s the bottom line here? The U.S. move has prompted its trade partners to respond in an extralegal fashion. And that, from the standpoint of the Trump administration, which is broadly skeptical of the multilateral trade regime, may be its greatest tactical victory. By forcing the hand of its trade partners, the Trump administration has shifted the exchange onto a field where there are few rules to constrain state behavior. The last time states played out their commercial grievances on that field was in 1930, when the Smoot-Hawley Tariff Act raised tariffs on hundreds of U.S. imports. That move didn’t end well.

In response to the ruling, a spokesman for the Office of the U.S. Trade Representative predictably reiterated that the WTO “has no authority” to review national security issues and announced: “We do not intend to remove the Section 232 duties as a result of these disputes.” The spokesman also hinted the United States will appeal the ruling into the legal abyss of the absence of a sitting Appellate Body—which would have the effect of denying to the prevailing countries their treaty right of lawfully applying economic sanctions against the United States, should the U.S. choose not to comply with the panel rulings and recommendations. The U.S. view that the WTO has no legal authority to judge whether an action purportedly taken for national security reasons is truly a national security action, is, as the WTO panel ruled, patently incorrect under international trade law. If the United States were correct about this, then why are specific circumstances set out in the WTO treaty that spell out when measures that would otherwise be inconsistent with WTO obligations can be excused for national security reasons? The United States agreed to the limits set out in the exception in 1947, and these limits have been in the trade rules—to which the United States agreed—ever since. Twice in recent years WTO panels have affirmed these limits when the national security exception has been invoked. As I have explained and explored at length in a new Cato Institute policy analysis, acceptance by WTO jurists of the contemporary and bipartisan U.S. view that the WTO has no jurisdiction on such matters would open up a black hole of national security, in which professed national security measures of all kinds could swallow up the entirety of all WTO obligations—something the United States itself warned against when the national security exception was written at the creation of the multilateral trading system in the immediate aftermath of the Second World War.

The world has had a long history of trade wars with only one real conclusion—no one wins. In US history, the Smoot-Hawley Tariff Act of 1930 was the most egregious, with tariffs on over 20,000 goods. Enacted by Republicans through a Senate and House controlled by the same party, the Bill was signed into law by a reluctant president, Herbert Hoover. It is widely believed to have led to a decline in sentiment during the Great Depression, resulted in retaliatory action and a massive global trade decline in subsequent years, and (some believe) was a contributory factor to World War II. The case for global trade, made eloquently after World War II and codified first under the General Agreement on Tariffs and Trade (Gatt) and then under the World Trade Organization (WTO), is that interdependence increases prosperity and discourages martial conflict. In that sense, global trade and nuclear deterrence are the world’s best bets against a full-scale global conflict. The WTO concerns itself with the trade in goods, services and intellectual property between member nations (123 at last count). It provides a framework for dispute resolution and is designed as a referee in enforcing member-country adherence to agreements. While Trump has cleverly used a provision within WTO rules for national security to push through his tariff, it will almost definitely be challenged by China and/or Brazil in the WTO. By that time, it is likely that the tariff will have been diluted and/or withdrawn, with Trump having triumphantly demonstrated to his narrow political base that he has followed through on his campaign promise. The problem with that reckless demonstration is that it sets a precedent for other countries and is tantamount to allowing open season on a rules-based order that has generally served to increase prosperity around the world over the last 70 years. India will not be affected much by the tariff since it is not a major steel exporter to the US, even though in recent years, metal exports to the US have been on a high growth path off a very low base. Taken as a one-off idea, India can afford to take the back seat on this tariff. However, Trump’s metal tariff highlighted only a tax on goods. A trade war on services, primarily through a rhetoric on immigration quotas for skilled workers, has already been going on. That constant jawboning has already resulted in several large information technology (IT) service companies announcing plans for near-shoring from within the US and hiring a large number of staff there. Infosys, Tata Consultancy Services, Accenture and Cognizant have announced that they will hire tens of thousands of engineers in the US. On margin, the ratio of hires in the US to those hired (more traditionally) in India could well end up being north of 50%. This would be a radical change from recent operating procedure and will have an impact on the price of the service to the customer and the bottom line of these firms. It erases decades of progress made