.
Draft: 15 MAY
2025
This piece considers
the interaction of Indian copyright law and artificial intelligence, explores
how the law could be amended to make it more compatible with AI, and suggests
that AI has simply highlighted age-old concerns which the law has given rise to
rather than created a series of novel challenges.
* An early draft of this piece was originally published here as ‘Artificial Intelligence: The Making or Breaking of Copyright Law‘ on April 8, 2025.
Contents
1 Assisted Creativity and Copyrightability
2 Privileged Creators and Trade
3 The Statutory Anticipation of AI
4 Acknowledged Authorship
5 The Data underlying Computational Creativity
6 Rethinking the Scope of Protection
Two
‘female forms’ which ‘mov’d, and breath’d, in animated gold’ supported
Hephaestus, ‘Sovereign of the fire’ and god of craftsmen, as he approached
Thetis, Achilles’ goddess mother, who was to commission armour for her son.
These female forms had been forged by Hephaestus and, we are told, had ‘voice, and sense, and science’
bestowed upon them by the gods.[1]
Hephaestus’
helpers are widely believed to have been androids, and, yet, the Iliad describes the figures which adorn the
warrior Achilles’ shield, also forged by the god of craftsmen, in remarkably similar
terms. This leads one to wonder if Hephaestus’ women of gold were, perhaps, not
androids at all but dynamic figures on greaves, armour protecting tibiae, which
quite literally supported Hephaestus by holding him upright and facilitating
mobility. Such an interpretation does not seamlessly align with the helpers’
grasp of science but neither does it disturb the practice by which living
beings, whether gods or humans, may labour assisted to fabricate works which
would, were they made in our own time, likely be eligible to be protected by
copyright and, possibly, other intellectual property laws.
Nonetheless,
despite assisted creativity being well established, there is considerable
contemporary debate about which uses of technology are legitimate. Social media
platforms often host heated discussions in arts and crafts groups about which
handicrafts merit being described as having been handmade. Many knitters, for
example, appear to treat knitting using a loom with derision; they refuse to
recognise items made using looms as truly being handmade. From their comments,
it appears that they view even manual looms, unlike the knitting needles they
themselves use, as machines rather than as tools.
It
is not entirely clear where the dividing line between a tool and a machine lies. The former is defined by the
Cambridge Dictionary as ‘a piece of equipment that you use with your hands to
make or repair something’[2] while the
latter is defined as ‘a piece of equipment with several moving parts that uses
power to do a particular type of work’.[3] The definitions provide little clarity since even common pieces of equipment like
electric drills and some sewing machines, such as those which are handheld and
manual, straddle the line between tools and machines.
Focussing
on aids which help bring tangible works into being, the difference between
‘tools’ and ‘machines’ matters not least because it is far more likely, in our
time, that items made by hand rather than by machines would be considered to be
eligible to be protected by copyright. In the case of looms, an
intricately-patterned scarf woven on a handloom used to make cloth would not be
considered to be anything but handmade and, if its design were original,
potentially copyrightable. A scarf made on a knitting loom, however, could
easily be alleged to not have been handmade by self-proclaimed purists of the
craft and may struggle to cross the threshold of copyrightability since such
items tend to have designs which are fairly predictable.
Copyright
law leans towards protecting the products of human labour especially if, in
their fabrication or creation, human ingenuity has been applied. In doing so,
it has come to protect a mixture of works, tangible and intangible, depending
on the circumstances: not just the content of books and other literary works
but also computer programmes (deemed to be literary works by the law),[4] musical
and dramatic works, sound recordings, films, artistic works (including
photographs, sculptures, and works of craftsmanship) and computer-generated
works. It requires these works to be either original or non-infringing
depending on their nature,[5] to have
been created independently of other pre-existing or contemporaneous works,[6] and to not be so banal as to have their treatment of a subject be unavoidable.[7]
Perhaps
betraying its origins as a right printers lobbied for, manifested through the
genesis of Great Britain’s seminal 1709 Statute of Anne, the law pertaining to copyright
treats trade in books as a given, allowing the content of literary works as
well as other works which can be printed on to paper or like media, physical or
electronic, to retain copyright even if copies of the works are mass-produced
and, in the case of works of art, are in two-dimensional form.[8] In
contrast to this, copyright law is less enthusiastic about trade in tangible
works which would ordinarily merit copyright protection (especially if they are
three-dimensional and visually-appealing); it tends to protect the works
themselves rather than content embedded in them, and to limit the extent to
which they can be protected by the 1957 Copyright Act, the primary statute
which governs the subject in India, upon their having been legally reproduced
more than 50 times by an industrial process,[9] which, in practice, is when such works tend to enter commercial channels.
A
less cynical outlook would perhaps explain the disparities in the treatment of
various kinds of works, and copyright law’s choice of which mass-produced works
to shun, not with reference to the origins of copyright law but with reference
to the purpose of the grant of copyright protection. Considered through a
utilitarian lens, that copyright law tends to protect original, contemporary books
containing literary, dramatic, musical, and artistic works without reference to
how many copies of them are produced is unsurprising; copyright is ostensibly
intended to protect the revenue streams of authors and those works which
benefit society at large, particularly if that benefit is achieved through the
facilitation of access to academic learning, a purpose books are ordinarily
intended to help fulfil.
An
introduction to the earliest Indian copyright law,[10] an 1847 statute, noted that it would be expedient to establish a copyright law
in those parts of India governed by the East India Company to encourage
learning[11] while,
over a century and a half later, the overriding purpose of the 2012 amendment
to the Indian copyright statute, the 1957 Copyright Act in force in the
Republic of India since 1958, was to protect the interests of musicians and
lyricists in the film and music industry.
Closely
linked to the broader societal benefits of protecting copyrightable works is,
of course, the need to protect the creators of those works: in that swoop, the
law protects authors (and, in recent decades, performers),[12] especially if they are natural persons, by helping them monetise their work
through the grant of exclusive rights in their creations and performances.
The
law’s close links with the potential for monetisation, demonstrated not least
by the ease with which copyrights can be licensed, bought and sold, may have resulted
in those creators, who can competently leverage markets and social networks not
to mention navigate legal labyrinths, being able to financially benefit from
having their works be protected by copyright far more easily than those who
cannot.
