Artificial intelligence is everywhere. And yet, the experts tell us, it is not yet actually anywhere. This is because we are yet to achieve artificial general intelligence, or artificially intelligent systems that are capable of thinking for themselves and adapting to their circumstances. Instead, all the AI hype--and it is constant--concerns narrower, weaker forms of artificial intelligence, which are confined to performing specific, narrow tasks. The promise of true artificial general intelligence thus remains elusive. Artificial stupidity reigns supreme.
What is the best set of policies to achieve more general, stronger forms of artificial intelligence? Surprisingly, scholars have paid little attention to this question. Scholars have spent considerable time assessing a number of important legal questions relating to artificial intelligence, including privacy, bias, tort, and intellectual property issues. But little effort has been devoted to exploring what set of policies is best suited to helping artificial intelligence developers achieve greater levels of innovation. And examining such issues is not some niche exercise, because artificial intelligence has already or soon will affect every sector of society. Hence, the question goes to the heart of future technological innovation policy more broadly.
This Article examines this question by exploring how well intellectual property rights promote innovation in artificial intelligence. I focus on intellectual property rights because they are often viewed as the most important piece of United States innovation policy. Overall, I argue that intellectual property rights, particularly patents, are ill-suited to promote more radical forms of artificial intelligence innovation. And even the intellectual property types that are a better fit for artificial intelligence innovators, such as trade secrecy, come with problems of their own. In fact, the poor fit of patents in particular may contribute to heavy industry consolidation in the AI field, and heavy consolidation in an industry is typically associated with lower than ideal levels of innovation.
I conclude by arguing, however, that neither strengthening AI patent rights nor looking to other forms of law, such as antitrust, holds much promise in achieving more general forms of artificial intelligence. Instead, as with many earlier radical innovations, significant government backing, coupled with an engaged entrepreneurial sector, is at least one key to avoiding enduring artificial stupidity.
INTRODUCTION I. IP FOR AI A. Patents 1. Patentable Subject Matter 2. Disclosure Requirements 3. Novelty and Nonobviousness 4. Patenting Elasticities 5. Summary B. Trade Secrecy 1. Threshold for Protection 2. (Non)Disclosure Requirements 3. Duration 4. Patenting Elasticities Revisited C. Copyright II. AI'S INDUSTRIAL ORGANIZATION A. The Theory of the Firm B. AI's Industrial Organization 1. AI's Intellectual Property Conundrum 2. Theory of the AI Firm III. IMPLICATIONS A. Strengthening AI Patents B. Strengthening Antitrust Laws C. Government AI CONCLUSION INTRODUCTION
Forms of artificial intelligence (AI)--computing systems that perform tasks that normally would require human intelligence--are everywhere. (1) AI determines what appears in our news feeds, (2) which ads we are served, (3) our search results, (4) and how personal assistants such as Siri and Alexa respond to us. (5) AI also increasingly determines our credit scores, (6) mortgage and loan interest rates, (7) insurance premiums, (8) how much we pay for goods and services, (9) where our money is invested, (10) and our job prospects. (11) In the criminal justice context, AI plays a growing role in determining who to police and, ultimately, what criminal sanctions to impose. (12) AI plays a vital role in foreign intelligence and national security matters as well. (13) AI is projected to affect every industry and sector of society, if it has not already; (14) it is fast becoming the most important technological development in some time. (15) Indeed, some have dubbed AI the most central part of the "fourth industrial revolution." (16)
Because of its growing ubiquity and importance, AI has attracted the attention of legal scholars. (17) Privacy scholars, for instance, have largely bemoaned the lack of transparency and accountability associated with AI systems, with one prominent scholar referring to the world that we now live in as a "Black Box Society." (18) Other scholars have worried about the biases that afflict many AI systems, (19) and yet others have analyzed who should take responsibility when AI runs amok. (20) Intellectual Property (IP) scholars have also examined a number of important IP-related questions, (21) including whether IP rights should apply to the products of these autonomous, automated systems. (22) Are patents and copyrights justified, for instance, in cases where the AI system, rather than a human subject, creates the outputs? (23)