towards a more flexible and global labour market arrangement for IT services firms and directly conflicts with the spirit of the WTO. It also weakens global cooperation under the WTO and could potentially lead to a breakdown in that institution. Free trade may occasionally create a problem related to inconsistent distribution of prosperity—that is, create winners and losers. Rather than scrap the notion of free trade, it is better to alleviate the issues of those relatively disadvantaged. In a back-door tariff, India had itself set minimum import prices on steel last year (Japan and Taiwan are in WTO consultations) and must phase these out if we are to have any credibility in protesting protectionism. India can and should take the high road against protectionism. It needs global trade, human-capital-based exports and financial capital import to buttress its path to prosperity. In its absence, India will have to make do internally on all three. India’s capacity for interstate trade, the opportunity for large-scale internal labour mobility of skilled and unskilled workers, and its sourcing of internal capital will not be sufficient to deliver prosperity to millions of households in one or two generations. India is a disproportionate loser in a protectionist world and must take a very strong and principled stand against it. The detour towards protectionism will eventually amount to naught, as evidenced by history, but the time taken will cost India dear. To mix some metaphors, India must go to war to support the continuation of a rules-based order for goods and services trade

Abstract The recent “US–China trade war” has aroused concern over trade-related environmental impacts. This study built a multiregional computable general equilibrium model to simulate environmental impacts of the “US–China trade war” under different scenarios of tariff and nontariff barriers and the battlefield spreading ranges. The present study found that although the trade war will cause a global economic downturn, which will seemingly reduce environmental pressure globally, global carbon emissions are expected to increase rather than decline. On the one hand, the CO2 emission increase caused by land-use changes in Brazil and Argentina will far exceed the emission reduction because of decreased global production. On the other hand, some countries/economies especially those developing countries such as Vietnam, Russia, and India will face emission increases driven by scale effects. Countries such as Korea, the UK, and France will enjoy a reduction in emissions driven by structural effects. China and the US will face a reduction in production and CO2 emissions, but their CO2 emission intensities will rise. The results remind us that as global production and supply chains are formed, it is important to closely monitor trade-related environmental impacts. Efforts should be made to balance the interests of trade and the environment. This publication is licensed under the terms of your institutional subscription. Request reuse permissions. 1. Introduction ARTICLE SECTIONS The US and China are the world’s two largest economies, and the trade volume between the two countries accounts for 23% of world trade. Especially after China’s accession to the WTO in 2001, bilateral trade between China and the US has developed more rapidly. The value of their bilateral trade in goods rose from $80.6 billion in 2001 to $584.8 billion in 2017. (1) The rapid and stable growth of US–China trade is related to the considerable differences in their production factors and the strong complementarity in their industrial competitiveness. (2) However, trade friction has always existed, especially in recent years, the trade friction between the two countries escalated into a “trade war” (see S1, Supporting Information for a timeline of the US–China trade war). The battlefield of the trade war has continued to expand to Mexico, the European Union, and even the world. The trade war has shocked the trade and economic paradigm of the two countries and even the world and increased the uncertainty of global economic development. (3−16) Although many believe that the US–China trade war will depress global economic production and consumption, reduce related emissions, and benefit the global environment at large, (17−19) the prospective outcome could be complicated and uncertain. The policy shock imposed on trade could lead to changes in production volume and structure and then change the energy and resource use of one country and the world and result in changes in the global CO2 and pollutant emissions and environmental quality. (20−35) The most direct and obvious impact in the short term is Brazil’s environment. The increased soybean production in Brazil caused by Chinese increased soybean import demand may exacerbate deforestation in Brazil, destroy biomes, and release large amounts of carbon. (7) Lin, et al., (17) Liu, et al. (18) and Du et al. (19) used different computable general equilibrium (CGE) models and set tariff reduction scenarios to explore the environmental impact of the US–China trade war and found the trade restrictions brought by the trade war will reduce global carbon and pollutant emissions due to the economic losses. However, it may be not conducive to the development of clean energy in less-developed regions. Considering the complexity of global environmental impacts of international trade, it is of great interest to explore the impacts of the US–China trade war on the changes in global production (supply chain) and then on the