In
practice, this has meant that, in making choices about whose works merit
protection, the law, anchored in patriarchy and ableism despite attempts in
recent decades to dislodge it from its moorings, has developed a tendency to
protect those works historically associated with men for the most part, and to
prioritise the protection of works of individual genius. As a corollary, those
creative items which have historically been overwhelmingly made by women are
also those which are least readily protected by copyright law: quotidian
cookery is unlikely to be protected but the creations of (once often male)
chefs may be eligible to merit protection, for example, and mere gardens tend
to remain unprotected but vistas shaped under the supervision of (once ordinarily
male) landscape architects can often be protected.
Strangely,
items made using looms and needles can far more easily be exceptions to the
general rule that copyright has a limited role to play in the protection of so-called
women’s arts. Needlework, brocades, lace, and tapestries all have the potential
to be considered works of art, perhaps because, although the media differ,
their visual impact can often be rendered using paints which men have
traditionally favoured.
Or
perhaps there is an older tale to tell, which copyright law has subconsciously
imbibed: Pandora, handed over to the titan Epimetheus, was moulded by
Hephaestus using clay and water. In Theogony
and Works and Days, Hesiod tells us that she was animated, and given
gifts by the gods; the goddess Athene taught her needlework and weaving.[13] It may
not be mere coincidence that these are the two ‘women’s’ crafts which copyright
law can most easily be called upon to protect.
Although
she is commonly referred to as the first woman, at no point does Hesiod tell us
that Pandora is a woman. Instead, she is described more as a prototype for
women than as a woman herself. Although she is clearly said to have been given
the ability to speak, not once does she actually speak in Hesiod’s telling.[14] And, so,
it may well be that these crafts, which women once predominantly engaged in,
were, in times of myth, passed on to them not by a woman but by an android, a
physical manifestation of artificial intelligence, resulting in their being
treated as more than banal ‘women’s work’, the products of mindful creativity
(or, at any rate, creativity derived from an android courtesy of the gods)
rather than mindless drudgery (undeserving of legal protection).
While
it has never been unusual for human beings to create works with the assistance
of technology, until recently, it has not been common for the general population to see technology create works unassisted only upon being prompted to do so by humans. Some of
the technology which allows for the fabrication of tangible works, such as that
used for 3D printing, has been relatively easily accessible by the public since
about 2009 when an early patent for fused deposition modelling, a 3D printing
technique, expired.[15] Other
technologies, such as generative AI, used to create ‘intangible’ works such as
text, audible music, and visual art which are recorded on or enjoyed through
physical and digital media without necessarily being tangible themselves, have
only exploded in popular consciousness since about 2023, in no small measure,
off the back of the release of a free version of ChatGPT based on GPT 3.5
towards the end of 2022.
In
the early 1990s, Mark Weiser pointed out: “The most profound technologies are
those that disappear. They weave themselves into the fabric of everyday life
until they are indistinguishable from it.”[16] This has certainly held
true in the case of AI, as artificial intelligence is commonly called: many
forms of AI, such as grammar- and spell-checks, are ubiquitous being all but
universally used; they are rarely thought of as being artificial intelligence
at all. The partial invisibility of AI on one hand contrasts with inescapable
visibility on the other hand brought about by news of developments in the area
frequently coming to the fore of public discussion. This makes it difficult not
just to define AI in any meaningful way for the purposes of the law which will
ultimately likely have to settle on a single set of standards and policies to
govern legitimate uses of the technologies involved without finely categorising
AI and finetuning the law for each category, an endeavour which would, in any
case, likely fail given that scientific developments invariably to outpace
legal developments.
The
emergence of GenAI, as generative AI tends to be referred to, in public use and
discourse seems, at first glance, to have been anticipated by a provision of
the 1957 Copyright Act introduced to the statute in 1994 which states that ‘in
relation to any literary, dramatic, musical or artistic work which is
computer-generated, the person who causes the work to be created’ is its
author.[17] However,
it may not be easy to identify which works could be considered to have been
computer-generated not least because a definition of ‘artificial intelligence’
has proved to be elusive, and policy documents (such as the Report on AI
Governance Guidelines Development released by the Indian Ministry of
Electronics and Information Technology in January 2025)[18] have demonstrated a tendency to forge ahead and discuss how AI should be
governed without delving into the issue of what constitutes AI in the first
place.
As
far as authorship is concerned, it is not immediately apparent that the 2018
Practice and Procedure Manuals published on the website of the Indian Copyright
Office are consistent with the mandate of the 1957 Copyright Act simply because
the manuals are non-committal about who the authors of musical works may be
(with reference to whether they are natural or legal persons),[19] indicate
that the authors of artistic works must always be human,[20] and assert that only human beings can be authors of literary works.[21]
As
such, the law does not clearly allow for the authorship of entirely
computer-generated literary, dramatic,[22] musical
or artistic works to be attributed to humans. Nonetheless, it would appear that
the absence of categorical assertion, perhaps inadvertently, allows for the
authorship of such computer-generated works to be attributed to companies; it
is not clear how the statutory position of attributing the authorship of computer-generated
literary, dramatic, musical and artistic works to the persons who cause them to
be created and the aversion to such persons being natural persons in practice could
otherwise be reconciled.
This
is in contrast to the treatment which both the 1957 Copyright Act and the 2018 copyright
practice manuals have accorded to works which are often composite in nature,
namely: sound recordings and cinematograph films (referred to as ‘composite
works’ in this piece); the statute states that their producers are their
authors[23] while the
practice manuals allow ‘producing companies’ to be named as their authors.[24] Thus, in
the case of composite works created without the support of artificial
intelligence, it appears that copyright law allows both natural and legal
persons (or, in the case of the latter, corporate entities, at any rate) to be
considered to be authors since there is no restrictive statutory provision
limiting author-producers to either natural or legal persons. The statute itself
is silent on the subject of who the authors of computer-generated composite
works may be but, despite the absence of explicit prescription, it could
perhaps be inferred that either human beings or legal persons could be
considered to be the authors of computer-generated composite works simply given
the absence, too, of explicit proscription.
