Against the backdrop of DPIIT’s Working Paper on AI and Copyright, Akshat argues that both the paper’s proposed framework and a broader turn to copyright are ill-suited to address labour displacement caused by GenAI and risk harming the very creators they claim to safeguard. Akshat is the founder and counsel at AASA Law Chambers. He is currently a PhD candidate at the University of Cambridge. He would like to add the following acknowledgements and disclaimer- “Credit to Profs. Oren Bracha and Talha Syed for ideas, and for “The Work of Copyright in the Age of Machine Reproduction” and “LPE of IP.” Credit to Swaraj Barooah for engagement. I have no monetary interest on either side. Disclosure: I represent Flux AI Labs, pro bono, in ANI vs. OpenAI before the Delhi High Court. Views expressed here are my own.“
Part 1: Copyright is the Wrong Answer
By Akshat Agrawal
The Fallacies
Commentators on the DPIIT Working Paper are widespread. Nikhil Pahwa argues that the Working Paper does little as it provides a way out for Generative AI companies that industrialize plagiarism. He says that “Copyright is the enforcement of consent” and the DPIIT paper strips them of this ability to consent voluntarily. Some say that the Report is constitutionally vulnerable as it strips creators of their constitutional right under Article 300A of the Constitution, and some say that the Report is not ethically good for creators. Ambika, a co-blogger at SpicyIP, has argued that the model that the Report proposes is ignoring various non-economic motivations that keep creators away from licensing their works by proposing mandatory licensing.
As a creator myself, I believe that Generative AI does threaten the displacement of labour across vocations, and on a very large scale. But both the DPIIT Working Reports solution, as well as expansion/use of copyright to address this urgent issue is riddled with arguments that, once considered, show copyright cannot deliver what it promises, and would harm the very humans we are trying to protect
The first part addresses these arguments. And Part II provides a lens, beyond copyright, to attempt addressing the underlying concerns.
The Consent Framing
The “copyright as enforcement of consent” framing is intuitively strong but only as a matter of rhetoric. It sounds right. But substantively isn’t. Copyright has nothing to do whatsoever with consent in the way, for instance, privacylaw understands it. Controlling information intrinsic to personhood (for instance, health data) is peculiar to privacy law and alien to copyright.
Copyright relates to a resource, intangible expression, which once published exists in the world as a public act (does not mean it’s in the “public domain”). A published article is not “part of” its author the way health data is part of a patient. The work has been externalized and communicated, and is now transactable as a commodified “unit” in the market. It is purely a market-oriented solution giving limited transactional exclusivity.
Here is the crucial difference: privacy-based consent (bodily or informational autonomy) is inalienable, non-transactable, making it property’s antithesis. I cannot consent to sell my bodily autonomy for monetary gain. Such transactions would be void, following John Stuart Mill: “The principle of freedom cannot require that he should be free to not be free.” Which is why privacy is “fundamental” making non-transactional consent mandatory. Copyright is the opposite. It is a (“property”) claim, not a personal right. It is about market participation in information goods. The very fact that you can sell your copyright proves it has nothing to do with consent-in-the-privacy-sense. And this makes all the difference – as unlike privacy – ownership in expressive resources can be consolidated within a single unit through assignments in a company’s favour.
The consent framing does the argumentative heavy lifting precisely because it makes the demand for control feel like respect for personhood rather than a demand for market exclusivity.
The Ignorance of Non-Rivalry
“You can look at the flowerpot in my courtyard, but you can’t take it. Copyright is the lock on the gate.” This metaphor reveals a fundamental conceptual error.
Take the flowerpot, and I no longer have it. Economists call this rivalry; my possession necessarily excludes yours. The entire apparatus of physical property law exists because of this basic fact about tangible things – your use interferes with mine, and so I, who values it the most (and value is purely economic here) gets a claim to exclude you and others unless you pay up.
