Redaelli (2024) characterizes the behavior of generative artificial intelligence as preter-intentional — it incorporates and transcends the intentions of designers and users while remaining dependent on them — and introduces the 'intentionality gap' to capture why it is increasingly difficult to trace AI system behavior back to human intentionality. He explicitly brackets the legal question this raises: whether liability should be attributed to artificial agents at all, and he offers no criteria for applying the concept to real systems or disputes. The problem is stated as: determine whether and how the preter-intentionality of generative AI should reshape responsibility and liability attribution — centrally, whether artificial agents themselves can or should bear liability (or quasi-responsibility) for preter-intentional outputs, or whether such attribution is instead blocked and redistributed entirely across human parties.

Background

Generative artificial intelligence — systems such as large language models (LLMs, neural networks trained on massive text corpora that produce novel outputs in response to prompts) and text-to-image models — can produce text, images, and code that no designer or user specifically intended. In the philosophy of technology, intentionality is the philosophical term for the 'aboutness' or directedness of mental states and actions toward objects and goals; whether technical artifacts can possess a form of it is a long-standing debate. Postphenomenology, the school founded by Don Ihde that studies how technologies mediate human experience, supplies a key precedent: Peter-Paul Verbeek's notion of composite intentionality, the idea that in human–technology relations human and technological intentionality merge into a hybrid whole, so that artifacts contribute their own 'intentionality' to outcomes. Against this background, Roberto Redaelli (2024) argues that generative AI exhibits what he calls the intentionality gap: (1) as AI systems become more autonomous, it is increasingly difficult to assign responsibility for their actions to humans, and (2) as systems move beyond good old-fashioned AI (GOFAI, the symbolic, hand-coded style of AI in which the reasoning behind a result is inspectable), it is increasingly complex to reconstruct the reasoning behind their outputs. His proposed response is terminological and conceptual: generative AI behavior should be called preter-intentional, borrowing a legal-philosophical term anchored in article 42 of the Italian Penal Code (which distinguishes intentional, preterintentional, and culpable commission of a crime; 'preterintentional' means beyond or additional to what is intended). Preter-intentionality is meant to capture that AI behavior goes beyond (Latin: praeter) human intentionality while remaining linked to and dependent on it. The proposal responds to the older responsibility-gap literature: Andreas Matthias (2004) argued that learning automata create outcomes attributable to neither manufacturer nor user, so that a responsibility gap emerges; Filippo Santoni de Sio and Giulio Mecacci (2021) refined this into four gaps (including a traceability gap — the impossibility of tracing an AI system's output back to identifiable human decisions) and surveyed response strategies (human oversight, strict liability on producers, regulation); David Gunkel (2020) pressed the deeper question of whether responsibility-talk even applies to machines; Bernd Carsten Stahl (2006) had already proposed ascribing quasi-responsibility — a form of machine answerability that does not require moral personhood — to computers. Kostas Terzidis, Filippo Fabrocini and Hyejin Lee (2023) gave the competing reading that AI 'intention' is an unintended consequence, which Redaelli explicitly rejects. What Redaelli's paper does not do — by his own statement in the Notes — is address the legal sphere: whether liability should be attributed to artificial agents, and, more broadly, how responsibility and accountability should be redistributed in practical ethical and legal frameworks when generative behavior is intentionally non-predetermined yet dependent on human design and use. The legal backdrop makes this timely rather than academic: the EU AI Act (2024) imposes risk-tiered duties on providers and deployers, the proposed EU AI Liability Directive was withdrawn in February 2025, and civil liability for AI outputs is consequently fragmented across national regimes; existing doctrinal tools (fault-based liability; strict liability, imposed without proof of fault; product liability for defective goods; vicarious liability, under which one party answers for the acts of another) all presuppose human or corporate agents.

Problem Statement

Determine whether, and how, the preter-intentionality of generative AI — behavior that incorporates and transcends the intentions of designers and users while remaining dependent on them — should reshape the attribution of responsibility and liability. The central question is the one Redaelli explicitly brackets in the Notes to his paper: (1) decide whether liability (or a Stahl-style quasi-responsibility, a form of machine answerability that does not require moral personhood) can and should be attributed to artificial agents whose outputs are preter-intentional, or whether preter-intentionality is normatively inert for liability attribution and instead rules out such attribution, redistributing it entirely across human parties. Two related open directions are motivated by the source: (2) how responsibility and accountability should be redistributed in practical ethical and legal frameworks among developers, deployers, and users, given that generative behavior is intentionally non-predetermined yet dependent on human design and use — Redaelli describes the responsibility gap as the difficulty of finding a criterion for the redistribution of responsibility between human agents and AI systems; and (3) how preter-intentional behavior might be identified in concrete generative-AI cases and disputes (for instance, a defamatory hallucination — a fabricated, reputation-damaging statement presented as fact by a model — a copyright-infringing output, or a harmful chatbot induction).

References

  1. Redaelli R (2024) Intentionality gap and preter-intentionality in generative artificial intelligence. AI & Society 40(4):2525–2532. https://doi.org/10.1007/s00146-024-02007-w
  2. Verbeek PP (2011) Moralizing technology: understanding and designing the morality of things. University of Chicago Press. https://press.uchicago.edu/ucp/books/book/chicago/M/bo11309162.html
  3. Matthias A (2004) The responsibility gap: ascribing responsibility for the actions of learning automata. Ethics and Information Technology 6(3):175–183. https://doi.org/10.1007/s10676-004-3422-1
  4. Santoni de Sio F, Mecacci G (2021) Four responsibility gaps with artificial intelligence: why they matter and how to address them. Philosophy & Technology 34(4):1057–1084. https://doi.org/10.1007/s13347-021-00450-x
  5. Gunkel DJ (2020) Mind the gap: responsible robotics and the problem of responsibility. Ethics and Information Technology 22(4):307–320. https://doi.org/10.1007/s10676-017-9428-2
  6. Stahl BC (2006) Responsible computers? A case for ascribing quasi-responsibility to computers independent of personhood or agency. Ethics and Information Technology 8(4):205–213. https://doi.org/10.1007/s10676-006-9112-4
  7. Terzidis K, Fabrocini F, Lee H (2023) Unintentional intentionality: art and design in the age of artificial intelligence. AI & Society 38(4):1715–1724. https://doi.org/10.1007/s00146-021-01378-8
  8. Faroldi FLG (2021) Considerazioni filosofiche sullo statuto normativo di agenti artificiali superintelligenti. Revista Iustitia. https://unipv.unifind.cineca.it/resource/item/510071
  9. Sarosiek A (2025) Regulatory intentionality in artificial systems. Proceedings of the 1st International Online Conference of the Journal Philosophies (MDPI Proceedings 126(1):16). https://doi.org/10.3390/proceedings2025126016
  10. Naumov P, Tao J (2025) Responsibility gap in collective decision making. Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence (IJCAI-25), pp. 4606–4614. https://doi.org/10.24963/ijcai.2025/513
  11. Brandão Bárrios L (2024) Are autonomous acts of artificial intelligence civilly responsible? In: Artificial Intelligence and Legal Responsibility. Iberojur Science Press, pp. 64–75. https://doi.org/10.62140/lbb642024
  12. Nannini L (2025) When less regulation means more complexity: the EU AI Liability Directive withdrawal and its impact on European technological competitiveness. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society (AIES '25) 8(2):1848–1861. https://doi.org/10.1609/aies.v8i2.36679