Further,
there is no guidance in the 1957 Copyright Act about what ‘causing a work to be
created’ by a computer means for the purposes of the statute. As a result, it
is also possible that statutory silence could be interpreted to mean that no
person, human or legal, is the author of composite computer-generated works,
and that the software programmes which generate them are to be considered their
authors; the statutory requirement that a person be the author of
computer-generated literary, dramatic, musical, and artistic works does not
apply to computer-generated sound recordings and cinematograph films. If it were
the case that software programmes could be considered to be the authors of
composite works, it is unlikely that works so created would be protected by
copyright.
It
is important to determine who the authors of potentially copyrightable works
are simply because authorship is often only one short step away from ownership:
the author of a work is usually the first owner of copyright in the work.[25] Currently, where the statute is unclear, and computational creativity has
played a role in bringing copyrightable works into being, the authorship and
ownership of those works tend to be determined by standard form contracts which
govern the use of the relevant software.
Unfortunately,
this is not a foolproof approach to determining issues of either authorship or
ownership: not only may such agreements ultimately fail the test of
conscionability should they be challenged (which would make their provisions
unenforceable), but also, given that, as a group, they have often been drafted
in a field rife with statutory uncertainties, they may simply fail to be
consistent with what are ultimately judicially interpreted to be statutory
imperatives. There is therefore good reason to amend the law to clarify who the
authors and owners of works created by or with the assistance of computational
creativity would be, ensuring that technology companies which are the most
powerful players in the field do not simply override the rights of end-users
and appropriate their putative rights.
Unfortunately,
having come into being as a somewhat mercenary right which tradesmen lobbied
for, copyright law does not have a solid foundational understanding of what
authorship entails. Instead, to this day, it relies on a rudimentary understanding
of who may be recognised as an author (in its quest to facilitate trade) which,
entirely unhelpfully, defines the authors of literary, and dramatic works as
the authors of those works,[26] and veers
away from popular understandings of authorship primarily to accommodate
commercial demands through its recognition of producers as the authors of what
would once necessarily have been expensively-made composite works.[27]
The
law tends to treat all natural creators at par, equally but not necessarily
equitably, occasionally making allowances for them not with regard to their own
innate abilities but with reference to the kinds of works they may create; the
so-called Bollywood amendments which were introduced to the 1957 Copyright Act
in 2012, for example, primarily attempted to ensure that composers and
lyricists working in the film and music industry, being the authors of
underlying works in cinematograph films and sound recordings, would be treated
fairly.[28] The same
set of amendments also introduced provisions related to disability to the
Indian copyright statute[29] although the
fact that it focussed on treating people with disabilities as consumers of
copyrighted works rather than as creators of copyrightable works is inescapable:
the statute enhances the ability of people with disabilities to engage with existing
copyrighted content by allowing works to be reformatted for their benefit but
it contains no particular provision to enhance the ability of people with
disabilities to step into the role of legally-acknowledged authors perhaps by
lowering the threshold of copyrightability should they create works using
assistive technologies.
The
technological assistance available to authors has come to take many forms most
of which defy heterogeneous categorisation: amongst them are a combination of
physical tools at the disposal of authors, artificial intelligence which may be
used to support (and occasionally displace) natural and legal persons who are
creators, and assistive technologies aimed at enhancing the ability of people with
disabilities to engage with the world but which can also aid currently-abled
people to do so.
How
a broader range of allowances to support creative endeavours could be framed remains
an open question, and, so far, the best efforts of the law to align itself with
technological advancements, and to ensure that all creators are treated fairly,
have proved to be inadequate. Volition is more frequently recognised in the
context of infringement than of authorship with its likely having underlain
both the understanding that the commission of secondary copyright infringement
(by allowing a place to be used to communicate copyrighted works to the public
for profit) only gives rise to liability if the person committing such
infringement is aware or at least has reasonable grounds to believe that the
communication would be infringing[30] as well as the development of exceptions to liability for copyright
infringement enjoyed by intermediaries which merely provide neutral networks
through which infringing copies of works may have been transmitted.[31] Nonetheless, ‘volition’ may be a useful tool for developing a theory of
authorship.
It
has been suggested that the intention of authors at the time of creating
copyrightable works may play a role in determining the legal protection
accorded to their works.[32] However, while
discussing the relationship between copyright and designs law, the Delhi High
Court noted: “The artist’s intent at the time of creation of the artistic work
is indecipherable at the best of times. Artists are governed more often than
not by their emotions and moods and whatever be the intention at the time of
the creation of the artistic work cannot, in our view, determine the nature of
protection available to the artistic work.”[33] That being the case, it is unlikely that authorial intention or the conscious
volition of creators could be used as the basis to validate legal authorship in
Indian jurisprudence.
Coherent
extrapolated volition, the machine ‘counterpart’ of conscious, flawed human volition (extrapolating human conduct in ideal circumstances),
too, is, by extension, unlikely to be meaningful in terms of acknowledging
authorship for the purpose of copyright law. This aligns with the law’s
reluctance to recognise humans as the authors of computer-generated literary,
dramatic, musical and artistic works but it does nothing to present a solution
to the question of which factors should be considered to determine how authors
should be recognised given exponential increases in the use of technology,
particularly artificial intelligence, in creative processes.
Even
though Indian law seems inimical to recognising conscious volition as the basis
for acknowledged authorship, it may be possible to argue that functional
volition (understood with reference to the underlying processes through which
copyrightable works may be brought into being by creators) could legitimately
form the basis on which authorship is acknowledged by the law. Not only would
this help align aspects of copyright law relating to authorship and
infringement with each other by having a common thread run through each of them
but it would also allow for copyright to be shaped to account for the
capabilities of individual authors and the allowances which may need to be made
to ensure that they, and their use of technology, are considered fairly.
