Of Mice and Academics: Examining the Effect of Openness on Innovation
GAICIL Insights
Lock up the lab mice, and science narrows as well as slows.1 Of Mice and Academics: Examining the Effect of Openness on Innovation, by Fiona Murray, Philippe Aghion, Mathias Dewatripont, Julian Kolev and Scott Stern, examines the tension between intellectual property, meant to reward inventors, and access restrictions that can slow the work of others. Openness, by lowering access costs to existing research, can enhance both early and late stage innovation through greater exploration of novel research directions.
The evidence is a natural experiment: late-1990s agreements involving the National Institutes of Health that reduced academics' costs of access to certain genetically engineered mice. Difference-in-differences estimates show increased openness encouraged entry by new researchers and exploration of more diverse research paths. It did not reduce the creation of new genetically engineered mice. Overall follow-on research increased after the agreements. More scientists, more directions, and no shortage of new tools.
Diversity is the distinctive finding. Openness was associated with a wider range of research paths, reflecting more horizontal experimentation. That highlights an often-overlooked feature of early stage knowledge: it is non-rivalrous and can in principle be applied across many later-stage domains. Open access sends more researchers off in more directions. Closed access herds them down the same corridor.
Exploration and exploitation get compared. Where pre-existing IP restrictions limited access to research materials, openness mainly increased more basic and higher-quality follow-on research. A neglected cost of strong IP restrictions emerges: lower exploration, leading to reduced diversity of research output. The lost experiments of a closed system are invisible because they never happen.
The parallel with artificial intelligence is nearly exact. AI research depends heavily on access to foundational tools: models, datasets, code and computing resources. Debates about open versus closed AI models echo the questions examined here. Openness of upstream tools can encourage entry by new researchers and exploration of more diverse directions, enriching the research ecosystem. Today's model weights play the role of yesterday's engineered mice.
Diversity matters especially for AI. Progress benefits from many researchers exploring different approaches and applications. Restricted access to foundational resources may concentrate research along fewer paths, shrinking the variety of ideas tested. Openness broadens exploration and raises the odds of discovering valuable new directions. A monoculture of research is as fragile as a monoculture of crops.
The no-decline result answers the usual objection. A common worry is that openness erodes incentives to create upstream tools. In this setting, the supply of new genetically engineered mice did not fall. For AI, openness need not undermine incentives to develop foundational models and datasets, though each context's specifics matter. The feared trade-off did not appear here.
Policy implications follow. IP and access policies for research tools should weigh effects on exploration and diversity alongside effects on tool creators' incentives. Facilitating access to foundational research resources, particularly for academic researchers, can foster more diverse and higher-quality follow-on research.
The design is reusable. Studying a change in access terms for specific research tools identifies the causal effects of openness on follow-on research. Similar evaluations of AI access policies, such as releases of open models or datasets, could provide valuable evidence. Every open-weights release is a potential natural experiment.
At bottom, openness can enhance innovation by letting more researchers explore more directions. Lower access costs encouraged entry, diverse exploration and high-quality follow-on research without reducing new tool creation. As AI becomes foundational across science and the economy, policies on access to models, data and tools will shape innovation's diversity and dynamism, and this study offers rigorous evidence that openness can drive progress.
One question to put to the open-versus-closed AI debate: count the research directions as well as the risks. The mice suggest that the true cost of closing a foundational tool is measured in experiments nobody runs, and that cost never appears on any balance sheet.
Introduction
This paper argues that openness, by lowering costs to access existing research, can enhance both early and late stage innovation through greater exploration of novel research directions. We examine a natural experiment in openness: late-1990s NIH agreements that reduced academics’ access costs regarding certain genetically engineered mice. Implementing difference-in-differences estimators, we find that increased openness encourages entry by new researchers and exploration of more diverse research paths, and does not reduce the creation of new genetically engineered mice. Our findings highlight a neglected cost of strong intellectual property restrictions: lower levels of exploration leading to reduced diversity of research output. (JEL I23, O31, O33, O34)
Takeaway
In this paper we argued that openness of upstream research does not simply encourage higher levels of downstream exploitation: it also raises the incentives for additional upstream research by encouraging the establishment of entirely new research directions. We tested this hypothesis by examining a “natural experiment” in openness within the academic community: NIH agreements during the late 1990s that circumscribed IP restrictions for academics regarding certain genetically engineered mice. First, we found, not surprisingly, that there is an increased overall level of follow-on research taking place after the NIH-DuPont-JAX openness agreements. Building on this initial result, we explored the particular margins where this increased innovation is taking place, developing measures of innovation that allowed us to test the predictions of our theory. First, we obtained robust evidence that increased openness was associated with the exploration of a wider range of more diverse research paths, which in turn reflects an increase in horizontal experimentation. This finding highlights a feature of early stage knowledge overlooked in many of the current models of innovation - the fact that it is non-rivalrous and as a consequence, can, in principle, be applied across multiple later-stage domains and applications. Second, when we compared the impact of openness on horizontal exploration versus vertical exploitation we found that on balance, when pre-existing IP restrictions limited access to research materials (rather than simply serve as a threat of potential future enforcement), the main impact of openness was an increase in more basic and more high-quality follow-on research publications.1
In contrast, when prior arrangements (informal or formal) had allowed for access even with some threat of enforcement, the openness shock would mainly affect the more applied follow-on research. Our results highlight that the current literature on intellectual property and innovation has neglected a key potential cost of intellectual property - the limits that IP rights may place on the diversity of research that would otherwise be pursued by follow-on innovators taking a single powerful idea and experimenting across multiple research lines. Our research in this paper could be extended in several interesting directions. A first avenue would be to reassess the Bayh-Dole Act in light of our analysis. Indeed our results highlight one of the possible dangers of excessive IP enforcement: if IP is used to restrict openness particularly at very early stages of the research line, then it is possible that the rich array of exploration projects that are key to diverse follow-on innovation will be stifled.
Notes
- Fiona Murray et al., Of Mice and Academics: Examining the Effect of Openness on Innovation, 8 Am. Econ. J.: Econ. Pol'y 212 (2016), https://doi.org/10.1257/pol.20140062. Takeaway source: Conclusion, reproduced from a free author or repository copy (working-paper version). ↩
Cite as: Fiona Murray et al., Of Mice and Academics: Examining the Effect of Openness on Innovation, 8 Am. Econ. J.: Econ. Pol'y 212 (2016), https://doi.org/10.1257/pol.20140062
Header image: Archives of the Lister Institute; laboratory Wellcome L0021989.jpg by Unknown author, CC BY 4.0, via Wikimedia Commons.
