>>Legislation and Regulation of AI Systems
Regulating AI: An equation with too many unknowns
Disruptive innovation is always a great challenge for legislation. The "pacing problem" (Marchant et al., 2011), the fact that technology and law develop at uneven paces, reaches its ultimate point here, since AI develops exponentially. Still, social, economic, and legal systems change incrementally. The expression "law lag", found in the literature, vividly describes the laborious process of law keeping pace with the rapid progress of technology. The gap between AI technologies and legislation sometimes leads to the perception that traditional legislation cannot tackle this new challenge (Cerca et al., 2015) or that regulating AI is impossible. Thus, the notorious Collingridge dilemma for the social control of technology escalates drastically regarding AI.
The complex and opaque networks that have developed in the AI industry pose even greater difficulty in handling this problem. Furthermore, AI, which is established as a general-purpose technology, broadens and escalates the above problem, as it aims to cover undefinable fields and can be used in unpredictable ways with unforeseen social impacts. Moreover, it interferes with human experience and decision-making processes, affecting our cognitive function and awareness of its consequences.
How, then, could our conventional legal system solve an equation with so many unknowns?
General rules or a new legal framework?
First, we should address whether legislative intervention is required. It is argued that the law (general rules) is clear, competent, comprehensive, and complete, and that the emergence of new technologies does not present anything truly novel. The challenge lies in ensuring the proper enforcement of existing general legal rules when resolving emerging issues, rather than creating new rules. Frank Easterbrook, in the context of cyberlaw, advanced the notion that “the best way to learn the law applicable to specialised endeavours is to study general rules” (Easterbrook, 1996, p. 207). In this context, the Universal Declaration of Human Rights has been suggested as an appropriate framework for addressing the potential impacts of AI, as it has been adopted by many countries over the decades (Donahoe & Metzger, 2019). However, some argue that existing legislation provides limited protection and is inadequate to address the specific challenges that AI systems may present.
Assuming that a new legislative initiative is necessary, what is its appropriate form and content? In recent years, various legal systems have responded to this question in different ways. Guidelines, standardisation protocols, information-gathering mechanisms, and behaviour modification techniques based on ethical principles have been published or recommended. However, many concerns have been expressed regarding the capacity of AI Ethics (soft law) to protect citizens from violations of their rights, highlighting the danger of facilitating “ethics-washing” (N. Van Dijk et al, 2021). Moreover, many frameworks and models have been suggested in the literature regarding the regulation of AI. Among the proposed frameworks, the concept of dynamic and interactive regulation could fit better in the field of AI, as it is based on learning by doing and continuity of regulatory relationships (Kaal & Vermeulen, 2017).
The EU Answer
The European Union responded to the aforementioned challenge by taking many initiatives that led to the publication of soft law. The European Commission organised the High-Level Expert Group on AI, which published its Ethics Guidelines for Trustworthy AI (2019) and Policy and Investment Recommendations for Trustworthy AI (2019). In 2020, the Commission published a White Paper on AI (A European Approach to Excellence and Trust). The EU has decided to tackle this challenge by taking the next step, publishing a “regulation”, a legal text that has entered into force on a set date in all 27 Member States, and it will have binding legal force throughout the EU, meaning that natural or legal persons will be able to invoke those rights before national courts and tribunals.
According to the official announcements of European authorities, the AI Act (Regulation 2024/1689) is introduced as the first-ever comprehensive legal framework on AI worldwide, guaranteeing the health, safety and fundamental rights of people, and providing legal certainty to businesses across the 27 Member States. Although there have been other similar initiatives worldwide (e.g. the US National AI Initiative Act, which became law on 1 January 2021), the AI Act is a very influential starting point. Furthermore, it aims to place the EU in a position of “leadership by example” for the good governance of AI technologies and will extend the so-called Brussels effect (Brandford, 2020).
Its goal is to foster trustworthy, human-centric AI by classifying applications into risk levels (unacceptable, high, limited, and minimal) and banning some practices (e.g. social scoring and manipulative AI), while imposing strict requirements on others to ensure safety and protect fundamental rights.
This initiative to regulate AI will be added to the General Data Protection Regulation (GDPR), the Digital Services Act, and the Digital Markets Act, which aim to regulate online platforms and services, and are part of what is known as "digital constitutionalism," which describes the legislative stance of the EU to address the challenges of the emerging digital era.
Our position
Despite its ambitious goals, the AI Act consists of many vague terms that leave too much room for uncertainty and loopholes (L. Floridi, 2021). First off, it is unclear whether the taxonomy of AI systems, categorised as high-, low- or zero-risk, will work in practice. No AI system is going to introduce itself as a high-risk system aiming at social scoring and manipulation.
Second, and perhaps most importantly, the Act lacks comprehensive technical standards. It stresses the importance of human oversight, yet it does not demand understanding and technically describing the underlying technology. This raises the question of whether it is possible to successfully regulate AI applications while ignoring the underlying technology. In other words, legislative interventions require the ability to accurately define, categorise, and communicate what the underlying technology comprises (Hin-Yan Liu, 2023). This is a challenging project for lawmakers and audit bodies, since the method and the data utilised to train the system are complex and opaque.
Conceptual confusion and uncertainty about the underlying technology (which is partly reflected in the awkwardness of defining ΑΙ in legal texts) entail uncertainty about the specific risks involved in the design and deployment of AI and will affect the effectiveness of any conformity assessment and of the sanctions that should be imposed. If we treat the underlying technology as being akin to magic, it will be impossible to tackle the consequences of its deployment effectively. Crucially, treating AI as magic is incompatible with the core principles of our democratic societies, such as transparency, accountability and procedural fairness.
REFERENCES
Almeida P. et al, Artificial Intelligence Regulation: A framework for governance, 2021, Springer.
Bradford A., The Brussels effect: How the European Union rules the world, 2020. Oxford University Press.
Cerka P., Grigiene J., & Sirbikite G., 2015, Liability for damages caused by artificial intelligence. Computer Law & Security, Review, 31(3), 376–389.
Collingridge D., The social control of technology, 1981, Palgrave Macmillan.
Donahoe, E., & Metzger, M. M., 2019, Artificial intelligence and human rights. Journal of Democracy, 30(2), 115–126.
Easterbrook, F., Cyberspace and the Law of the Horse, 1996, University of Chicago Legal Forum, 207.
Floridi L., The European Legislation on AI: a Brief Analysis of its Philosophical Approach, Philosophy & Technology, 2021, 34:215–222, Springer.
Liu H., What makes AI regulation so difficult?, in Artificial Intelligence for human-centric society: The future is here, 2023, European Liberal Forum.
Marchant, G., Allenby, B., & Herkert, J. (Eds.), The Growing Gap Between Emerging Technologies and Legal-Ethical Oversight: The Pacing Problem, 2011, Springer
Rahwan, I., Society-in-the-loop: Programming the algorithmic social contract. Ethics and Information Technology., 2017, 20, 5–14.
Van Dijk N., S. Casiraghi & S. Gutwirth, “The Ethification” of ICT Governance. Artificial Intelligence and Data Protection in the European Union, Computer Law & Security Review, 43, 2021.