Another
alternative may be to consider authorship not in positive terms by attempting
to recognise it with reference to the use of technology but in negative terms
by describing when authorship would not be acknowledged by law on account of the
circumstances in which content had been created; as Elena Cooper
has pointed out: “…the challenges to authorship may stem from the objectives of
particular practices, informed by their context (e.g. online community resource
as opposed to art gallery display), rather that the use of digital technology per
se.”[34] Such an
approach could align with an emphasis on functional volition, and recognise the
disparity between syntactic and semantic comprehension demonstrated by the Chinese
Room Argument[35] (in which,
in essence, it is imagined that a person who does not know Chinese may render
the false impression of understanding the language by ‘conversing’ using Chinese
characters chosen on the basis of detailed instructions in English).
In
so re-considering authorship with reference to the circumstances of creation, technology
per se could at least be partially dislodged as a mast upon which
copyright law’s extremely rudimentary understanding of authorship pivots. For
example, text written as part of an open project could simply not be protected
by copyright at all instead of having to be freed through various ‘copyleft’
licences. Thus, copyright law could be restructured to potentially remove from
the scope of copyright protection the sundry, make doctrinal issues about what ‘copyleft’
stands for (other than being opposed to conventional copyright) irrelevant, and
harken to another time when content was protected upon registration. However,
unlike registration, considering the circumstances of creation instead of the
completion of a formal legal process would not disadvantage those creators unfamiliar
with the legal system or creators with specific requirements or disabilities
which were alleviated through the use of assistive technologies.
The
manner in which people use assistive technologies to create copyrightable works
ranges from using technologies to generate creative content in its entirety to accomplishing
disparate (subsidiary) tasks embedded within broad creative endeavours to
create works (undoubtedly of their own making) with the support of technology. Assistive
technologies themselves include a variety of tools from tangible mechanical apparatus
at one end of the spectrum to intangible artificial intelligence (embedded in
hardware though it certainly is) at the other end of the spectrum. While in the
case of tangible tools, the difference between ‘tools’ and ‘machines’ plays an
important in determining the copyrightability of works, in the case of intangible
tools such as AI systems, questions regarding the copyrightability and the use
of AI must not only consider that difference but also, bearing it in mind, ask how
AI interacts with pre-existing copyrighted content, go on to question what
precise use AI is put to by its users, and finally determine if the output
which comes into being with the involvement of AI merits copyright protection.
The
first of these questions requires an analysis of what data or content is fed to
AI to train it to act in specific ways, for what purpose the AI deals with the
content, and how it treats the content it is fed to formulate outputs.
Strangely enough, if one were to disregard the fact that the statute does not
deal with minor infractions,[36] it is
just one provision of the 1957 Copyright Act which deals with all these aspects
of the issue: Section 52 which enumerates exceptions to copyright infringement.
Protected
works other than computer programmes, as Section 52(1)(a) of the 1957 Copyright
Act says, may be used without authorisation for private or personal purposes
including research, for the criticism or review of the works themselves or
other works, and for the reportage of current events, current affairs and
public lectures. The provision does not differentiate between commercial and
non-commercial use, and although its emphasis on private and personal uses in
its first subsection may lean towards such a demarcation, it is also possible
that open artificial intelligence systems used by closed groups may satisfy the
requirement that use be ‘private’ as opposed to being public to benefit from
the exception carved out by the provision. Notably, the other two subsections
of Section 52(1)(a) of the 1957 Copyright Act dealing with review and reportage
do not contain analogous restraints. All three subsections do, however, require
that unauthorised use be fair to fall within the scope of the provision; Indian
statutory law does not define ‘fairness’ in this context but an understanding
of fairness drawn from § 107, Title 17, USC, which explains ‘fair use’ in the
context of US copyright law, has been incorporated into the Indian corpus juris through case law.
Section
52 of the 1957 Copyright Act carries on listing exceptions to copyright
infringement from subsection (1)(a) all the way to subsection (1)(zc), finally
ending with subsection (2) which extends the application of the exceptions
carved out in Section 52(1) both to translations of literary, dramatic, and
musical works and to adaptations of literary, dramatic, musical, and artistic
works in the same way that they apply to the works themselves. The exceptions
cover a wide range of subjects and circumstances but their tenor remains
remarkably consistent in that they explain how the copyright in specific kinds
of works can be used without authorisation, and with limited regard to the
identity of users. Their consistency may make them appear to be rigid but their
interpretation in the real world has proven to be less than inflexible. Indian
courts have, for example, recognised ‘transformative use’ as a valid defence to
counter allegations of copyright infringement even though the term quite simply
does not appear in the 1957 Copyright Act, much less in Section 52.[37]
There
is also the fact to consider that the exceptions to copyright infringement
appear to step straight from one work being used or copied to create another in
a manner where proximity would, but for their existence, lead to the commission
of infringement. While such an arrangement functioned smoothly in a world that
had not seen the proliferation of AI, it does not work as smoothly in an
AI-ridden world.
It
is not one step from having so-called training data as AI inputs to the
manifestation of an AI output. The lack of proximity between input and output,
when it comes to AI, could be read in two ways: either that proximity (closely
linked to the expressive use of one work in the creation of another) is
required for infringement to occur and, therefore, the lack of proximity means
that no infringement has occurred, due to which the exceptions to copyright
infringement in the 1957 Copyright Act tend to ignore situations in which there
is no proximity, or, alternatively, that choosing not to deal with situations
which lack proximity is a considered choice meaning that the lack of proximity
is no defence to allegations of infringement.
It
is extremely unlikely that there could have been much thinking along these
lines when the 1957 Copyright Act was first drafted simply because the relevant
technology had barely begun to be developed, and, despite the statute having
been amended several times since then, there does not appear to have been much,
if any, attention paid to the possibility of AI gorging content as training
data. This is far from the first time that the law has had to find ways to
catch up with technology, and although the initial steps to do so may be taken
by the courts, ultimately, the question of how to handle AI may well fall to
Parliament.
One
way in which the law could handle copyrighted content being used as training
data would be to craft a new exception to copyright infringement to allow for
such use. A broad exception would, however, be unlikely to be fair simply
because it would facilitate extremely well-funded AI companies using content
belonging to others without either authorisation or remuneration. If the choice
to expand the statutory exceptions to copyright infringement were made, it
would therefore be fitting to frame an exception to infringement to account for
the wishes of authors and rights owners (which may not always align with each
other), and to ensure that the law would not inadvertently come to either
facilitate unjust enrichment or block technological progress.
