Understanding AI and Its Impact
At its core, Artificial Intelligence (AI) can be understood as systems (algorithms) that perform statistical learning and inference, using various types of data and relying on dedicated computing power. Think of it as a sophisticated cloud infrastructure made up of three essential components: data, advanced algorithms, and the powerful computers needed to process them. This technology has gained immense momentum since 2010, driven by breakthroughs in machine learning and deep learning.
Many experts consider AI a "General Purpose Technology" (GPT), much like the steam engine or electricity in past centuries. This means AI has the potential to be used widely across many sectors and can continuously evolve, leading to new innovations and increased productivity throughout the economy. It acts as an "enabling technology," opening up new possibilities rather than just providing final solutions. AI profoundly impacts how we produce knowledge, not only by helping researchers manage the "explosion of data" (finding more "needles in a haystack") but also by suggesting new combinations of existing knowledge to create entirely new insights (predicting "which combinations of existing knowledge will yield valuable new knowledge").
However, the rapid rise of AI also comes with concerns. Some argue that AI is not as widespread as believed, primarily being adopted in limited sectors like ICT, professional services, and finance, often as an addition to existing systems rather than a complete overhaul. More critically, the development of AI has led to worries about its application and accessibility. There's a growing dominance of large technological companies in controlling crucial AI resources—data, computing power, and even the top researchers. This is partly due to the "data hunger" of AI systems, as generating predictions relies heavily on access to vast amounts of private data and the infrastructure to process it, which is often controlled by these big-tech firms. This dynamic creates a situation where companies provide hardware and data, while universities supply trained personnel, leading to increasing collaboration between them.
This growing interdependence raises significant issues:
Exclusion of non-elite universities: Scientists from less prestigious institutions may be left behind, creating a technology gap.
Narrowing of research focus: A small group of powerful players might dominate AI research, potentially limiting the diversity of knowledge produced. Industrial AI research often focuses on topics with short-term profit potential, such as computer language, computer vision, and recommendation systems, sometimes neglecting long-term societal impacts.
Threat to university autonomy: There's a fear that increased industry collaboration could push university research towards a "third mission" – a focus on commercial, profit-oriented outcomes – which is seen by some as outside the traditional scope and potentially undermining academic freedom.
However, the paper challenges this last concern, arguing that such collaborations are not unprecedented and do not necessarily undermine university autonomy. To understand this, we must look to the history of university-industry links in the US.
A Historical Perspective on University-Industry Relationships in the US
The evolution of universities and their engagement with industry in the US has gone through several distinct phases.
The Prewar Period (before WWII): Deep Industry Ties Universities, even in their earliest forms in the 12th century, were rooted in a "utilitarian soil," serving societal needs by educating professionals in medicine and law, and contributing to economic development. The German Humboldtian model, which combined teaching with research to train bureaucratic and professional elites, was later adopted in the US.
In the 19th and early 20th centuries, as the US industrialized, the need for trained workers and engineers grew immensely. The Morrill Act of 1862 was a landmark, establishing "land-grant colleges" dedicated to practical education in agriculture and mechanical arts. These institutions, like the University of Akron supporting the rubber industry, were designed to provide industry with necessary skills. Universities served a "third mission" by transferring agricultural knowledge to farms, directly contributing to productivity.
Formal links between universities and the private sector were crucial due to limited government funding. Curricula were often tailored to industry needs, and large firms like Du Pont and AT&T sponsored graduate fellowships to secure future talent. In-house R&D laboratories began to emerge within large corporations (e.g., DuPont, Kodak, General Electric) at the end of the 19th century. These labs focused on converting scientific knowledge into marketable products, and universities were seen as a source of talented individuals who could be recruited full-time. Patents were aggressively used by large firms to protect their innovations and maintain market power. In essence, the prewar system was largely industry-driven, with universities playing a vital role in training the workforce and, in some cases, acting as external research labs for new companies.
The Postwar Period (1945-1980s): Federal Ascendancy and Basic Research World War II marked a significant shift. The success of projects like the Manhattan Project highlighted the critical role of scientific research, leading to a greater involvement of the federal government as a primary funder of research in both academia and industry. Vannevar Bush's 1945 report proposed extensive federal support for university research, especially in military and medical fields.
This period saw a dramatic increase in federal R&D funding, growing from around $44 billion in the 1950s to over $165 billion by 2022. A key aspect was the shift towards basic research within universities, driven by the belief that fundamental scientific inquiry was best conducted in academic settings. Industry, meanwhile, focused more on improving existing products and commercialization. This influx of federal funds somewhat weakened the aggressive university-industry ties seen before WWII, as universities no longer needed to pursue corporate partnerships as actively for funding. This era also saw the rise of new, innovative small firms, partly due to a more relaxed intellectual property rights (IPR) regime, which encouraged broad diffusion of knowledge. The "third mission" was still present, though perhaps less overtly, tied to national priorities like health and defense.