total emissions and their spatial distributions. However, the current limited studies on the environmental impacts of the US–China trade war only focused on the impacts of either fossil fuel energy use (and sometimes industrial process) emission or deforestation causing emissions separately, without a combination of the two aspects. Besides, in the policy shock scenario settings of most of the previous trade-economy or trade-environment impact research literatures, only tariff barriers were taken into consideration but nontariff barriers (NTBs) were neglected. Currently, the global tariff level has been generally low, and the NTBs have stood out as important impeding factors on the international trade. (36−39) In addition, during the trade war, China and the US had interactions with other economies, for example, the US signed trade agreements with Mexico, Canada, and Japan but intensified trade conflicts with the EU, and so forth. These issues are also among the influencing factors to the economic-environmental outcome of the trade war. Thus, this study comprehensively considers the potential emissions from fossil fuel combustion, industrial processes, and land use changes and adds NTBs and the multilateral trade practices to the trade war scenario settings, which makes this research more comprehensive and different from the existing studies. Based on a chain reaction assessment framework of “trade policy/measures → trade and economic impacts → environmental impacts” to carry out analysis, (40) this study built a multiregional CGE model and set up different trade war scenarios to simulate and explore the impacts caused by the US–China trade war on the global environment. 2. Methods and Data 2.1. Model Framework 2.1.1. GTAP Model The main body of the chain reaction method in this study was a classic multiregional and multisector comparative static CGE model, GTAP, which was developed by the Global Trade Analysis Project at the Purdue University, (41) whose database has been updated to version 10. (42) Like other classical CGE models, under the assumption of perfect competition and constant returns to scale, the GTAP model uses utility- and profit-maximizing behavior of consumers and producers as the optimal conditions. (43) For the producer side, firms combine the intermediate inputs and the primary factors to produce output. The production structure is based on a three-level nested constant elasticity of the substitution function. The top level nest is a composite of intermediate demand and value added. The second level nests decompose the intermediate demand into demands for various individual intermediate goods and value added into demands for various individual factors, respectively. The third nest decomposes the demand for individual intermediate goods into domestic and imported composites. This study adopted the standard assumption in GTAP where there is no substitution between the intermediate goods and input factors (represented by value added) and between different intermediate goods. In this model, household and government consumers maximize utility subject to budget constraints. The utility function for private spending is based on the constant differences of elasticities implicitly additive expenditure (CDE) function. The utility function for government expenditure is based on a Cobb–Douglas structure, where the spending shares are fixed. Household and governmental savings are subsequently translated into investment. Households, governments, and firms choose to buy a combination of domestic goods and imported goods following Armington assumption, with full consideration of prices and associated transport costs, export taxes, and tariffs. Countries/economies and regions are linked together through multilateral trade. Households bring together all factor incomes, and the government brings all fiscal revenues from all tax incomes, which are distributed over the three forms of final demand: private household expenditures, government expenditures, and household and government savings, so as to maximize the constructed regional utility. This study employed the full GTAP 10 database, which describes the world economy for 2014 and distinguishes 65 sectors in each of the 141 countries/economies. In the present study, the original database of GTAP 10 was aggregated into 23 regions, each with 24 economic sectors, with an emphasis on the main trade partners of China and the US (see Tables S1 and S2, Supporting Information). 2.1.2. Policy Shocks Although the current GTAP-E model has the energy and CO2 emission data, it lacks some key environmental and nature resource information, such as the local pollutant information and carbon emissions from the LULUCF (land use, land use change, and forestry). This study establishes a soft link between trade and economic behavior and the environmental impacts with reference to Song et al. (32) Policy shocks of tariff and NTBs were introduced into the GTAP model to obtain the economic impact αr,i,j, which is the change rate of economic variable j (such as output, import and export volume, etc.) in sector i of region r. The absolute impact can be calculated by Zr,i,j = αr,i,j × Vr,i,j, where Zr,i,j is r country/economy’s value or quantity change in economic variable j in sector i and Vr,i,j is r country/economy’s status quo value or quantity of economic variable j in sector i. Finally, the change in the emissions of pollutant k in sector i of country/economy r caused by the policy shock was calculated by Pr,i,k = er,i,k × Zr,i,output, where er,i,k is