How Should Public Administrations Foster the Ethical Development and Use of Artificial Intelligence? A Review of Proposals for Developing Governance of AI

Abstract

Recent advances in AI raise questions about its social impacts and implementation. In response, governments and public administrations seek to develop adequate governance frameworks to mitigate risks and maximize the potential of AI development and use. Such work largely deals with questions of how challenges and risks should be managed, which values and goals should be pursued, and through which institutional mechanisms and principles these goals could be achieved. In this paper, we conduct a systematic review of the existing literature on the development of AI governance for public administration. The article describes principles and means by which public administrations could guide and steer AI developers and users in adopting ethical and responsible practices. The reviewed literature indicates a need for public administrations to move away from top-down hierarchical governance principles and adopt forms of inclusive policy-making to ensure the actionability of ethical and responsibility principles in the successful governance of AI development and use. By combining the results, we propose a CIIA (Comprehensive, Inclusive, Institutionalized, and Actionable) framework that integrates the key aspects of the proposed development solutions into an ideal typical and comprehensive model for AI governance.

Governance of Responsible AI: From Ethical Guidelines to Cooperative Policies

“…Therefore, the research contributions and suggestions on public AI governance have proliferated during the last few years (Cath, 2018;Cath et al, 2018;e.g., Taeihagh, 2021), with the role of public sector being under particular scrutiny (de Sousa et al, 2019;Kuziemski and Misuraca, 2020;e.g., Zuiderwijk et al, 2021). It is possible to divide these contributions on public AI governance into three categories (see Sigfrids et al, 2022).…”

“…In others, all AI systems should be subjected to relatively strict regulation and enforcement (Wallach and Marchant, 2018;Yeung et al, 2019). Overall, the contributions in this first category are concerned with the societal impacts of AI and see that negative impacts should be avoided with informed policy, which is supported among others by impact assessments and stakeholder consultations (Sigfrids et al, 2022).…”

“…It has been also suggested that a specific regulatory agency or relevant administrative bodies should have responsibility of approval of algorithms, supervision of AI developers, impact assessments, as well as certification and testing of algorithms (Bannister and Connolly, 2020;Dignam, 2020). In general, the contributions of this category argue that governance should support the operationalization of principles of good governance and AI ethics (Sigfrids et al, 2022).…”

“…Furthermore, apart from the current legislation that to some extent concerns AI and steers its implementation (e.g., GDPR and the general principles of good public governance), recently proposed legislation on AI 1 , and some program or organization based initiatives to our knowledge, there is yet no systematic empirical or real-life experiment on putting these suggestions systematically and comprehensively into practice. Thus, for the time being, the suggested models have been more or less hypothetical, based on rational scrutiny and aimed at facilitating policy and AI governance development (Sigfrids et al, 2022).…”

“…Furthermore, their fuzziness together with an overproduction of overlapping documents has raised some doubts about the real intentions behind their adoption in terms of ethics washing (Bietti, 2020). While there are also suggestions to operationalize ethical principles (Yeung et al, 2019;de Almeida et al, 2021;Stix, 2021;Sigfrids et al, 2022), to our knowledge, there are yet no empirical evidence, pilots or experimentations on suggested solutions.…”