Image credit: Adobe Stock

Written by Walter G. Park, Professor of Economics, College of Arts and Sciences, American University, USA.

A Key Issue

Research on the impacts of artificial intelligence (AI) on the economy and the appropriate role of intellectual property (IP) rights in an AI-driven economy is currently at a nascent stage. Technologies based on artificial intelligence will continue to permeate throughout society and further affect jobs, production, incomes, income distribution, and global trade. Existing IP laws and institutions, and policy and regulatory authorities, are tasked with overseeing AI developments. This article asks the following question:  How does AI condition the optimal design of IP protection? Answering this will help explain how IP protection will best work to facilitate economic growth and development amid continuing advances in artificial intelligence.  I begin with a recap of what I believe is the conventional view of the appropriate role of IP protection for generic technological innovation, and then discuss how the conventional view might be adjusted in the case of AI technologies. I then conclude with some brief thoughts on the implications for economic development.

Conventional View

The prevailing view is that IP protection has both costs and benefits for social welfare. My recently edited Handbook discusses the various trade-offs entailed by intellectual property rights.[1] IP protection confers some market power on the IP owner. This limits the availability of a good — or access to the good — since it is not apt to be competitively supplied. Prices are in general raised as a result of IP protection, and for some consumers or users, a good will be less affordable, or not affordable at all.  Balanced against this cost is that IP protection can stimulate innovation and creative activity which enhances the quality of goods or expands the number or variety of goods. [2]

More specifically, one can dive more deeply and argue that IP protection affects the size and frequency of innovations. If either the duration or scope of protection is too low, innovations will arise few and far between; and what little do occur will be of modest size, since innovators will not invest in large or ambitious research projects or creative endeavors if the protection is too weak to generate sufficient returns (due to piracy, imitation, or infringement). As protection increases, innovations can become more frequent and more radical. The greater market exclusivity and opportunities for greater returns create incentives for innovators to invest in larger breakthroughs. However, if protection is too stringent, innovators face less competition, less threat from having their innovations displaced, and therefore may introduce new innovations more slowly, ones that might even be incremental improvements over existing innovations. Thus, one can imagine an inverted-U relationship between the level of IP protection and innovation and ultimately social welfare. That means that there exists some optimal strength of IPR protection for purposes of maximizing social well-being.

But this optimal strength should vary by economy. In particular, for developing countries, it should typically be lower than that for developed countries. Typically, the Global South conducts less R&D, owns less of the world’s IP assets, and has lower innovative capacities. A greater propensity for imitation and piracy exists in the developing world owing to the affordability issue under IP protection. Moreover, the market size of the Global North is much larger than that of the Global South. Because of this, most of the innovators recoup their returns in the developed country markets. This is one reason why IP protection should be stronger in the developed world.  And since the majority of innovating firms are based in the developed world, their profits add to the social welfare of the developed world and help offset losses to social welfare stemming from higher prices. The developed world, in other words, is better able to absorb the costs of IP protection.[3]

The optimal strength of IPR protection should also vary by industry, such as pharmaceuticals, transportation, computers, and electronics. The variation by industry is dependent on the technological characteristics.  For example, hardware versus software. If hardware is especially capital-intensive, this intensity can be a natural barrier to imitation. Another feature is the complexity or discreteness of products. Discrete products such as drugs are relatively easier to imitate. Complex products, in contrast, which comprise multiple pieces are harder to reverse engineer because the components would not easily reveal the nature of the composite product.  Whether industries produce goods with long or short product life cycles will also affect their dependence on IP protection. When product life is short, technological obsolescence may arise well before the product is imitated. Thus, some industries are more dependent on IP protection than others depending upon the ease of imitation. Ideally, therefore, the strength of protection should vary by industry, being stronger in those sectors facing higher imitation risks.

