Events

Moderation: Klaus Wiedemann

To eliminate AI content risks, the Chinese government has embraced a zero-risk policy, culminating in a sweeping list of prohibited outputs outlined in Article 4 of its Interim Measures. Jiawei critiques this approach, placing large language models (LLMs) within the broader information ecosystem and exposing fundamental flaws in the policy design. First, the pursuit of fully de-risking LLMs is unattainable. LLMs merely replicate the problems that have already existed in the information marketplace. As the whole information ecosystem is far from perfect, expecting LLMs to function as a zero-risk engine is unrealistic. Additionally, Jiawei argues that the Chinese zero-risk approach is unnecessary. An AI chatbot’s output competes not only with other chatbots but also with other information outlets. Market forces, therefore, have great potential to incentivize AI companies to improve their output and derisk their systems automatically, voluntarily, and continuously. Jiawei uses Article 4 of Chinese Interim Measures as a case study to further illustrate how to design a scientific framework to govern AI risks and avoid setting unrealistic standards and imposing onerous duties on AI companies.

In an experimental idea generation context, we study the effect of input incentives, which reward time invested to generate ideas, and output-based incentives, which reward the number of innovative ideas generated, compared to simple fixed wages. Overall, we find that both input and output incentives increase the average number of innovative ideas to a comparable extent. Under input incentives, this increase is due to a rise in effort duration, i.e., an increase in the time spent generating ideas. For output incentives, this increase is due to both heightened effort duration and effort intensity, i.e., the number of ideas generated per unit of time. The optimal incentive scheme seems to depend on the context. We find that output incentives result in innovative ideas at lower costs. However, output incentives also cause a shift in the types of ideas generated, with individuals focusing on less time-intensive ideas. (Joint work with Kay Blaufus and Kevin Piehl)

Contact person: Svenja Friess

Subscription to the invitation mailing list and more information on the Seminarseite.

The common practice to include firm fixed effects in empirical research may eliminate the explanatory power of important economic factors that are persistent. To illustrate this point, we review the intuitive R&D-patent relation in recent studies and present a surprising pattern that R&D input only positively explains patent output in half of prior regression estimations. This “missing link” can be attributed to the persistence of R&D and patents that causes the between-firm variation to be absorbed by firm fixed effects. We consider modified Hausman-Taylor estimates and advanced machine learning methods, and find that both methods lead to a consistent positive R&D-patent relation. In particular, the advanced machine learning methods select some firm dummies that have explanatory power for patent output, and exclude other firm dummies are not informative for – and may even bias – the identification. This paper thus offers two methods to serve as a “second opinion” for empirical researchers working with explanatory variables that strongly correlate with between-individual unobservables.

Contact person: Daehyun Kim

Subscription to the invitation mailing list and more information on the seminar page.

We use a production function framework to estimate how government-funded R&D leads to the creation of new scientists and science. Focusing on public investments into specific energy technologies by the US Department of Energy (DOE), we leverage quasi-experimental variation in R&D investment that emerges when Congress appropriates more or less funding than what the DOE requests to estimate the effect of public funding on dissertation flows and recover the marginal costs of producing new (potential) scientists as proxied by dissertation flows. We also document the relationship between dissertations (potential scientists) and patents (new science). Our use of the production function framework allows us to recover technology-specific productivity trends and estimate the number of scientists and amount of science that would have been produced under alternative funding allocations. Given how noise in the budgeting process is prevalent, our adaptation of the production function framework could prove useful in many other settings.

Contact person: Ulrike Morgalla

Subscription to the invitation mailing list and more information on the seminar page.

University-based scientific research has long been argued to be a central source of commercial innovation and economic growth. Yet at the same time, there have been long-held concerns that many university-based discoveries never realize their potential social benefits. Looking across universities, research and commercialization activities such as start-up formation vary tremendously – variation that could reflect the composition and orientation of faculty research, university-level factors such as patenting and licensing efforts, or broader place-based factors such as location in a technology cluster. We take a first step towards unpacking this heterogeneity in university commercialization by analyzing how the propensity of academic research to spill over to commercial innovation changes when academics move across universities. Our estimates suggest that at least 15–25% of geographic variation in commercial spillovers from university-based research is attributable to place-specific factors.

