In light of the Copyright Office’s refusal to register Dr. Stephen Thaler’s AI-generated artwork, holding that DABUS cannot be recognised as its author, Vikram Raj Nanda and Sumukhi Subramanian unpack the decision. They examine what it gets right on originality, authorship and ownership, and where its reasoning on human creative control and AI disclosure leaves questions unanswered. Vikram is a fourth-year student at National Law School of India University, Bengaluru, with a keen interest in IP law, Competition Law, and Arbitration. Sumukhi is also a fourth-year student at National Law School of India University, Bengaluru, and is the current Editor-in-Chief at the Indian Journal of Law and Technology (IJLT), NLSIU.
Originality Without Authorship: Analysing the Copyright Office’s Decision in Thaler
By Vikram Raj Nanda and Sumukhi Subramanian
In an interesting development surrounding the evolving debate on AI-generated works and copyright, the Copyright Office has refused to grant copyright for Dr. Stephen Thaler’s application for registration of the artwork titled ‘A Recent Entrance to Paradise’, generated autonomously by his AI system, DABUS. This decision has been handed down after the Delhi High Court’s direction to the Copyright Office to decide this matter (for context, see the earlier post here). In this post, we shall discuss what the Copyright Office gets right and what it fails to address.
Brief Overview of the Case
The applicant, Dr. Thaler, had listed the AI system, DABUS, as the ‘author’ of the artistic work, and claimed that he was the owner. He claimed that while he was responsible for the creation of the AI system, it was the latter that was autonomously responsible for the generation of the artwork, without any human interference, and hence remained the legally recognisable ‘author’ of the work.
Interestingly, the Copyright Office explicitly granted Dr Thaler the opportunity to amend his plaint and list himself as the author of the work in question, perhaps to indicate that it might help his application (this shall become relevant below). However, he refused to do so, continuing with his position that DABUS was the author of the artwork. As an alternative claim, the applicant argued that even if he were to be listed as an author of the work in question, the Registrar may, in exercise of his discretion, list in the ‘remarks’ column of the Register or in the Certificate of Registration that the artwork was generated by DABUS.
Summarily described, the Registrar’s decision proceeds through four principal inquiries:
– First, it distinguishes the questions of originality, authorship and ownership, recognising that these constitute analytically distinct stages in determining copyrightability.
– Second, it considers the originality of the impugned artwork and whether the absence of real-time human intervention affects the application of the originality standard.
– Third, it examines authorship and ownership, holding that DABUS, not being a recognised juristic person, cannot be treated as the author, while Dr. Thaler, as the person who ‘caused the work to be created’, may be recognised as such. It further holds that the claim that DABUS was the author while the applicant was the owner is inconsistent with Sections 17-19 of the Act.
– Finally, the Registrar considers and rejects the applicant’s alternative request to record, in the remarks column, that the work was generated by DABUS. We examine each of these aspects in turn.
Originality, Authorship and Ownership
Significantly, the Copyright Office has expressly recognised the sequential inquiry at the heart of copyright, and recognised that originality, authorship and ownership are distinct elements therein (this distinction is reflected in Sections 13, 17 and 19 of the Act; see here). The fact that a work is sufficiently original or creative for the purposes of copyright does not automatically answer the question of who its author is, nor does identifying an author resolve the distinct question of who is the owner of the copyright. As noted by Atilla in her article here (p.14), the US cases concerning Dr Thaler’s application saw the courts sidestepping this sequential inquiry. Rather than first examining whether the work satisfied the threshold of originality from which authorship subsequently arises, the absence of a human author was collapsed into and deemed determinative of the question of originality as well.
Consequently, recognising the operation of copyright’s fundamental sequential framework allowed the Registrar to acknowledge that while the work in question may be an original creative work, the question of authorship and subsequent vesting of ownership interests remained analytically distinct.