Influence of AI

Artificial intelligence technologies should influence the optimal levels and design of intellectual property protection in diverse ways. They tend to raise both imitative and innovative capacities. First, AI raises imitation risks since the technologies make it easier to produce copies of IP works, mimic original creations, and engage in reverse engineering. These risks may require a more stringent enforcement of IP to the degree that they discourage or pre-empt innovation.  On the other hand, AI can enhance the productivity of research and production, and improve the ability of developing economies to catch up technologically. For example, AI aids immensely in recognizing and analyzing patterns in large datasets and seeking solutions to complex problems more speedily. AI may generate its own creations or innovations endogenously. Like plant variety innovations, which carry the seeds for further innovations and reuse, AI has the ability to create (or be programmed to self-create) original technologies beyond what the human programmer/creator might have intended or imagined resulting. High valued AI innovations may also require sufficiently secure IP protection to motivate their creations but the ease of innovation due to AI may call for a relaxation of stringent levels of protection.

Compounding the issue of the appropriate design of IP protection is that AI has both hardware and software features. Software includes the algorithms and the data, while the hardware includes the supercomputers with vast numbers of GPU’s, the robots, chips, solid state drives, the tensor processing units. Software is typically easier to reproduce but the codes border very closely to being material of mathematical nature that is unpatentable in many jurisdictions, unless some legal requirements as to “technical effect” are demonstrated. (Of course, software does enjoy copyright protection.) Hardware, as was pointed out earlier, may inherently be harder to imitate if it is capital-intensive and complex, and thereby associated with high setup costs. Having both hardware and software components make calibrating levels of IP rights around AI innovations uniquely challenging.

From a North-South perspective, the Global North is likely to widen its technological lead over the Global South, given that most of the AI innovations occur and are owned by firms in developed countries, such as the U.S. While China, from the developing world, is offering a formidable challenge to the U.S., other developing economies are likely to see an increase in their technological balance of payments deficit to the developed world. Developments in AI (and any stronger IP protection for AI) may well worsen the terms of trade for the Global South and widen the skills gap between the South and North unless AI knowledge, training, and computer science skills are transferred or diffused widely around the world, and unless IP rights on AI are well-balanced.

A balancing of the costs and benefits of IP protection for AI should take into consideration a number of factors. First, the value and efficacy of AI as a tool for creativity and innovation depend on AI systems having full access to data — that is, the training data. Withdrawing information or material from the training dataset because copyright owners refuse to let their works enter the knowledge base, unless they receive compensation, will make AI less effective as a resource.[4]

Second, an anti-commons effect also threatens the efficacy of AI. While there are legitimate concerns about privacy, ethics, and safety, AI tools and innovations are likely to be underutilized if policymakers over-regulate. Appel et al. (2023) have identified a number of important concerns and legal risks with AI.[5] But methods to mitigate the risks, such as permitting artists to opt out of image generators, developing audit trails (to source the provenance of AI-generated content and the platform with which it was developed), disclosures, agreements, and generative AI check lists, must not be such as to be burdensome and raise the transactions costs of deploying AI.

Third, AI challenges the non-obviousness standard of patentable innovations.[6] If AI assisted human innovation, how much credit should the human inventor receive and what was the investment cost of the innovation if AI facilitated it? (For example, a major cost component of pharmaceutical R&D cost are the clinical trials. If AI can prevent multiple failed trials, the upfront fixed cost of R&D can be lowered significantly.) Moreover, what might seem complex and nonobvious to a human inventor might be obvious to a supercomputer that can better discover emergent patterns.

In view of the potential economic benefits of AI-related innovations, the debate over IP might center upon the commercialization costs — of delivering these benefits to end users. Even if the innovation is less costly to create due to AI’s assistance or generation, or the innovation is less non-obvious from the vantage point of an AI model, the commercialization of it may require large investments. The question then is whether a period of market exclusivity is necessary for the investors to recoup those upfront costs. This is a question that resembles the conventional rationale for IP protection, but focused on the introduction of the technology into the marketplace. Exclusivity is especially needed if the imitation and infringement risks are on balance greater for AI-related innovations.

A close eye on levels of industry concentration is needed as the key players involved in commercialization acquire market exclusivity. Currently, the market is oligopolistic with companies such as Microsoft and NVIDIA possessing significant market shares of AI-related products. A Goldilocks rule should govern the temporary market exclusivity: not too strict as to diminish the drive to innovate when competition is limited and not too weak as to dissipate the returns needed to recoup the upfront costs. Non-exclusive licensing to enable other participants to market the AI-based products will no doubt mitigate concerns about concentration.

