Conference Agenda

Agile methods of regulation and civil society participation are increasingly identified as institutional mechanisms that may assist in addressing the governance challenges associated with artificial intelligence (AI). These considerations are frequently addressed at a theoretical level and often examined independently. This paper integrates these perspectives to explore how agility and civil society participation are beginning to be organized and how they may contribute to AI governance.

We begin the discussion by elucidating the concept of AI governance. Subsequently, we present an integrated perspective suggesting that incorporating agility and civil society engagement can effectively address some of the complexities of AI governance, particularly those arising from significant uncertainty during the early stages of technology development and the widening gap between fast-paced technological innovation and slow governance responses (the “pacing problem”). We also identify limitations and shortcomings of both agile governance and civil society participation.

The term AI governance is challenging, if not impossible, to operationalize at a general level. We use concepts of innovation economics, business ecosystem governance, and policy analysis to identify specific, limited instances where appropriate governance measures promise to better align outcomes with notions of social welfare or human flourishing. We argue that traditional concepts of externalities, public goods, and competitive choke points provide robust guidance to identify many of these, often local, cases. Where new insights are needed are discussions on the desirable direction of innovation and whether governance should influence it.

One particular challenge of identifying the need for governance and appropriate responses is the high level of uncertainty under which emerging technologies evolve initially. Risk-based approaches to governance fail at that stage of technology development because risks and opportunities are not sufficiently known. Applying the precautionary principle will often result in overly stringent regulation. In contrast, a fully unregulated system may develop in irreversible, undesirable directions. This dilemma, that both too little and too much governance may result in lower rates or unwanted directions of innovation, is a key insight of recent innovation research.

Agility and civil society involvement can contribute to avoiding these outcomes by enabling adaptive learning to establish guardrails and dynamic responses to problems. We explore the advantages and shortcomings of these approaches based on a thorough review of the research literature and policy documents. Agile governance refers to institutional arrangements and instruments that can adapt more flexibly and faster than traditional forms of government policy, such as legislation and regulation. Proposals include “soft law”, regulatory sandboxes, outcome-based regulation, or data-driven governance. All these approaches are informed by evidence and rely on informational feedback loops that enable learning about intended and unintended consequences, implicated values, and on the workability of governance interventions.

Civil society organizations, in our paper broadly understood as “an institutional realm of private associations, voluntarism, and active citizens” (Torfing, 2020) that exists independently of government and private corporate can make important contributions to such agile governance methods. Among other contributions, civil society organizations bring knowledge and values to governance discussions that traditional government and corporate actors may not consider. Such organizations may represent groups that are harmed by AI, they may advocate for vulnerable groups, enhance accountability, and ensure that AI development remains aligned with human rights and social values. They also may mitigate concerns about the capture of agile regulatory arrangements by corporate and government players.

As part of our research, we conducted a thorough review of policy reports and already established policy frameworks on how they envisage inclusion of civil society organizations. The paper develops a typology of civil society involvement at the global, national, and local levels. Civil society organizations are characterized by their origin (e.g., existing information policy organizations that expanded into AI, newly formed initiatives), by their main contributions (often helping to shape a shared policy vision), their funding sources, and the formal institutional arrangements that guide their contributions to AI governance. Several national and international bodies have established processes that give civil society organizations voice.

We conclude the paper with a review of the limitations and shortcomings of civil society participation and agility in AI governance. Among other limitations, agile governance methods are critically dependent on effective monitoring and feedback mechanisms, which often are not implemented properly; civil society faces resource constraints and may also be captured by special interests.

The novel contributions of the paper include the development of a differentiated approach to AI governance and an in-depth exploration of how agile institutional designs and civil society participation can contribute to mitigating challenges created by uncertainty and the pacing problem.