the emission coefficient of pollutant k in sector i of r country/economy and Zr,i,output is country/economy r’ s output variation in sector i. 2.2. Trade Measure Scenarios The baseline scenario is an equilibrium without policy shocks, which was the original balance in GTAP. Three trade policy shock scenarios with tariffs and NTBs were set that depended on the depth and scope of the US–China trade war (see S3, Supporting Information): Scenario 1: The trade war involves the US and China only. Policy shocks for this scenario were set based on tariffs mutually imposed after September 24, 2018. (44,45) In this scenario, the US imposes an extra 25% tariffs on $50 billion and extra 10% tariffs on $200 billion of goods imports from China, and China imposes extra 25% retaliatory tariffs on $50 billion and extra 5 and 10% retaliatory tariffs on $60 billion of products imported from the US. Both sides impose a certain degree of NTBs on each other. Scenario 2: Policy shocks for this scenario were based on the situation after May 13, 2019, before the 13th round of China and US trade negotiations. (45,46) This scenario also involves new trade agreements between the US and other countries. The US imposes an extra 25% tariffs on $250 billion of imports from China, and China imposes extra 25% tariffs on $50 billion of imports from the US and another four tariff rates, that is, 5, 10, 15, and 25% on $60 billion of imports from the US. At the same time, the new trade agreement among the US, Canada, and Mexico, that is, the USMCA, and the trade agreement between the US and Japan were taken into consideration. Scenario 3: The trade war becomes more intense and expands to other economies. It is assumed that the US and China impose extra 25% tariffs and more NTBs on all traded goods from each other. In addition, this scenario also assumes that the US imposes extra 5% tariffs on all imports from Mexico, extra 10% tariffs on EU aircrafts, and extra 25% tariffs on EU agricultural products. (47) 2.3. Effect Decomposition Decomposition analysis is an important tool for understanding and quantifying the driving forces behind changes in related indicators. (48) Index decomposition analysis (IDA) is widely used in driving force decompositions for carbon emission changes. (49−53) The logarithmic mean Divisia index (LMDI) (54) is a common arithmetic method for solving IDA problems. In this study, we use the IDA-LMDI method to decompose the driving forces of trade war-induced changes in CO2 emissions from fossil fuel combustion and production processes in specific countries/economies. For the total CO2 emissions Q, the IDA equation in the period dimension can be written as follows (1) where Xt is the total output in period t, which denotes the scale effect; Sit is sector i’s output share in the total output in period t, which represents the structural effect; and Eit is sector i’s CO2 emission intensity per unit output in period t, which represents the technical effect. This study adopted the standard assumption in GTAP where there is no scope for substitution between intermediate inputs and does not allow energy substitution within a sector, meaning that CO2 emissions per unit of the sectoral output in a specific country/economy in the short term are constant or that the short-term technical effect is zero. The change in the total CO2 emissions in period t relative to period 0 can be expressed as follows (2) According to Ang, (54) the LMDI decomposition results for each factor are as follows (3) (4) (5) where ΔQscl is CO2 emission change from the scale effect; ΔQstr is CO2 emission change from the structural effect; and ΔQtech is CO2 emission change from the technical effect. 2.4. Calculation of CO2 Emissions from Land Use Change Except for the CO2 emissions from fossil fuel combustion and industrial processes, the change in land use was also a major source of carbon emissions. This was especially true for countries such as Brazil, where approximately 58% of the Brazilian territory is covered by natural and planted forests, (55) making it an important carbon sink. The greatest carbon emissions caused by land use changes are attributed to the conversion between native vegetation and cropland. (56−58) So we focused on land use changes caused by agriculture production. When focusing on land use changes brought by the US–China trade war, the most substantially affected agricultural product is soybean, which accounts for around 70% of China’s imports of the US agricultural products in 2017. Brazil and Argentina are the second and third largest soybean exporters to China following the US and are the major soybean export beneficiaries from the US–China trade war. Forests in Amazon and Cerrado area in Brazil and Chaco forest in Argentina have suffered a lot from deforestation because of soybean expansion. (59−61) These forests have extremely high carbon content and rich biodiversity, playing a significant role in global carbon balance. (62−65) Studies have shown that changes in the structure of soybean trade caused by the trade war can affect soybean plantations in Brazil (3,7) and Argentina, (8,66,67) if China withdrew soybean orders from the US and turned to Brazil and Argentina to fill in the supply gap, and it is very likely to increase deforestation and carbon emissions. Though there might be the possibility for Brazil and Argentina to increase soybean exports in other ways such as importing soybeans from the US for re-export to China instead of increasing soybean production in Brazil and Argentina, however this could not