A digital constitutionalism framework for AI: embedding constitutive and limitative rules into AI architectures

In the last few years, there has been a growing awareness across all stakeholders involved in the development and implementation of AI that the full potential of this technology can only be realized through aligning AI design and usage with legitimate social values (Floridi et al. 2019). As a result, we have witnessed a flourish of initiatives setting ethical codes and good governance principles for AI development (Fjeld et al. 2020), which nonetheless seem unable to fill the "principle-to-practice" gap (Mittelstadt 2019, Morley et al. 2021), raising doubts about being mere "ethical washing" initiatives (Greene 2019). This paper argues that a digital constitutionalism approach constitutes a more effective way to developing a trustworthy and human-centric framework for AI governance, providing an already widely recognized set of standards supported by national and international institutions (Yeung et al., 2019). However, a traditional human rights approach may result ineffective because most of the violations and harms to people integrity and autonomy occur at an opaque technological layer of governance, outside of public awareness and scrutiny (Lessig 2006, Musiani et al. 2015). Thus, this paper propose a suitable digital constitutionalism framework for developing a trustworthy and human-centric framework for AI governance, moving from Gunther Teubner’s Societal Constitutionalism approach (2012). It will be argued that: i) fundamental rights could be conceived as counter-institutions contrasting the expansionistic and harmful tendencies of digitalization; ii) to be effective fundamental rights must be embedded into the socio-technical architecture of digital systems; iii) the constitutionalization of digital system occur when fundamental rights give rise to both constitutive rules that allow the deployment of digital potential and limitative rules setting the borders of digital systems; iv) the constitutionalization of digital systems is a hybrid process involving different actors, normative sources and institutional logics, such as international organization, governments, standardization bodies, technical communities, with different roles and responsibilities. The paper then illustrates how a digital constitutionalism process is actually taking place in the AI field. It will be discussed how existing human rights standards apply to artificial intelligence, an review ad hoc initiatives by international organizations (UN, UNESCO, CoE, OECD, etc.) The study then analyzes how this human-centric normative framework is being transposed into: i) legislation (such as the European Artificial Intelligence Act); ii) international standards (such as the IEEE p700 series, or ISO/IEC Joint Technical Committee SC 42); iii) operative technical and organizational arrangements. The analysis outline a framework in which international human rights law and international organization define high-level principles, governments define less abstract requirements and accountability mechanisms, while technical communities and standard setting organization translate the latte former into operationalizable technical specification and organizational governance mechanisms. Finally, this work warns about observable risks of mere technological solutionism, poor accountability and special interest capture, in this process and outlines some recommendations on how to increase the degree on integration and coordination between the different components of the AI ecosystem to strengthen the implementation of effective constitutive and limitative rules.