The Revised Social Contract (1990s-Today): A Return to Responsiveness Since the 1990s, there has been another shift, often termed a "revised social contract". Universities are now expected to be more responsive to the industrial and economic impacts of their activities and face increased accountability for public funds. The paper argues that this is not a radical departure, but rather a return to historical norms; the "third mission" has always coexisted with teaching and research, with the 1945-1980s period being a temporary deviation where direct economic returns were less immediately expected.
AI R&D: "Back to the Origins?"
The central argument of Borsato and Llerena is that the current dynamics of university-industry collaboration in AI R&D largely echo the prewar period.
Commonalities with the Prewar Era:
Close Collaborations and Talent Recruitment: Just like before 1940, there are tight collaborations between universities and industry in AI. Engineering departments at top universities (like Harvard and MIT in the past) are working closely with private companies. Companies fund fellowships and actively recruit PhDs, scientists, and engineers directly from universities, making universities important sources of talent and external monitoring for industrial research labs.
Dominance of Large Corporations: A significant portion of AI R&D is conducted by very large companies, with limited reliance on smaller firms, much like the prewar era where industrial research was concentrated in large enterprises in sectors like automotive, petrochemicals, and electrical engineering. Independent research organizations are complementary, not substitutes, for in-house labs.
IPR Regime Resemblance: The current intellectual property rights (IPR) regime in AI, with large firms accumulating patents to secure strategic assets and maintain market power, resembles the prewar period. This can create a "virtuous cycle" for patent holders but a "vicious" one for others lacking IPRs.
Data Accessibility as a Driver: Limited access to public sector data often compels university researchers to collaborate with private firms that possess vast, accessible datasets, for example, from social networks.
Funding Reliance: The paper hypothesizes that current federal investments in AI R&D might still be insufficient, pushing universities to rely more on private funds and their critical assets, mirroring the prewar situation where decentralized and limited government funding necessitated private sector linkages.
Peculiarities of Current AI R&D (Key Differences): Despite these similarities, AI R&D also presents unique characteristics:
Universities as Training Hubs: Universities are increasingly tailoring their role in AI innovation towards research-based training of skilled labor, rather than solely focusing on developing new scientific and technological knowledge (except for very fundamental, upstream research).
Focus on Fundamental Science for Broad Application: The workforce trained for AI research in industry is not just taught coding; they receive a deep understanding of fundamental sciences like physics, chemistry, and mathematics. Firms seek scientists with a broad knowledge base applicable to a wide range of tasks, aligning with the "Humboldtian perspective" of comprehensive education.
Broader Goals beyond Firm Profits: Unlike the early US NSI, AI R&D is much less exclusively "firm-oriented." Significant resources are dedicated to achieving Sustainable Development Goals (SDGs), which have a far broader scope than past industrialization concerns.
Increased Stakeholder Involvement: The number of actors involved in AI R&D has expanded beyond just universities and private firms to include associations, governments, and even supranational entities.
These dynamics have implications for how knowledge and labor are organized. The AI innovation system encourages researchers to invest in their human capital early, leading to job opportunities and inter-organizational collaborations. Universities are crucial in providing the skilled labor necessary for technological advancement. In this context, the paper suggests that cognitive processes (skills) à la Babbage may drive the division of labor, rather than the traditional Smithian view where the division of labor dictates specialized knowledge. In other words, having specific skills comes first, and that then determines how tasks are divided.
Conclusion
The research by Borsato and Llerena offers a compelling perspective on the evolving relationship between universities and industry in AI R&D in the US. While concerns exist about industry dominance and the potential for a narrow research focus, the paper argues that the increasing commitment of universities to a "third mission" – contributing to the economy and society – is not new. Instead, it represents a return to a social arrangement that largely prevailed before WWII, where close university-industry links were common and often beneficial. Universities are seen as evolutionary entities that continuously adapt to their changing environment.
However, the current AI landscape also introduces unique elements. Universities are increasingly concentrating on research-based training that equips individuals with a fundamental scientific understanding, making them highly adaptable to industry needs. Moreover, the goals of AI R&D are broader, encompassing global challenges like the Sustainable Development Goals, and involving a wider array of stakeholders than ever before. This unique blend of historical patterns and novel characteristics underscores the dynamic nature of innovation systems and the indispensable role of universities in shaping the future of AI. The ongoing challenge will be to ensure that the "triple helix" of university, industry, and government fosters innovation while also serving the broader public interest and strengthening governmental capabilities in this critical technological domain.
