In recent years, AI-based generative systems have produced creative outputs that are increasingly difficult to distinguish from those of human artists. Cases in which such outputs become objects of institutional endorsement and controversy-competition...
In recent years, AI-based generative systems have produced creative outputs that are increasingly difficult to distinguish from those of human artists. Cases in which such outputs become objects of institutional endorsement and controversy-competition awards, copyright disputes, and transactions in the art market-are rapidly accumulating. This development raises questions that go beyond the technical issue of whether AI can generate creative products. It instead presses a set of foundational normative questions: who should count as the agent of artistic creation, and under what conditions praise and blame-as well as the attribution of responsibility-can be justified. Rather than confining the debate to a binary choice between AI as a mere tool and AI as a creative agent, this dissertation investigates the affirmative possibility of recognizing AI as a creative agent, by analyzing the philosophical conditions under which creative action can be responsibly attributed and the extent to which AI can satisfy those conditions. To do so, it adopts the debate on free will as its primary framework, since that debate has long examined the conditions of action-attribution and responsibility.
Chapter I clarifies the concept of agency presupposed in dominant discussions of creativity and argues that standard approaches-taxonomies of creativity, cognitive-scientific models, and neuroscientific explanations-remain insufficient to normatively justify AI creative agency. While these approaches can provide descriptive accounts of creative processes, they do not supply criteria for deciding whose action an output should count as, or who is an appropriate target of praise and blame, reward and sanction. The dissertation therefore reframes the problem of AI creative agency in terms of action attribution and the conditions of responsibility, and argues that the free-will debate provides philosophically fruitful resources for analyzing AI-generated artistic action. It then offers an overview of major positions in the free-will debate-libertarianism, determinism, and compatibilism. On this basis, it rejects libertarianism insofar as it explains creation and responsibility by appeal to a transcendent self-origination (causa sui), thereby narrowing the criteria required for real artistic practice and institutional design. It instead adopts compatibilism as the basic standpoint for analyzing creative agency, since compatibilist accounts aim to explain and secure substantive freedom and responsibility even in a deterministic world.
Chapter II analyzes the capacity conditions of free will with a focus on Daniel Dennett’s compatibilism. Dennett famously reconstructs free will as the “free will worth wanting,” rejecting the demand for a transcendent self-origination and emphasizing a naturalistic account of agency. On this view, freedom is not a mysterious metaphysical power but a complex of capacities-centrally including self-control and reasons-responsiveness-that enable agents to govern their conduct in socially and institutionally embedded contexts. Dennett also emphasizes the gradual, developmental character of agency through bootstrapping, a process by which agents “bootstrap” themselves into increasingly robust forms of responsibility and self-governance. Drawing on this framework, the dissertation examines skeptical challenges often posed to compatibilism, including the manipulation argument and the problem of luck, and argues that Dennett’s compatibilism can still provide a substantive account of an agent’s freedom and responsibility within deterministic causal structures.
The dissertation then applies these capacity conditions to AI creative systems. It argues that current AI can implement limited self-control insofar as it can modulate output tendencies through feedback and interaction. It further contends that, with respect to reasons-responsiveness, AI can partially instantiate functional structures that register normatively relevant considerations and adjust outputs accordingly. From the perspective of Dennettian bootstrapping, AI autonomy may be understood not as a fixed property but as a practical capacity trained and developed within sociocultural contexts. Nonetheless, at the present stage, AI lacks robust metacognitive self-monitoring and faces clear limits as an agent that can autonomously form and articulate normative reasons. Even so, the dissertation argues that where self-control and reasons-responsiveness are structurally implemented above a certain threshold, there is room to recognize partial (rather than full) creative agency in AI-generated artistic action.
Chapter III extends the free-will discussion to the problem of desert and responsibility. According to Dennett, responsibility is not justified by basic desert-the idea that an agent deserves praise or blame simply in virtue of having performed the action-but rather by forward-looking functions that sustain trust, cooperation, and the practical regulation of social life within moral, legal, and institutional practices. Building on Dennett’s notion of non-basic desert and its forward-looking justification, the dissertation analyzes how praise and blame, reward and sanction can be justified in artistic contexts. Through concrete cases-competition awards, copyright registration and cancellation, and disputes over stylistic appropriation-it shows how the single premise that “AI is merely a tool” generates institutional inconsistency and responsibility gaps. It then proposes that a more coherent approach is to recognize AI creative agency as partial and conditional, depending on the scope and level to which relevant capacity conditions are satisfied, and to attribute credit and responsibility within that bounded range in ways that are practically and institutionally more consistent.
Finally, in the latter part of Chapter III, the dissertation examines the free-will skeptic’s concept of answerability as a complementary device. Free-will skepticism rejects responsibility grounded in basic desert and reconceives responsibility as a dialogical structure consisting of demands for reasons, responses to those demands, and subsequent normative improvement. On the influential forward-looking articulation of answerability, responsibility practices are grounded not in desert but in three non-desert-invoking moral desiderata: protection, reconciliation, and moral formation. This forward-looking answerability-responsibility offers practical insight by showing that, even without granting AI metaphysical standing, AI can be incorporated into practices of explanation, justification, and correction, thereby sustaining responsibility practices. However, while acknowledging the significance of an answerability-based practical model of responsibility, the dissertation argues that a wholesale abandonment of desert discourse risks weakening the norm-forming role of praise and blame in artistic communities and leaves little room to treat AI as an agent to whom creative credit can be attributed.
In conclusion, this dissertation retains compatibilism as the basic framework for explaining AI agency, while proposing a hybrid approach for transitional circumstances: depending on the extent to which AI satisfies relevant capacity conditions, the skeptic’s concept of answerability should be introduced as a supplementary device for responsibility allocation and institutional design. This proposal is grounded in the judgment that current AI does not yet sufficiently satisfy the capacity conditions of self-control and reasons-responsiveness. Nevertheless, if AI systems gradually develop to satisfy capacity conditions comparable to those of human creators, there may be greater room to attribute responsibility to AI on the basis of non-basic desert. Ultimately, the central issue is not the absolute metaphysical question of whether AI truly has free will, but the normative question of under what capacity conditions and within what institutional contexts it is justified to treat AI as a creative agent-a question that should stand at the center of contemporary art institutions and policy design.