This summer, two open-access academic articles were published regarding authorship in the era of generative artificial intelligence. One approaches the subject through the philosophy and genealogy of authorship. The other compares results produced by generative systems with human interpretations in the visual and performing arts. Read together, they propose a useful distinction: the creative process may be shared among many people, tools, models, and files; accountability for the work is not shared in the same manner.
They do not say exactly the same thing. Nor do they resolve the legal question of who obtains copyright over a specific work. Their convergence lies elsewhere: they reject the notion that a generative system should be treated as an autonomous creative agent and place judgement, selection, and accountability upon the individuals involved.
Michael Uebel and Bilal Hamamra published 'Authorship After Generative AI: Distributed Creativity and Relational Responsibility' in Philosophy & Technology on 12 August. The article questions the romantic myth of the sovereign and self-sufficient author, though not to declare the end of authorship. Their argument is that creativity has always depended on infrastructures, archives, and distributed labour that the figure of the solitary author tends to obscure.
From this premise, the authors argue that speaking of "AI authorship" confuses categories. A statistical system may intervene decisively in a production chain, yet it cannot occupy a position of accountability, face a claim, or justify its decisions. Nor does it take part in a relationship of vulnerability or dialogue of the kind the authors associate with responsibility.
Their proposal is to understand authorship as the ethical and epistemic stewardship of a hybrid writing system. This concept does not return all merit to an isolated individual. It acknowledges that there is distributed agency and technical opacity, but it demands that there be identifiable individuals who are accountable for the results mediated by artificial intelligence.
This formulation has a practical consequence. If a work depends on models trained with third-party materials, platforms, instructions, selection, editing, and the context of publication, acknowledging this network does not make it possible to conceal the person who decides to use it. Distributing creativity is not equivalent to dissolving accountability.
Tsehaye Haidemariam published 'After generative AI: authorship, labour, and cultural governance' in AI & Society on 29 June. The article adopts an empirical approach and compares results from systems such as Stable Diffusion and Sora with human interpretations in visual and performative fields.
The study combines methods of semantic analysis, latent traits, inverse image similarity, and motion tracking to observe how originality, authorship, and bodily expressiveness are organised. According to its results, the models achieve high fidelity to the brief and a recognisable stylistic coherence, yet they remain conditioned by the patterns of the training data and by probabilistic recombination. Human interpretations exhibit greater variation, responsiveness, and affective nuance.
The central finding does not consist of measuring what portion of the result belongs to the person and what to the machine. The article locates the relevant difference in the work that links the operations: iterating, selecting, evaluating, and providing a framework for the result. Creation appears as a distributed flow, but human judgement performs a constitutive function within it.
Hence, Haidemariam describes generative systems as components that extend human work, not as autonomous creative agents. The conclusion does not assert that every human intervention suffices to produce an original work, nor that any generated result merits legal protection. It asserts something more constrained: in the cases analysed, creative value is organised through human decisions within a process shared with algorithmic tools.
In my view, the two articles distribute creativity and concentrate accountability, but they do so from different planes.
Uebel and Hamamra formulate a normative requirement: someone must be able to be held accountable for what is produced. Haidemariam offers an empirical description: in the cases studied, significant differences appear in the iteration, selection, evaluation, and human interpretation. The former explains why accountability cannot be delegated to a statistical system. The latter shows where human work is observed when production also depends on a model.
The two approaches have distinct limitations. The philosophical thesis identifies who must remain available to be held accountable, yet it does not, by itself, allow for the measurement of the degree of creative control a person exercised over a specific work. The empirical study analyses specific systems, materials, and metrics; it does not establish a universal rule for every technology or for all artistic practices. Its results depend upon what is being compared, how the test is constructed, and what is understood by variation, response, or nuance.
Read together, they serve as a useful corrective. Responsibility cannot be deduced from an experimental score. Furthermore, creative attribution should not be determined by disregarding what actually occurs during the work process.
One must not convert a philosophical thesis into a legal rule. Being responsible for publishing a work does not necessarily mean being its author for the purposes of intellectual property. Similarly, intervening in a process, writing instructions, or selecting an outcome does not, in itself, guarantee that a protected work exists or that the human contribution reaches the required level of originality.
In Spain, article 5 of the Intellectual Property Law considers the natural person who creates a literary, artistic, or scientific work to be the author. The European Union does not have a specific, closed rule for all content generated with AI, but the current framework and European case law continue to base protection on human creativity. The difficult question is not whether the system can be listed as an author, but whether a person's decisions are sufficiently expressed in the specific result.
The reviewed articles provide criteria for considering this question, rather than an automatic legal answer. Ethical custody, human judgement, and the capacity to be held accountable may coincide with legal authorship, but they are not synonymous. There may also be editorial, professional, or institutional responsibility for content that does not attain protection as a work.
This distinction avoids two shortcuts. The first consists of calling the machine an author because it participated in the production. The second involves attributing the entire work to the person who pressed the button without examining what decisions they made, what control they exercised, and what was expressed in the result.
For those who create with these tools, the practical consequence is not to draft a technological confession or to turn the work into a technical report. It is to be able to explain, when relevant, what function each part of the process fulfilled.
This may include:
- Which system and, if known, which version was used.
- What materials the person provided and under what rights.
- What role the instructions, references, or input images played.
- How many phases of generation, discarding, and editing occurred.
- What formal and conceptual decisions remained under human control.
- What transformations were performed outside the system.
- What limitations, errors, or uncertainties were detected.
- Who approved and published the final result.
Not all of this data must always accompany the work. However, it should be possible to reconstruct it when it affects attribution, research, a call for submissions, conservation, the employment relationship, or the rights of third parties. Useful transparency does not consist of listing every instruction given to a machine, but rather in documenting the decisions that allow one to understand where the work originates and who is responsible for it.
The two articles leave a demanding conclusion. Recognising that creativity was never completely individual does not force one to accept authorship without a subject. On the contrary: the more distributed and opaque the process, the more important it becomes to identify the individuals who select, justify, and assume its consequences.
About the open conversation
This review distinguishes the arguments made in the two articles from my joint reading of them and from the additional legal clarification. Both texts are open access. If you have read them, work with generative systems or document hybrid creative processes and would like to discuss this interpretation, write to me.
Sources
Repository associated with the article by Tsehaye Haidemariam.
Real Decreto Legislativo 1/1996, de 12 de abril, texto refundido de la Ley de Propiedad Intelectual, artículo 5 (Spanish Intellectual Property Act, Article 5).