More attention is also needed on technology standards and the role of standard essential patents. As AI-related innovations expand, and if incumbent firms continue their market dominance, there is a risk of locking into established technology standards which could be harmful if markets lock into inefficient standards.

Implications for Economic Development

AI development is currently heavily concentrated in developed economies. To date, no formal evidence has been gathered on the extent to which AI technologies have diffused to developing countries nor on the extent of indigenous AI innovation in the Global South. For sure, the Global South is heterogenous. Some are still agricultural-based and exhibit innovative capacities in that area, such as genetically modified food products. Others with experience in electronics and computers are part of the global value chain and can make value added contributions to AI development.

IP protection should be appropriate for providing multinational and other companies incentives to establish facilities in the South — to run the supercomputers, create global knowledge networks and research labs. IP and other policies should also create incentives for local firms and workers to invest in the skills needed. A strong motivation for human capital investment in AI-related skills would be the availability of high-value added jobs. The Global South can be participants in producing the intermediate inputs that go into AI manufacture. With labor cost advantages, the region can be a draw for outsourcing or offshoring. To the extent that AI-enabled supply chains help reduce firm costs, the savings can be used for innovation.

Other potential ways in which AI could benefit the Global South are that several middle-income economies have presence in the ICT industries and, with licensing or joint ventures, can access new AI innovations and in turn become exporters or licensors of AI technologies.[7] AI can also help with climate change mitigation and agricultural crop research, or with searches for vaccines, particularly for neglected tropical diseases, or resource management for minerals in Africa (such as cobalt, nickel, lithium).[8] As in the developed world, AI and digital platforms can offer workers in low-income countries work flexibility, such as remote or hybrid work, and consumers access to a more connected global, online marketplace.

As AI eventually diffuses widely in the Global South — and labor markets there are restructured due to automation and other disruptions — social safety nets, unemployment insurance, and retraining will play an important role.[9] These policies seem imperative if the Global South is to share in the benefits of AI innovation. Some income redistribution schemes, via the tax system, could be implemented whereby the gainers can compensate the losers.

AI technologies are transformative and some intellectual property regulations remain relevant while others may need to shift – if not in law, then in practice – such as greater uses of research exemptions and fair uses for human capital growth and the training of AI, newer standards of novelty and non-obviousness and industrial applicability. Something will need to evolve, whether it is the institutional regime or the innovation system.[10]

---

[1] See W. Park (2024) Handbook of Innovation and Intellectual Property Rights, Edward Elgar, U.K.

[2] These tradeoffs are addressed in Chapters 2 and 7 in the Handbook, op cit.

[3] International aspects of IP are addressed in Part 1, as well as in Chapters 14, 20, and 21 of the Handbook, op. cit.

[4] Training AI and copyrights is discussed in Chapters 11-12 of the Handbook, op. cit.

[5] Appel, G., Neelbauer, J., and Schweidel, D. 2023. “Generative AI Has an Intellectual Property Problem.” Harvard Business Review. https://hbr.org/2023/04/generative-ai-has-an-intellectual-property-problem.

[6] See Shemtov and Gabison (2022) “The Inventive Step Requirement and the Rise of the AI Machines.” In Abbott, R. (ed.) Research Handbook on Intellectual Property and Artificial Intelligence. Edward Elgar Publishing, pp. 423-442.

[7] For the potential of countries like India in ICT, university research and commercialization in Turkey, and start-up entrepreneurs in sub-Saharan Africa, see Part VI of the Handbook, op. cit.

[8] These sectors are discussed in Part V of the Handbook, op cit.

[9] See C. Freund (2023), “Advances in Technology Affect Trade and Vice Versa,” IMF Finance & Development, June, pp. 15-17

[10] More on exceptions and alternative forms of IP is addressed in Chapter 13 and Part IV of the Handbook, op. cit.

---