On 23rd and 24th of June, I participated in the 2nd Conference on Large-scale AI Risks (LSAIR) @ KU Leuven.

Despite the heat, it was a very engaging and stimulating two-day event, hosted in the beautiful UNESCO World Heritage’s Grand Beguinage.

The conference gathered together scholars, policymakers, and consultants, providing a diverse spectrum of expertise, approaches and viewpoints on AI regulation, and its technical, legal, political, societal, psychological and ethical implications.

As part of a philosophical strand, in my talk titled “Reframing AI Large-scale Risks: A Quantum-inspired Approach to AI regulation“, I addressed AI as a technology emerging at the intersection of the digital and language ecologies intended as complex (and thus, emerging and “unprestatable”) systems. Such complexity calls for speculative-first, empirical-later regulatory approaches to AI, which can complement ex-ante, risk-based ones.

Below, the abstract; further below, my slides.

This paper takes a philosophical perspective to reframe the concept of “large-scale risks” of Artificial Intelligence (AI), based on an onto-epistemological interpretation of quantum physics (Barad 2007; Wendt 2015). The goal is to advance a quantum-inspired approach to the regulation of AI, which is complementary to (and more adaptive than) the risk-based approach adopted, for instance, by the EU with the AI Act. The paper is divided into three parts.

Part 1 highlights – based on literature – how the risk-based approach informing the AI Act is posited on an ex-ante linear extrapolation of “known and reasonably foreseeable risks”, which cannot properly estimate risk magnitudes, especially when it comes to regulating general-purpose AI (GPAI). An example is the classification of AI-powered chatbots as bearing “limited risk”, but ultimately having produced some dramatic consequences, such as suicides among youngsters (The Guardian 2026). While scholars (Novelli et al. 2024a, 2024b) have developed scenario-based assessment models that aim to mitigate the unpredictability of GPAI’s applications and impacts, especially on rights and principles that elude a proper quantification (Sass et al. 2026), in this paper it is contended that these attempts remain insufficient to effectively regulate GPAI.

Part 2 of the article motivates and expands on this position by borrowing from the fields of complex systems (Gershenson & Heylighen 2003) and biology (Kauffmann 2019) to advance the claim that GPAI’s “large-scale risks” are inherently “systemic” and “unprestatable”. Being part and parcel of sociotechnical ecosystems, the applications and impacts of GPAI are not only hard to predict but, more radically, they cannot be estimated through linear extrapolations (e.g., cause-effect links, cost-benefit analyses) because they do not abide to entailing laws. In other words, GPAI’s potential outcomes cannot be pre-stated, as they change over time with and through – among other factors – the advancement of GPAI itself. Hence, a different approach to GPAI large-scale risks is required.

A possible path forward discussed in Part 3 comes from the onto-epistemological tenets of quantum physics. Notably, the fundamental indeterminacy and entanglement at the core of quantum physics can prove powerful conceptual tools for overcoming linear extrapolations of large-scale risks and deterministic decision-making entailed by these. Calls in this direction have started to emerge (Calzati & de Kerckhove 2024; Renda 2025; Meckel et al 2025) which signal how the philosophical tenets of quantum physics can help reframe tech regulation in terms, for instance, of superposed values, non-local effects and trade-offs, and complementary scenarios. The paper concludes by discussing one example where the quantum concepts of indeterminacy and entanglement were operationalized in the course on “data ethics” designed and taught by the author, together with a colleague, at Delft University of Technology.

Here the slides