be the case due to the fact that, first, the tariffs are supposed to be imposed on the US-produced product according to the principle of the origin of goods, (44,46,68) which means that the re-exported US-produced soybean must be also subject to tariff. Second, when checking the US and Brazil custom statistics, it was found that before and after the implementation of China’s retaliation trade measures on the US soybean in June 2018, the US soybean exports to Brazil has always been zero. (69,70) The Brazil custom statistics also proved that there is no change in soybean imports from other countries too. (68) Comparing soybean trade data of Argentina in 2019 and 2018, it was found that soybean export increased and import decreased, indicating that the increase of net export was due to the domestic production expansion. (1) Therefore, in the present study, the increase in carbon emissions from Brazil and Argentina land use changes triggered by the US–China trade war must be taken into consideration (see S4, Supporting Information for the detailed calculation process). Africa and Southeast Asia also potentially have problem of land use change emissions, which will potentially increase the agricultural output due to the trade war. However, in Africa, the output value of the agricultural sector will only increase slightly by 0.09–0.15% according to the present study. In Southeast Asia, the agricultural output value will only increase slightly by 0.08% in scenario 1. Therefore, we do not discuss the land use change emissions in these regions. 2.5. Data Sources We focused on the impacts of the trade war on global CO2 emissions and were also concerned about the ancillary impacts on the local air pollutants of sulfur dioxide (SO2), nitrogen oxides (NOx), carbon monoxide (CO), fine particulate matter (PM2.5), carbonaceous speciation (BC), and non-methane volatile organic compound (NMVOC) emissions. The emissions will be considered at the global, national, and sectoral levels according to data availability. The national and sectoral CO2 emission data in 2014 were drawn from the version v5.0 of the Emissions Database for Global Atmospheric Research (EDGARv5.0) (71) and the GTAP database. (42) The industrial air pollutant emission data in 2014 were also drawn from EDGARv5.0 global air pollutant emission data (72) (for the detailed data sources, please see S5, Supporting Information).

The WTO can ... ...cut living costs and raise living standards ...settle disputes and reduce trade tensions ...stimulate economic growth and employment ...cut the cost of doing business internationally ...encourage good governance ...help countries develop ...give the weak a stronger voice ...support the environment and health ...contribute to peace and stability ...be effective without hitting the headlines. When the world economy is in turmoil, the multilateral trading system can contribute to stability. Some would argue that this can even contribute to international peace. History is littered with examples of trade disputes escalating into armed conflict. It’s a claim that should not be exaggerated, but there is truth in it. If we understand why, we have a clearer picture of what the system actually does. Why was the system set up? Essentially, it was for two reasons. One was the big-picture need to avoid a repeat of the destructive trade tensions before World War II. The other was countries’ pragmatic desire for their producers to trade more easily. The result was the General Agreement on Tariffs and Trade (GATT), created in 1948, immediately after the war. Its success led to further reforms, and the World Trade Organization (WTO) emerged in 1995. An ever- increasing number of countries have agreed on trade rules that are now almost global. They are committed to the legally binding limits on their trade barriers and subsidies that they have also negotiated. Sticking to these is self-interest because countries want their trading partners also to play by the rules and stay within their commitments — and just in case pressure from domestic interests is too great, the information on raised trade barriers is shared globally through regular monitoring. By and large, peer pressure works. Two of the most fundamental principles of the trading system are at work here: helping trade to flow smoothly, and providing countries with a constructive and fair outlet for dealing with disputes over trade issues. Before: when protectionism backfired. The early 1930s saw a devastating trade war.In the Great Depression, fear that imports would throw more people out of work led governments to raise their trade barriers, thus setting off a vicious cycle of retaliation. This simply worsened unemployment. The world economy spiralled downwards, eventually contributing to the outbreak of World War II. Protectionism can easily plunge us into a situation where no one wins and everyone loses. After: restraint and confidence. Post-war, under the GATT/WTO system trade surged. More importantly, it has been much more stable, even during economic crises. Agreed rules and confidence-building are key. Confidence helps to avoid the no-win trade wars witnessed in the 1930s. When governments believe that others will keep their trade barriers within agreed limits, they will do the same. They will be in a much better frame of mind to cooperate with each other. The WTO trading system plays a vital role in creating and reinforcing that confidence. Particularly important are negotiations that lead to agreement by consensus and a focus on abiding by the rules.