On Originality
Regarding the question of originality, it was held that “work is not deprived of originality merely because technology played an indispensable role in its production” and that the lack of real-time human intervention at the final stage does not defeat originality (p.47-49). The mere fact of the system operating through ‘statistical processing’ and ‘pattern recognition’ would not render the work devoid of creativity. It was reiterated that originality requires independent creation, accompanied by a minimal degree of non-trivial skill, judgment, and ‘creative’ effort. Thus, the objection as to originality was answered in favour of the applicant.
The ultimate ruling aside, this raises a further question about the process through which an AI-generated work acquires its ‘creative’ or ‘original’ character. If an independently generated output can satisfy the originality threshold despite the absence of real-time human intervention, what characteristics of the computational process are relevant to describing that process as creative?
The technical account of DABUS presented to the Registrar describes a sequential process involving training, association formation, perturbation, monitoring and detection, stabilisation and reinforcement, and output generation (p.16). The system was said to form associative patterns from visual and linguistic material, generate candidate pathways through controlled variations in its internal activation states, identify emerging pathways, and stabilise those reaching a specified threshold. The resulting configuration was then converted into the final artistic output.
This sequence – typical of several contemporary AI models – can be compared with certain accounts of human creative activity involving the generation, modification, and selection of possibilities. The comparison, however, does not necessarily establish that the underlying processes can equally be described as creative. For instance, Matteo Da Pelo’s exploration of whether there can be creativity without cognition, the application of the Minimal Cognitive Grid’, treats contemporary generative systems as ‘non-cognitive systems’ – i.e., artificial neural networks process information through mathematical relationships between weighted connections rather than through biological cognition. Similar arguments have been advanced in other literature as well (see Anna Shtefan’s work here). Based on this, the mere fact that an AI output’s creativity may resemble human creativity, that itself would not necessarily imply that there is a similar underlying ‘creative’ process.
This raises several questions that remain unsettled: does ‘creative process’ necessarily involve cognition, or can the term also encompass a generative process that produces creative or original expression that resembles cognitive output without the concomitant cognitive state? Alternatively, if creativity is assessed only through human intervention, does that mean the ‘creative process’ is solely limited to the human inputs, and can this be called creative at all?
Some guidance may be found in Eastern Book Company v Modak, which required independent creation accompanied by a minimal degree of creativity (specifically rejecting trivial/mechanical variations, instead requiring “careful consideration, discernment, and choice”) [p.39-41].
This provides a useful lens to examine the present decision, without necessarily resolving the question in favour of either human or ‘artificial’ creativity. As outlined above, DABUS’s process involves several steps, however, it fundamentally involves stochastic variation which operates computationally, without cognitive judgement. Thus, the question is consequently whether the concepts of selection, discernment and non-trivial variation in EBC can meaningfully describe such a computational process, or whether those concepts necessarily presuppose a human exercise of judgment.
Furthermore, the problem might be compounded by the black-box character of such models. An observer can identify the input, architecture, and output without necessarily being able to provide a complete account of why a particular internal configuration emerged rather than another. As has been argued, claims of ‘emergence’ by AI systems (the spontaneous origination of an idea beyond its training data) must be treated with caution, and the mere inability to explain a black-boxed output should not lead to the output being attributed to autonomous creative agency, especially analogous to the kind of creative contribution contemplated by copyright law. A system may produce outputs exhibiting originality without engaging in the intentional activities associated with human creativity (which are arguably mirrored in the EBC standard).
On Authorship and Ownership
The more difficult question, however, concerns the treatment of authorship. Under Section 2(d)(vi) of the Act, authorship vests in the ‘person who causes the work to be created’. This phrase, it was noted, reflected a legislative intention to distinguish between the technological mechanism for the creation of a work, and the person in whom authorship vests. The latter, it was held, could not be traced to simply be one of the contributors or the ‘last or most immediate act in the creative chain’, rather requiring a causal nexus to the creation of the work (p.91).