5.1
An Exception to Infringement
The
primary challenge to contend with is the almost ubiquitous nature of copyright:
copyright has come to protect almost all content, including the most banal,
provided it meets decidedly low thresholds of copyrightability. Although
doctrine somewhat raises the bar through the scènes à faire principle and the
like, per the 1957 Copyright Act, sound recordings and cinematograph films must
merely be non-infringing to benefit from copyright protection,[38] while
literary, dramatic, musical, and artistic works must be original to be
protected by copyright,[39] with
judicial interpretation resulting in ‘originality’ rarely requiring much beyond
having a work originate from an author. This has resulted in very little
contemporary content not being protected by copyright.
One
possibility would be to have the law treat works differently depending on
whether or not they were commercially available at the instance of authors, and
on how the relevant AI systems were structured. Although far from a foolproof
mode of differentiation, it would seem reasonable to ordinarily assume that
significant effort had been made to create content legally made available for
sale or hire, and it would seem to be fair to exclude such content, including
trade fiction and commercial animation, from the scope of a broad exception to
copyright infringement which allowed copyrighted content to be used as training
data by AI systems.
Simultaneously,
it would also seem justifiable to include works which were not commercially
available within the scope of a possible exception to copyright infringement
allowing protected content to be used as training data by AI, provided that the
works were not tagged with an explicit indication that they were not intended
for such use, not simply on account of the practical challenges involved in
tracing the competent persons to grant licences in respect of all content which
was not commercially available, negotiating licence terms, and paying fees but
also because of the inescapable possibility that many authors may have little
desire to be traced.
Copyright
law would need to walk a fine line between protecting authors’ revenues and
their privacy; it is entirely possible that some authors (such as those who
have been subject to gender-based violence in conservative societies and who
speak about their experiences to raise awareness) may desire anonymity for
their advocacy. Forcibly shredding the privacy of authors at the altar of
copyright maximalism is unlikely to benefit either authors or society at large
which may ultimately be deprived of their voices.
5.2
Consensual Licensing
When
it comes to content which is commercially available, too, the law would need to
tread carefully noting that within the category lies, on one hand, content
published subject to contractual negotiation which has been made immediately
available for monetisation whether or not its authors have received pecuniary
remuneration such as text published by traditional academic and trade
publishers, and, on the other hand, content published subject to contractual
imposition (including through various software EULAs[40] and website Terms of Service) which could be monetized but may not have been
published with the intention of having it enter commercial channels such as what
is likely to be a significant fraction of end-user posts on social media
platforms.
In
the case of content published subject to contractual negotiation, it is probable
that the identity of authors would be known to publishers while, in the case of
content published subject to contractual imposition, such knowledge cannot be
taken for granted, and any framework to obtain authors’ consent would therefore
have to be developed cognizant of the need to avoid unmasking the identity of
authors who would prefer enhanced privacy, if not total anonymity. Nonetheless,
in both cases, it would be fair to require that authors’ consent for the use of
content they have created as training data AI be obtained.
Ensuring
that publishers acting as content aggregators do not unilaterally license
content en masse to AI companies
without consent from and remuneration payable to the authors who have created
it may require an amendment to the 1957 Copyright Act to ensure that signed
grants of rights to publishers, whether historical or not, are not interpreted
to allow such conduct.[41] Further, while tweaking the provisions in the copyright statute pertaining to
contracts would probably be adequate to obtain informed consent in the case of
content published subject to contractual negotiation, it would probably be
necessary to include an additional provision either to the statute itself or in
subordinate legislation to ensure that users subject to contractual imposition were
in fact aware of the possibility that content they created and uploaded could
be used to train AI, and to give them the option to refuse to allow their
content to be so used.
5.3
Non-Voluntary Licences
Finally,
is the question of the manner in which AI systems should benefit from
derogations from the usual flow of copyright law whether through exceptions to
infringement, extraordinary contractual perquisites or otherwise.
Looking
at the issue through a utilitarian lens, it could be argued that, to benefit
from derogations without charge, an AI system should need to be open and have
subscription levels where end-users (being members of the public) could
meaningfully use it without payment. Failing that, if AI systems were some
combination of closed, proprietary and chargeable, there could conceivably be
established a mechanism through which statutory licences could be obtained in
respect of them to legalize their use of certain forms of protected content as
training data.
Such
a layered approach is not unknown to copyright law: the provisions in the 1957
Copyright Act which help to make copyrighted works accessible to people with
disabilities are contained in an exception to copyright infringement which
focuses on personal, private and not-for-profit access to copyrighted works in
streams independent of the usual commercial channels of trade,[42] and in a
compulsory licence which allows for-profit operations.[43] In essence, the law facilitates the possibility of the grant of a licence in
cases beyond the purview of the statutory exception to infringement.
Copyright
law generally does not operate independently: although its substantive
provisions are contained almost exclusively in the 1957 Copyright Act, in
practice, its procedural provisions, particularly relating to rights ownership
and exploitation, come into their own at the confluence of contract and
copyright law. As a result, it is not exceptional for copyright law to call
upon contract law, and there appears to be no reason why it could not do so
with reference to AI.
The
proliferation of AI, with creators of all stripes using it, has forced
copyright law to grapple with determining not only the extent to which AI
programmes themselves should be protected by copyright law but also the extent
to which works may be brought into being with technological assistance without
sacrificing their eligibility to be protected by copyright (and without their
creators sacrificing the right to be acknowledged as authors).
Thus,
the extent of copyright protection needs to be considered both qualitatively
and quantitatively with analyses referring to eligibility for protection and,
where protection is merited, the duration of such protection respectively. With
regard to the latter, it would be prudent to rethink the term for which
copyright protects software in general (within which AI is included), and the
term for which copyright protects the output of AI particularly when it comes
to the output of generative AI.
In
the case of AI systems, given the rate at which we now know that technology can
progress, it would seem that lowering the term of copyright for software to a
number of years comparable to that for which patents or semiconductor
integrated circuits are protected would likely achieve the dual aims of
protecting investment in research, and not supressing further research by
allowing those who do not own the rights in specific AI systems to further
develop them after a certain number of years. Dealing with the term for which
the output of AI should enjoy copyright protection, however, requires one to
simultaneously deal with the issue of whether such output is or should be
copyrightable in the first place.
