The central aim of this book was to provide an accessible and comprehensive overview of the legal, ethical and policy implications of artificial intelligence and algorithmic systems more broadly. As the various chapters in this book have shown, these technologies have a significant and growing impact on all domains of our lives, which makes it increasingly important to map, understand, and assess the challenges and opportunities they raise. This calls for an interdisciplinary approach, which is why this book brought together contributions from different disciplines with the goal of advancing our understanding of AI’s societal impact, as well as examining how the current legal framework deals with this impact, and where it falls short in protecting the core values of our societies. To conclude this book, I first wish to take a step back and highlight some of themes that were common across its chapters. Second, I want to draw attention to a few gaps that require further exploration in the future.

An important thread across this book was the observation that AI and algorithmic systems do not exist in a legal vacuum. In particular, Parts II and III of the book provided an overview of the prevailing normative landscape, looking at both crosscutting legal areas and sectoral domains. While this resulted in a description of a patchwork of rules that apply to the technology’s development and use, sometimes even with tensions or inconsistencies across domains, it simultaneously showcased the possibility of mobilizing existing legal norms toward better protection against AI’s adverse effects. Arguably, this possibility is too easily overlooked in today’s public debate, as all eyes are on the European Union’s new AI Act. Laws that were enacted prior to the advent of AI may not be perfectly tailored to its idiosyncrasies and the way in which it causes harm, yet they still have an important role to play, and should be invoked and enforced to make most out of their protective provisions.

At the same time, practically all authors identified certain shortcomings in the legal framework, especially when it comes to the protection of individual rights. In some instances, the substantive gaps within the – often convoluted – existing legal patchwork were the culprit, while in other instances, the authors instead hinted at a lack of enforcement of existing rules, which is no less of a concern. The new AI Act is frequently invoked as an attempt to address the former, but it is not free from criticism either, and the effectiveness of its enforcement mechanism over the longer term is far from evident in light of the questionable regulatory architecture it adopted. While this does not mean that we should despair at the state of Europe’s legal framework for AI – there is, in fact, much to be hopeful about in light of the EU’s (albeit imperfect) attempt to protect human rights against AI’s risks – it does underline the fact that we should not put all our hope on new or better legislation, or on legislation altogether. The AI Act has its limits, and other regulatory frameworks can complement its provisions, but they have their limits too. No law is perfect, as it is always a compromise solution that must meet the political reality, and that has inherent tensions between the demands of generality and specificity, and between predictability and flexibility.

Rather than losing faith in the power of the law altogether, this should instead incite us to also consider how law complements and interacts with solutions outside the legal realm. Several contributions for instance made the case for more AI literacy and education, responsibility-taking, awareness raising, public deliberation, and the encouragement of critical engagement more generally. Part I of this book has also shown how philosophical and ethical reflections on AI can inspire, inform and contribute to those aims, as well as strengthening the normative underpinnings of the existing laws. Indeed, the study of other disciplines – covering not only ethics and philosophy, but also sociology, anthropology, psychology, economics, cognitive science, and many others – are prerequisites for any sound critique and improvement of the normative framework that governs AI as a technology, its impact on society, and especially the human action involved in it.

In addition, they can also enable us to reflect on our approach to AI more generally, and the ease with which we seem to be ready to relinquish our agency in setting out the role and functions we want this technology to have in our society. The idea that AI overcomes us, and that all we can do as a society is run behind the facts and try to come up with some imperfect legal rules to control it, is both erroneous and dangerous, as it unwittingly deprives us from our individual, collective and societal self-determination vis-à-vis technology, which remains a human-made and human-caused object. Reclaiming this agency is important, but it requires a more nuanced understanding and debate about AI’s short and long-term societal effects, to which this book aspired to contribute.

There are, however, also some themes that fall outside the scope of this book, and that require further exploration in light of their importance for the AI governance debate. Let me point out two in particular. The first pertains to the jurisdictional area of attention. Indeed, when examining AI through the lens of law, ethics and policy, and assessing whether new or updated norms and theories are warranted, it is important to ask at which jurisdictional level those updates should take place. This book is heavily Europe-focused, and primarily analyses the normative framework of AI in the EU legal order. In this regard, this book is not an exception, as many contributions in the sphere of AI governance have a markedly Western perspective. In fact, most AI ethics guidelines originate from the US and Europe,Footnote ¹ despite the fact that these jurisdictions represent less than 20% of the world population.

Admittedly, the EU was the first to adopt a cross-sectoral AI-specific regulation, and could hence arguably offer important insights for the AI governance debate (though other jurisdictions, such as China in particular, already adopted domain-specific AI rules prior to the AI Act’s adoption). It is, however, rather evident that governance solutions which might (theoretically) work for Europe, will not necessarily work in other parts of the world – nor should we assume this to be the case. As noted elsewhere, “the economic, social, legal and political situation of countries strongly differ. Hence, the manner in which countries will be affected by AI – both in positive and negative ways – will inevitably differ as well.”Footnote ²

Already today, it is extremely clear that the benefits of AI are unevenly distributed not only within states, but also among states. Nevertheless, voices of the global majority are still given less space and weight, despite the alleged “universality” of the global AI governance discussion, and despite the (direct and indirect) extraterritorial reach of the EU’s standards.Footnote ³ While this book does not pretend to adopt a universal perspective to the challenges that AI raises, it remains important to acknowledge the virtues as well as the limitations of examining those challenges from a European lens, and to assert the importance for readers to complement contributions like this book with the study of works that illuminate perspectives beyond the Global North.

The second aspect that this book does not discuss, but that may be useful to reflect upon, pertains to the role of international organizations in the governance of AI. Not only national and supranational entities (such as the EU), but also international organizations have increasingly been awarded with a mandate to develop policies (and even laws) to guide AI-related human behavior toward one that is “ethical” and “trustworthy.” Some of these developments are briefly touched upon in this book’s contribution – from the Council of Europe’s new Convention on AI, human rights, democracy and the rule of law, to the UN’s attempts at governing lethal autonomous weapons, and UNESCO’s role in the ethics of AI in education. However, a systematic analysis of the practical and political impact of those organizations in governing AI, and the way in which they succeed or fail at fostering (more universal and potentially inclusive) norms to protect societal values, would be an important complementary contribution to consider in this context.

Finally, let me close with a remark that may sound obvious, but which has important repercussions that merit being pointed out nevertheless. AI as a technological phenomenon is continuously in flux, and so are the law, ethics and policy that relate to it. And while we can to some extent identify and anticipate the various risks and harms that are associated with its irresponsible development and use, there are undoubtedly also problems as well as benefits that we do not yet grasp or that we cannot yet predict. The launch of AI governance debates and initiatives at several jurisdictional levels can certainly be helpful in discovering, framing and analyzing the concerns we wish to address and the advantages we wish to gain, but we must equally be mindful of the fact that there is much information we do not have, and that there are also unknown unknowns, especially over the longer term.

It is therefore essential that we clarify, especially for ourselves, which values we wish to hold on to amidst the potential changes that societies might undergo, whether impelled by technology or by other developments – and it is precisely there that law, ethics and policy play a vital role. This role will, however, only come to fruition in its fullest force when imbued with the openness to keep on listening to and learning from other disciplines and jurisdictions, and when accompanied by an appropriate level of (regulatory) humbleness.