Despite this attention, many crucial questions remain. What, for instance, is the best innovation policy for spurring radical AI innovation? If AI is the most important technological development in some time, as some claim, (24) then better understanding what innovation policies are best suited to ensure its success is vital. Otherwise, artificial stupidity, rather than true artificial general intelligence, will continue as the norm. (25) Indeed, despite the incessant hype about and ever-growing uses of AI, many AI experts lament a lack of any real progress in the AI space. (26) As one such expert recently opined, "People worry that computers will get too smart and take over the world, but the real problem is that they're too stupid and they've already taken over the world." (27) While access to vast amounts of processing power and data have enabled applications of some basic AI techniques to perform specific tasks, a more general form of AI, capable of thinking for itself beyond those specific tasks, still eludes us. (28) Our computerized world is thus plagued with an artificial stupidity confined to carrying out particular, narrow tasks, and not often very well. (29) This is not to claim, of course, that narrower, weak forms of AI are not often quite valuable--we all benefit from them in a variety of ways. (30) But it is to say that realizing the full potential of AI--by acheiving stronger, more general forms of AI--requires us to examine how best to spur its further development.
Scholarly conversations about how best to incentivize AI innovation have been lacking. (31) Some of this neglect may owe to the fact that AI systems are comprised of software, hardware, data, and other technologies that have been around for some time. (32) Hence, whatever innovation policies we have had for these types of technologies may be good enough for the AI systems that utilize them. But in this Article, I argue that the nature of many AI systems challenges some of the basic assumptions underlying traditional United States innovation policy as reflected in its intellectual property laws, thus necessitating a reexamination of several of those assumptions. And doing so is not some niche exercise, because AI is not some niche technology. Instead, because AI increasingly pervades nearly every major modern-day technological system, (33) the innovation policy ramifications that I discuss in this Article will tend to apply more broadly to technological development in general.
Traditionally, IP laws form a vital, and perhaps the most important, part of United States innovation policy. (34) Hence, this Article examines several different forms of IP rights, including patents, trade secrecy, and copyrights, and assesses how each is likely to affect developers of AI systems. Doing so reveals a number of important implications for AI and technological innovation going forward.
First, patent law, often viewed as a key vehicle for incentivizing inventive innovation, (35) is often a poor fit for incentivizing radical A1 innovation. This is so for a number of reasons, including the Supreme Court's recent patentable subject matter rulings that have made effectively patenting software innovations, of which AI innovations are a subset, more difficult. (36)
Second, trade secrecy, often viewed as the primary alternative to patent protection, provides AI developers with several key advantages in comparison to patent protection. (37) For instance, keeping an AI system's technical details under wraps is often a key competitive advantage because doing so can provide significant lead-time advantages. (38) Trade secrecy allows AI developers to keep their systems secret, whereas patenting the same invention requires disclosure of key technical details as part of the patenting process. (39) Furthermore, even in cases where parties may prefer patent protection, (40) the relative ineffectiveness of AI patent protection means that more AI innovators are likely to choose trade secrecy over patent protection for their AI innovations. Thus, trade secrecy is increasingly displacing patent protection as the preferred form of legal protection in a growing number of AI contexts. (41)
Third, these IP realities mean that the AI industry is likely to become increasingly consolidated as a limited number of large, incumbent firms dominate it. (42) The reasons for this relate both to the IP choices available to industry participants as well as the nature of AI innovation in general. For instance, in industries with "weak appropriability regimes," large incumbent firms have an easier time fending off would-be competitors in part because ineffective patent protection means the potential rivals have greater difficulty realizing the value of their innovations. (43) Ineffective patent protection is also likely to contribute to higher costs of parties doing business with each other, so that market participants are more likely to "vertically integrate" than strike...