In this regard, the order surveys the relevant jurisprudence (Aalmuhammed v. Lee, 9th Cir. 2000) and repeatedly invokes the idea of the human being as the ‘effective cause’ or ‘mastermind’ of the work – the person “who conceived, superintended and exercised ultimate creative control over the work as a whole” (p.102). The difficulty is that, having identified this standard, the order appears to reach the conclusion that Dr. Thaler may be recognised as the author due to his role as the “initiator” of the work (p.108). It is noted that Dr. Thaler conceived and created the system, configured its operation, personally supplied the visual inputs comprising photographs taken by him, curated the linguistic inputs, supplied textual descriptions linking the visual and linguistic material, and initiated the process through which the particular work came into existence. However, being the person who conceived of the system or the one who set in motion a process is not necessarily equivalent to being the ‘mastermind’ or ‘effective cause’ of the particular expression produced at its end. Indeed, the very cases cited suggest that authorship turns on the exercise of creative control over the expressive elements of the work, rather than simply on being the person from whom the project originated.
This becomes considerably more important because AI-mediated creation can disaggregate the different stages of creative decision-making. The ‘mastermind’ may therefore not necessarily be the person who first initiated the process. The more precise question would be: who exercised creative control over the expressive elements of the particular work for which copyright is claimed? Thus, Thaler provides less assistance than suggested, as the Registrar may have treated ‘initiator’ and ‘mastermind’/‘effective cause’ as interchangeable categories.
Another difficulty concerns the Applicant’s claim to ownership. Section 17 of the Act makes the author the first owner, while Sections 18 and 19 govern subsequent assignment. Dr. Thaler claimed that DABUS was the author, while accepting that it was not a legal person and could neither hold copyright nor assign the rights to him. This creates a dead-end: if DABUS is the author, it cannot transfer the copyright to Thaler; if copyright is to vest directly in Thaler, the statutory basis for departing from Section 17 must be established. No such basis was identified. Importantly, the Registrar also held that ownership of DABUS itself cannot fill that gap. Emphasis was also placed on the public function of the Register, as it identifies not only a work’s technological provenance, but its legally competent rights-holder, in case any member of the public seeks to utilise the work in any lawful manner (p.174).
The same logic informs the Registrar’s treatment of DABUS’s legal personality. Section 2(d)(vi) identifies the person to whom authorship is attributed; it does not create a new class of legal persons. It therefore cannot supply the legal personality necessary to make DABUS a rights-bearing author. The Registrar opines that legal personality must precede the grant of authorship. To recognise DABUS as an author is a choice that belongs to the Legislature instead.
On Disclosure
As noted earlier, the applicant alternatively requested that he be recorded as the author, provided the Register included a remark that the work was generated by DABUS. The Registrar declined, holding that he could not reconstruct the application or make a suo motu amendment that the applicant had not unequivocally sought. Nor could the “Remarks” column in Form XIII be used to indirectly confer statutory status, authorship, or legal personality upon DABUS.
However, there may be some value in such disclosures. For instance, the US Copyright Office requires applicants to disclose more than a de minimis amount of AI-generated material and identify the human contribution. This permits registration to cover the human-authored elements while excluding the AI-generated material. The significance for India lies in recognising disclosure as beneficial, as it promotes transparency about a work’s provenance, and may become increasingly valuable as disputes over AI-generated works grow. It would allow the Register to record the factual circumstances of creation without requiring the Registrar to determine AI authorship or its copyright consequences. More fundamentally, because the Register serves a public function, as noted earlier in the opinion itself, its accuracy and completeness matter beyond the applicant’s immediate claim.
Conclusion
The decision presents a significant development for copyright law. In recognising that an AI-generated work may satisfy originality, it may have opened the door to accommodating increasingly autonomous forms of creation within the existing statutory framework. As AI-generated works become more common, these questions will require clearer answers, and this decision will provide a strong framework for addressing them.
