Role of the state and responsibility in governing artificial intelligence: a comparative analysis of AI strategies

ABSTRACT

Technologies based on artificial intelligence (AI) represent a crucial governance challenge for policymakers. This study contributes to the understanding of how states plan to govern AI with respect to the role they assume and to the way they develop AI in a responsible manner. In different policy instruments across 22 countries plus the European Union, there is considerable variation in how governments approach the governance of AI, both regarding the policy measures proposed and their focus on public responsibility. Analysing a set of policy instruments we find multiple modes of AI governance, with the major difference being between self-regulation-promoting and market-based approaches, and a combination of entrepreneurial and regulatory governance approaches. Our analysis also indicates that the approach to public responsibility is largely independent of the chosen policy mix of AI governance. Therefore, responsibility seems to be a cross-cutting issue that cannot be tied to a specific approach of states towards technology.

KEYWORDS:

• Artificial intelligence
• AI governance
• policy instruments
• state types
• responsible research innovation
• technology assessment

Acknowledgements

The authors thank the two anonymous reviewers for the time and efforts spend on commenting previous versions of this manuscript. The final paper profited immensely from their feedback and suggestions. We would also like to extend our thanks to Jeremy Richardson, Berthold Rittberger and Tim Büthe for their constructive comments and for steering the manuscript through the publication process. Finally, we gratefully acknowledge the support by Sarah Kluge, Margareta Maier, Bernadette Spitzer and Lukas Wörlein during the process of data collection.

Notes

1 Following our previous work (König et al., 2022, p. 11),

we understand AI systems as implementations of digital technology that are designed to interact with a given environment and to realize predefined goals by processing data in ways which allow the system to modify its behaviour (Russell & Norvig, 2016). The program that guides the behaviour of the system is informed by data fed into the system and patterns derived from this data.

2 Section one in the online supplemental material outlines the case selection strategy. Out of 51 states that belong to the EU, OECD or G20, 13 countries did not have any official AI strategy at the end of 2020. This changed for some of these countries, such as Great Britain, in the following months; however, data collection ended at the end of 2020 so that strategies passed after this date were not included. From the remaining 38 countries, we excluded those which did not have a standalone AI strategy or had passed only sector-specific ones. Some countries have unofficial AI strategies which are not passed by government bodies and are therefore excluded. We treat the EU as a sovereign entity which is a member of the OECD and part of the G-20; more importantly, it is a key actor that shapes the development, deployment, and governance of AI.

3 Some instruments that did not directly show in the data have been left out. One example are state-owned companies. It is contended that those companies played a big role in strategy making in Russia (Petrella et al., 2021). The actors making the strategy are not identical with the instruments in the strategy. The fact that the Russian strategy opted for self-regulation might benefit the interests of state-owned companies as they have been part of self-regulatory efforts in the past.

4 Our analysis only includes comprehensive national strategies, i.e., a government strategy which cover the full spectrum and not only focuses on a specific policy sector like industry or research and development. For more information which countries drop from our sample, visit section two in the online supplemental appendix. Why certain states have not passed such a strategy is beyond the scope of our analysis. One possible interpretation is that countries doing so take an intentional decision by following implicitly or explicitly a laissez-faire state approach of non-intervention. Other explanations could be low state capacity, a low level of AI development in a country, being a general laggard concerning the digital transformation, among others. For the interpretation of national AI strategies including only a few policy instruments, the fallback option of having no strategy at all should, however, be kept in mind.

5 We focus on key components of the expected instrument mix for each state type. This does not mean that, for example, a regulatory state might not also provide financial incentives – e.g., via regulated tariffs or allowed return to investment – or that a market-oriented state would not build institutions to establish or protect specific market mechanisms. However, we would argue that a high usage of these instruments is not an essential component of the respective state type.

6 The same applies to the pair of market-oriented and self-regulation-promoting states.

Additional information

Notes on contributors

Christian Djeffal

Christian Djeffal is Professor for Law, Science and Technology at Technical University of Munich (Germany). Markus B. Siewert is Senior Lecturer and Managing Director of the TUM Think Tank at the Munich School of Politics and Public Policy, Technical University of Munich (Germany). Stefan Wurster is Professor for Policy Analysis at the Munich School of Politics and Public Policy, Technical University of Munich (Germany).