Assuming
that works meet the criteria to be protected by copyright, to answer the
question of how long copyright protection should last, it would probably be
possible to argue that where works are entirely created by GenAI upon being fed
prompts by humans, the duration of protection should be much lower than when works
are created by human beings with or without the assistance of technology. This
would divide works into two groups: that which had come into being upon the
execution by AI of instructions from human beings and that which had come into
being upon the execution of activities by humans possibly with the support of
AI; the degree of support would not matter if such a bifurcation were adopted.
At a theoretical level, ignoring the degree of technological assistance
employed to create works would mirror copyright law’s refusal to engage in the
qualitative analysis of specific works to determine their copyrightability on
the basis of artistic merit and, on a practical basis, concentrating only on whether
or not technological assistance had in fact been employed to bring works into
being would demonstrate cognisance of the difficulties in assigning a
quantitative test to establish the degree of technological support relied upon
to create a work without, at the very least, also engaging in qualitative
analyses of the work.
As
difficult as it may be to determine for how long a copyrightable work should be
protected, it is probably even more difficult to determine if a work should be
protected by copyright. While there is no doubt that AI programmes generally
merit copyright protection, the copyrightability of the output of AI programmes
is fraught with concerns primarily evolving from two issues: the legality of
the manner in which AI treats content, and the legitimacy of the uses to which
humans put AI with ‘legitimacy’ being understood to include not just legal but
also ethical considerations. Unfortunately, these issues raise questions which
do not have easily discernible answers: firstly, the question of what exactly
AI does with content which has been fed to it as training data, and, secondly,
the question of what exactly humans who use AI are doing with both AI and its
output. Threshing out these two issues is an exercise complicated by the facts
that we do not truly have a workable theory of authorship to guide us, that we
are far from certain about the extent to which the framing of prompts amounts
to an act of authorship for the purposes of copyright law, and that we rarely
seem to know what happens to data within an AI system.
The
operations of AI systems tend to be opaque, especially in the case of
proprietary AI, and, so, it is often challenging to determine if, in its
operation, AI winds up committing copyright infringement or other legal wrongs.
It would, therefore, probably be prudent to frame supplementary guidelines
governing how legal AI systems may operate, particularly predictive AI systems
which are used in situations where they could have (sometimes devastating)
real-world consequences for human beings beyond simply creating content that
would be considered artistic; if one takes a step back and looks at the big
picture, it is difficult to argue that copyright infringement is the primary wrong
that legal policy should address in this context. For example, AI systems which
are used to attempt to accomplish such tasks as determine eligibility for
parole, assess the risk of recidivism, or analyse the manner in which an
epidemic may spread should, ideally, be rigorously tested so that the risks of
employing AI are well established, and humans are in a position to make
informed decisions factoring in the likelihood of AI output being incorrect.
This may require not only framing policies on how the data inputted into AI
systems may be treated and processed (notably in terms of confidentiality and
copyright) but also framing policies to ensure that AI systems themselves stand
up to scrutiny such that claims of their efficacy are not divorced from
reality; this may necessitate having the operations of AI systems not operate
completely with an impenetrable ‘black box’.
While
forcing (often corporate) owners of AI systems to divulge enough about their
proprietary software to ensure that those systems actually work as claimed and
do so in a legal manner may require ‘diluting’ their intellectual property
rights, it should be possible to encompass within the copyright framework
(which protects software in India) mechanisms beyond those which are extant to
protect information which has been publicly divulged about AI systems (possibly
akin to those used by the legal regimes which govern patents and semiconductor
integrated circuits). Ultimately, the law will have to balance proprietary
rights and public interest to ensure that one is not privileged over the other.
As
far as the activities of human users are concerned: although it is unlikely
that end-users could be held liable for the operations of AI they did not
control, the purpose and manner in which they used AI would likely be factors
in determining whether they could be considered the legal authors of works they
had brought into being or, perhaps, had helped bring into being in conjunction
with AI. Complementary to the issue of whether or the output of AI could be
considered to have been authored by its end-user creator are the questions of,
firstly, whether or not the output could be considered to have violated the
rights of people (other than the end-user who had caused the work to come into
being) such as the right to privacy and the right of publicity, and, secondly,
whether or not the output could be considered to have infringed rights
subsisting in preexisting works.
In
some cases, the issues involved would be relatively easy to clarify. For
example, although copyright law has little enthusiasm to entangle itself with
morality, it is extremely unlikely that a non-consensual, explicit audio-visual
work generated by or with the assistance of computational creativity featuring
an identifiable woman would merit copyright protection (not least because it
would have almost certainly been created using preexisting imagery of the
woman); not only would a user who had created such a work not be recognised as
an author by copyright law for the purpose of protecting creative content but
such a user would also likely find themselves being accused of a range of other
legal wrongs including that of defamation which, in India, is a legal wrong
under both tort law and criminal law.
Thus,
the first issue relating to rights violations could be serviceably addressed by
drawing on not just various extant statutes which deal with the issues at play (such
as the right to privacy, and the right of publicity) but also by drawing on
tort law and, where criminal offences have not been alleged to have been
committed, contract law to examine if there were any way the lens of ‘consent’
could justify what would otherwise be unjustifiable rights violations. The
second issue, that of infringement, would be reasonably easy to address with
reference to sound recordings and cinematograph films simply because, under the
1957 Copyright Act, infringing works cannot have copyright subsist in them. In
the case of other works, however, the issue could become murky because copyright
law merely requires non-composite to be original; coupled with the doctrine of
independent creation which allows copyright to subsist in two similar or even
identical works if they have been independently created, it is not impossible
that copyright could subsist in what would, but for the doctrine of independent
creation, have been infringing. Here, too, copyright law is clearly not
designed to cope with AI.
