Politicians and experts are advocating for the development of “trustworthy” Artificial Intelligence (AI). However, according to researchers from the Leibniz Centre for European Economic Research (ZEW), appeals and regulations alone are not sufficient to guide popular chatbots and image generators from companies like OpenAI and Google in this direction.
Dominik Rehse, head of a ZEW research group focused on the design of digital markets, notes that there is a “market failure” with generative AI. Developers can currently generate significant profits with more powerful models, but they do not bear the costs of the societal risks associated with these technologies.
To address this imbalance, ZEW economists propose a new EU funding program in a policy brief. The program aims to create incentives for the development of safe generative AI. Rehse explains that the safety of these systems represents a socially desirable innovation, which individual development companies have not sufficiently provided. A similar issue exists in the production of certain vaccines, where such funding programs have already been successfully implemented.
The team’s idea is that interested developers participate in a competition funded and organized by the EU. Systems that meet specific safety and performance criteria would be advanced in stages. Upon reaching a milestone, developers would receive predetermined financial rewards. This approach reduces demand uncertainties for companies, as the industry knows from the start that the effort for innovation will be worthwhile. At the same time, the EU does not prescribe specific technologies, remaining open with few technical requirements.
According to the group, carefully defined safety and performance metrics are necessary. The robustness of these systems should primarily be ensured through transparent testing, where the public can scrutinize AI algorithms for misconduct or exceptional performance.
With such incentive mechanisms, the EU has the opportunity to develop innovations more targeted than before, emphasizes co-author Sebastian Valet. This approach could also allow the EU to carve out a niche in the generative AI market, where it currently plays only a minor role.