Director of international law at CIGI from April 2014 to February 2020. Oonagh has a B.F.A. (Hons.) from York University (1977). She obtained her LL.B. from Osgoode Hall Law School (1981) and was called to the Bar of Ontario in 1983. She obtained an LL.M. from the University of Ottawa (1990), a Doctor of Juridical Science (S.J.D.) from the University of Toronto (1994) and an M.B.A. from Queen’s University (2007). She has taught various courses in the Faculty of Law (Common Law) and Telfer Business School at the University of Ottawa and in the Department of Law at Carleton University. “Modernizing the World Trade Organization: A CIGI Essay Series” Centre for International Governance Innovation.

There is, of course, no bar to the Supreme Court adopting a Court of Appeal's reasoning as its own. Yet there is legitimate reason to question whether that was what the Supreme Court intended in 1983. First, as just shown, only the least authoritative of the three sources cited **385 actually supports the word “must,” while the two Supreme Court decisions employ “should.” Second, barely a month later, the Supreme Court—speaking though the same justice—filed another decision which used the “should” formulation from the 1955 Allen decision as quoted in Abbott.20 Third, the 1983 Allen decision involved retirees (and Flournoy the widow of a retiree), who historically receive a heightened degree of judicial protection. (See fn. 19, ante.) Fourth, and most significantly, the “must” formulation has never been reiterated by the Supreme Court, which has instead uniformly employed the “should” *699 language from the 1955 Allen decision. (Olson v. Cory, supra, 27 Cal.3d 532, 541, 178 Cal.Rptr. 568, 636 P.2d 532 [“Although an employee does not obtain any ‘absolute right to fixed or specific benefits ... there [are] strict limitation[s] on the conditions which may modify the pension system in effect during employment.’ [Citation.] Such modifications must be reasonable and any ‘ “changes in a pension plan which result in disadvantage to employees should be accompanied by comparable new advantages” ’ ”]; Legislature v. Eu (1991) 54 Cal.3d 492, 529–530, 286 Cal.Rptr. 283, 816 P.2d 1309 [quoting Olson]; City of Huntington Beach v. Board of Administration (1992) 4 Cal.4th 462, 472, 14 Cal.Rptr.2d 514, 841 P.2d 1034 [“changes in a pension plan which result in disadvantage to employees should be accompanied by comparable new advantages,” citing Allen v. City of Long Beach, supra, 45 Cal.2d 128, 131, 287 P.2d 765].)

It thus appears unlikely that the Supreme Court's use of “must” in the 1983 Allen decision was intended to herald a fundamental doctrinal shift. “Should,” not “must,” remains the court's preferred expression. And “should” does not convey imperative obligation, no more compulsion than “ought.” (Lashley v. Koerber (1945) 26 Cal.2d 83, 90, 156 P.2d 441; see People v. Webb (1986) 186 Cal.App.3d 401, 409, fn. 2, 230 Cal.Rptr. 755 [“the word ‘should’ is advisory only and not mandatory”].) In plain effect, “should” is “a recommendation, not ... a mandate.” (Cuevas v. Superior Court (1976) 58 Cal.App.3d 406, 409, 130 Cal.Rptr. 238.)

For a policy to be useful, it must be clear. However, not all policies are the same; they differ in terms of expected results, commitment of resources, and indication of importance or urgency. Therefore, it is important to simplify the language used in the General Plan and understand the distinctions between the different levels of policy. The following definitions of terms provide guidance in interpreting the policy language of the General Plan:

Ÿ Shall: Policies containing the word “shall” indicate that an action must be taken in all cases. This represents absolute commitment to the policy, and the expectation is that the policy will always be carried out.