On
one hand, while the end-user of a work may imagine that it has been
independently created, the AI itself may have been fed protected content as
training data. How the law treats the output of AI may have to factor in
whether or not the training data had been acquired legally, and, if so, the
terms under which it was so acquired. And, on the other hand, one of the
concerns which has repeatedly presented itself in relation to the output
generated by AI is that of the legitimacy and copyrightability of works created
in the style of contemporary authors. It may be possible to argue that where
those authors have, in any case, commercialised ‘their’ style such they employ
studios filled with artists to replicate it in a variety of ways, they have
also lost the right to protect their style being imitated by end-user creators
using AI. Such an argument could, by analogy, conceivably apply to the works of
any author has made their style formulaic and commercialised although there is currently
no body of judicial opinion which could be used to conclusively either
establish or demolish such an argument.
The
issue of the who was recognised as the author of works created with the
assistance of AI would likely depend not just on the extent to which AI had
been relied upon but also on the terms of use appurtenant to the AI. Even if
the output were copyrightable, it would not necessarily be possible for a
corporate owner of AI to be considered to be the author (except in cases where
the output was in the form of a sound recording or a film) although it is
possible that the company could, through its terms of service, require end-user
creators to assign all rights in the output to the company or to allow the
company to exploit the output in various stated ways. To avoid situations in
which companies took undue advantage of end-user creators and, possibly, unjustly
enriched themselves at their expense, it would likely be necessary to lay down
minimum standards for such contracts in subordinate legislation if not in
statute itself.
Further,
the use agentic AI alone would likely not result in the creation of a
copyrightable work. However, the use of agentic AI or AI which accomplished
specific and limited tasks in a broader creative process, which allowed the
argument to be made that AI had merely supported the creative endeavours of its
human users, could well result in the creation of works which were
copyrightable and whose creators could be considered to be legal authors. The
use of generative AI would, however, be far less likely to result in either the
creation of copyrightable works or in having human users who prompted the
creation of works be recognised as legal authors.
The
line between the two, however, is extremely thin and blurred. For example, AI
which harmonised a melody could, depending on one’s point of view be considered
to be either agentic or generative AI.[44] Further,
the mere harmonisation of a melody, it could be argued, does not merit
copyright protection by itself given that it must, in order to sound sonorous
to the human ear, abide by the rules of music theory. This leads to the
development of a situation where it is not obvious who the author of the
content should be considered to be or, for that matter, whether the content
brought into being is copyrightable at all.
Other
forms of computational creativity also raise similar concerns. Consider AI used
to develop recipes either autonomously or semi-autonomously upon having been
fed a range of recipes developed by humans along with rules drawn from hedonic
psychophysics, a field which explores what appeals to human beings: although
the AI could be used to autonomously create recipes, it is almost certain to
function much better in collaboration with human beings who know how to cook
than in isolation.
Alexandra Kleeman described how she ‘cooked some eggplant
fritters that made convenient use of every sad, wrinkling root’ in her refrigerator
using AI named Chef Watson to help combine an unlikely assortment into
the ingredients of a recipe.[45] Making the best use of leftovers and assorted remnants of vegetables is, of
course, not a new task: women have been doing so for centuries, if not longer.
Not once in history does there appear to have been a concerted push to
recognise the practice as worthy of consideration as labour which should be
renumerated, much less as a practice which results in the creation of recipes
which should, arguably, be worthy of intellectual property rights protection.
However, the insertion of computational creativity into what would once have
been primarily a domestic chore, and the commercial opportunities AI presents
by urging humans to think along new lines to develop recipes not just with
leftovers but with unusual combinations of ingredients, has changed the
equation. Although no-one may be entirely certain of what the intellectual
property rights position of content – in this case: recipes – generated either
by or with the help of AI should be, it is certain that AI has breached the
domestic arena, bringing such issues, likely inadvertently but nonetheless
firmly, into the realm of academic discussion,[46] by
raising questions about the authorship, ownership, and commercialisation of
so-called women’s work.
The
ambiguities associated with computational creativity and copyright law are not
new. They echo old uncertainties relating to the difference between ‘tools’ and
‘machines’, and the reluctance to treat ‘women’s work’ with any respect worth
mentioning, with much the same effect: there being a distinct lack of clarity
about who could be considered a legal author, what legitimate acts of
creativity encompass, and which works might be eligible for copyright
protection. So far, there has been no pressing need to answer any of these
questions: conventional commercial channels have managed to sidestep them since
answering them has not been essential to make money, and, whatever else
copyright may be, it is unarguably an economic right. With AI having entered the
picture, however, these old fault-lines can no longer be ignored simply because
AI systems, associated with large sums of money, seem primed to exploit them.
It may, however, be prudent not to have copyright law centre technology and to
consider sociology more closely, scrutinising consent, forcing equity where it
is absent, and examining the circumstances in which creative endeavours unfold.
While
it may appear that AI has caused tectonic shifts in copyright landscapes, it
may be more accurate to say that it has merely caused old fault-lines to appear
in stark relief. In doing so, it creates the opportunity to address lacunae in
the law and to recast it in an equitable mould. Whether that opportunity is
availed of or squandered remains to be seen.
References
[1] Pope, Alexander, trans. The Iliad of Homer. W. Suttaby, 1806. https://archive.org/download/homeriliad00home/homeriliad00home.pdf.
[2] “Tool.” https://dictionary.cambridge.org/dictionary/english/tool.
[3] “Machine.” https://dictionary.cambridge.org/dictionary/english/machine.
[4] Section 2(o) of the 1957 Copyright Act (India).
[5] Section 14 of the 1957 Copyright Act (India).
[6] See: the doctrine of independent creation
[7] See: the scène à faire principle and the merger doctrine
[8] In practice, some works such as pop-up books containing 3D art may be exceptions to the general rule that 3D art cannot be reproduced an uncounted number of times without potentially risking the protection of copyright law.
[9] LNG Express India Private Limited v. INOX India Limited. Supreme Court of India, CNR: SCIN010525222024, 15 April 2025. api.sci.gov.in/supremecourt/2024/52663/52663_2024_3_1501_60892_Judgement_15-Apr-2025.pdf. The case analyses Section 15 of the 1957 Copyright Act (India).
[10] References to specific laws, regulations, reports, and manuals in this piece are all references to Indian laws, regulations, etc., unless otherwise specifically stated.