Contact person: Marina Chugunova

Subscription to the invitation mailing list and more information on the seminar page.

Scalable and low-cost AI assistance has the potential to improve firm decision-making and economic performance. However, running a business involves a myriad of openended problems, making it difficult to know whether recent AI advances can help business owners make better decisions in real-world markets. In a field experiment with Kenyan entrepreneurs, we assessed the impact of AI advice on small business revenues and profits by randomizing access to a GPT-4-powered AI business assistant via WhatsApp. While we are unable to reject the null hypothesis that there is no average treatment effect, we find the treatment effect for entrepreneurs who were high performing at baseline to be 0.27 standard deviations greater than for low performers. Sub-sample analyses show high performers benefited by just over 15% from the AI assistant, whereas low performers did about 8% worse. This increase in performance inequality does not stem from differences in the questions posed to or advice received from the AI, but from how entrepreneurs selected from and implemented the AI advice they received. More broadly, our findings demonstrate that generative AI is already capable of impacting—though in uneven and unexpected ways—real, open-ended, and unstructured business decisions.

Contact person: Daehyun Kim

Subscription to the invitation mailing list and more information on the seminar page.

Biosimilars are a relatively new class of drugs designed to be nearly identical to existing (“original”) biologic therapies—the equivalent of generics for traditional drugs. Patients, payers, and policy makers have long anticipated their widespread market penetration due to their cost-saving potential. However, their penetration rate has been disappointingly slow. We present evidence that manufacturers leverage their broader commercial relationships with clinics to influence biosimilar penetration. For manufacturers of the original biologic product, this behavior entails a pull effect that hinders biosimilar penetration. For biosimilar manufacturers, a push effect that promotes it. Our results portray these as highly influential effects. One implication is that biosimilars commercialized firms with strong pre-existing relationships with clinics may penetrate the market significantly faster. Our findings also reframe the regulatory debate around the use of exclusionary contracts that restrict the use of biosimilars, shifting the focus away from the use of these practices to the economic fundamentals that make these practices acceptable for clinics. (Manuel Hermosilla, Joseph Levy & Kelly Anderson)

Contact person: Elisabeth Hofmeister

Subscription to the invitation mailing list and more information on the seminar page.

The conference will deal with two topics:

Standard Essential Patents Litigation in the Unified Patent Court

When the UPC Agreement was drafted, FRAND and Standard Essential Patents were not as important topics as they are today in patent litigation. The UPC Agreement therefore does not explicitly address what should be the legal framework for such proceedings. In one of the first cases dealing with SEP litigation, the Mannheim Local Division considered the negotiation framework established by the European Court of Justice in the 2015 case of Huawei Technologies Co. Ltd. v ZTE Corp. as relevant for FRAND licensing negotiations and shed some light on important issues of confidentiality in FRAND litigation.

The panel will discuss on what legal basis SEP cases should be litigated in the UPC and to what extent the UPC courts will not only grant or dismiss injunctions, but also define the conditions of FRAND licenses. In general, the question arises as to whether the UPC changes the way SEP litigation is conducted in the UPC member states.

Protection of Competition in AI Innovation

Given the significant economical and societal implications of AI technology, it is vital to ensure that competition in the field of AI, as well in the sectors where AI finds applications, remains fair, open, and ultimately serves societal values. Are competition law and policy well-equipped to address the specifics and challenges of competition in AI innovation?

The panelists are invited to share perspectives on this multi-dimensional topic, as well as to explore the interactions between competition law with other legal frameworks - particularly intellectual property rights, trade secrets, and recently adopted instruments for regulating the data economy - in the context of AI innovation.

Programme as pdf
 

Registration is closed.