[11] Copyright Act (India) 1847, https://web.archive.org/web/20160304212712/http://lawmatters.in/wp-content/uploads/2014/04/Indian-Copyright-Act-1847.pdf
[12] The protection of performers is beyond the scope of this piece although Indian copyright law does protect live performances in a manner which could be considered to be loosely analogous to the manner in which it protects works.
[13] Evelyn-White, Hugh G, trans. Hesiod: The Homeric Hymns and Homerica. 1914. Reprint, William Heinemann, 1929. https://archive.org/download/hesiodhomerichym0000unse/hesiodhomerichym0000unse.pdf.
[14] Francis, James A. “Metal Maidens, Achilles’ Shield, and Pandora: The Beginnings of ‘Ekphrasis.’” The American Journal of Philology 130, no. 1 (2009): 1–23.
[15] “Timeline of the 3D Printing History – ASME,” https://www.asme.org/topics-resources/content/infographic-the-history-of-3d-printing.
[16] Weiser, Mark. “The Computer for the 21st Century.” Scientific American 265, no. 3 (1991): 94–105.
[17] Copyright Act (India) 1957, s 2(d)(vi).
[18] Ministry of Electronics and Information Technology. (2025). Report on AI Governance Guidelines Development. https://indiaai.s3.ap-south-1.amazonaws.com/docs/subcommittee-report-dec26.pdf
[19] “Practice and Procedure Manual 2018: Musical Works.” Copyright Office, Government of India, 2018. https://copyright.gov.in/Documents/Manuals/MUSICAL_MANUAL.pdf.
[20] “Practice and Procedure Manual 2018: Artistic Works.” Copyright Office, Government of India, 2018. https://copyright.gov.in/Documents/Manuals/Artistic_Manual.pdf. In explaining how to fill the application for copyright registration, the manual states: “Correct and appropriate details of the Author (the person who has actually authored or was involved in the authoring of work. There can by more than one persons who can jointly and severally be the author of a work. A firm/organization cannot be an author of work by no means; this column shall always be filled with the name & details of a person.”
[21] “Practice and Procedure Manual 2018: Literary Works.” Copyright Office, Government of India, 2018. https://copyright.gov.in/Documents/Manuals/LITERARY_MANUAL.pdf.
[22] Although there appears to be no copyright manual published by the Indian Copyright Office specifically dealing with dramatic works in force, the definition of dramatic works in the 1957 Copyright Act reveals that dramatic works tend to possess a dual nature with their also, in the usual course, simultaneously being literary, musical, or artistic works. Due to this, it is possible to infer that the constraints which the 2018 practice manuals of the Indian Copyright Office apply to literary, musical, and artistic works also apply to dramatic works.
[23] Copyright Act (India) 1957, s 2(d)(v).
[24] “Practice and Procedure Manual 2018: Sound Recordings.” Copyright Office, Government of India, 2018. https://copyright.gov.in/Documents/Manuals/SOUND_RECORDING_MANUAL.pdf; “Practice and Procedure Manual 2018: Cinematograph Works.” Copyright Office, Government of India, 2018. https://copyright.gov.in/Documents/Manuals/CINEMATOGRAPH_MANUAL.pdf.
[25] Copyright Act (India) 1957, s 17.
[26] Copyright Act (India) 1957, s 2(d)(i).
[27] Copyright Act (India) 1957, s 2(d)(v).
[28] Saikia, Nandita. “The Bollywood Amendments: Film, Music and Indian Copyright Law (2010 to 2012).” SSRN Electronic Journal, January 1, 2010. https://doi.org/10.2139/ssrn.1566350.
[29] Saikia, Nandita. “Viewing Copyright Policy Through Music and Injury” April 1, 2022. https://copyright.lawmatters.in/2022/04/viewing-copyright-policy-through-music.html.
[30] Copyright Act (India) 1957, s 51(a)(ii).
[31] Copyright Act (India) 1957, s 52(1)(b) and (c).
[32] Buccafusco, Christopher. “A Theory of Copyright Authorship.” Virginia Law Review 102, no. 5 (2016): 1229–95.
[33] Microfibers Inc. v. Girdhar & Co., High Court of Delhi, RFA (OS) NO.25/2006, 28 May 2009.
[34] Cooper, Elena. “Reassessing the Challenge of the Digital: An Empirical Perspective on Authorship and Copyright.” In The Work of Authorship, edited by Mireille van Eechoud, 175–214. Amsterdam University Press, 2014.
[35] “The Chinese Room Argument (Stanford Encyclopedia of Philosophy),” October 23, 2024. https://plato.stanford.edu/entries/chinese-room/.
[36] Copyright Act (India) 1957, s 14.
[37] Syndicate of the Press of the University of Cambridge v. B. D. Bhandari & Anr., High Court of Delhi, RFA (OS) No.21 of 2009, August 03, 2011.
[38] Copyright Act (India) 1957, s 13(3).
[39] Copyright Act (India) 1957, s 13(1)(a).
[40] End User Licence Agreements.
[41] A proviso to Section 18(1) of the 1957 Copyright Act states that no assignment ‘shall be applied to any medium or mode of exploitation of the work which did not exist or was not in commercial use at the time when the assignment was made, unless the assignment specifically referred to such medium or mode of exploitation of the work’ but an analogous proviso does not appear in Section 30 of the statute which deals with licences.
[42] Copyright Act (India) 1957, s 52(1)(zb).
[43] Copyright Act (India) 1957, s 31B.
[44] Saikia, Nandita. “Priorities and Power: An AI Governance Proposal” January 13, 2025. https://copyright.lawmatters.in/2025/01/priorities-and-power-ai-governance.html.
[45] Kleeman, Alexandra. “Cooking with Chef Watson, I.B.M.’S Artificial-Intelligence App.” The New Yorker. November 20, 2016. https://www.newyorker.com/magazine/2016/11/28/cooking-with-chef-watson-ibms-artificial-intelligence-app.
[46] Somaya, Deepak, and Lav R. Varshney. “Ownership Dilemmas in an Age of CREATIVE MACHINES.” Issues in Science and Technology 36, no. 2 (2020): 79–85.