Now think about what happens when someone reads your news article and learns from it. You still have your news article. Nothing has been taken from you. All intangible resources in that sense are what economists call non-rival goods – unlimited people can enjoy them simultaneously without diminishing anyone’s enjoyment at the same time. This matters. A right to exclude others exists to protect one’s own use from interference. If reading your publicly available news article doesn’t stop you from reading it, doesn’t wear it out, doesn’t deplete it – then what interference are you being protected from? The foundational reason for a property claim evaporates. So, then what justifies copyright, which is also a property [exclusionary] right? The answer lies in a practical problem:
Because (i) creative works can be copied infinitely without depletion, and (ii) the cost of copying is cheaper than the cost of creation, creators face what economists call an “appropriability problem.” Imagine you spend a year writing a novel. The moment it’s out there, anyone can copy it and sell it for less than you can – because they didn’t have to spend a year writing it, while you have to recoup your costs. So then why would you bother creating in the first place?
Copyright grants limited exclusionary rights to solve this specific problem only. It has never meant to give creators control over every conceivable use or every derivable value. It is designed to address the gap between producing a work and copying it for distribution. Everything else remains free for the public to use. It is not a natural entitlement that comes from any concept of consent, personhood or anything else – even “moral rights” [distinct and separate from copyright], concerned with personhood, are not exclusion rights but merely attribution claims or claims against reputational harm.
Proponents assume that because creators work hard, they deserve control over all benefits flowing from their labour, per se. This assumption is widespread, yet fundamentally mistaken. Copyright protects expression, original communicative content by giving transactional exclusivity for a limited period (transactions in the marketplace are social- human to human). Not labour per se. Not on extraction of facts, ideas, or information (remember the hot news doctrine conflicts?)(read the difference between value creation and value extraction in Swaraj’s brilliant piece on Spicy IP here). The principle that labour alone does not create property rights is foundational. If every investment of effort created property rights, facts, ideas, and information would be carved into private tollbooths. The ability to build on what came before, to stand on giants’ shoulders, would be destroyed (see Woodmansee and Jaszi on the romantic authorship myth).
I can read your article, learn everything useful, and write my own on the same topic. I have appropriated your labour’s value without paying a rupee. That is not wrongful. It is precisely what copyright permits. Copyright has never condemned competition through independent creation by extracting mere fruits of labour. It, in fact, encourages it by keeping transformative adaptationsoutside its adaptation right.
The “Copy” Obsession!
Some argue that copying works into a computer for training triggers copyright regardless of what happens next. This reasoning succumbs to “copy fetishism“, the view that physical reproduction matters regardless of purpose or outcome.
But as stated above, copyright’s domain is expressive content – the communicative content that humans experience and enjoy. Physical acts matter only to the extent they involve anotherhuman creating a market for the protected expression – a copyright concern, or unlawfully accessing it – not a copyright concern but a contract law or a para-copyright i.e., anti-circumvention law concern. Making this a copyright concern, automatically, expands or bloats the scope of copyright law as against merely enforcing the existing doctrine.
Let me explain: When AI is trained, no human reads or is exposed to those works. No alternate market is created (markets fundamentally involve humans transacting with humans qua a particular resource – the commodified unit with an attached valuation). Copyright gives you exclusive rights over your song. It does not give ownership over the observation that most pop songs use four chords, even if the observation is made usingor based onyour song among others and is then used to create another four chord song that competes with yours. Commercial intent does not change this. A for-profit school teaching students to extract narrative patterns is not liable because tuition is charged. If the activity falls outside copyright’s domain, adding price tags changes nothing. Using copyright to curb use, per se, irrespective of whether humans are on the other transacting end of the use, or whether the use is of expression in the work as against extracting the unprotectable elements embedded within expression, for instance patterns across works etc., is expansion of copyright’s scope. [I’ve expanded on why training, storage, etc aren’t infringement in my paper co-authored with Sneha Jain here. See also Rahul Matthan here, and Shivam Kaushik here and here].
So far as fair use or fair dealing is concerned, I am very clear – this isn’t an exemption where we say “yes, this is a copyright harm, but we’ll excuse it for policy or free speech reasons.” It’s a determination that the activity falls outside copyright’s domain, its scope, entirely. Simply put, the interests copyright protects simply aren’t implicated when no human accesses the particular or a substantially similar expression – something that only comes into foray at the output stage.
Limits of the law: Addressing the “Scale” argument
“But value extraction at scale, some say, mandates a power law. And Copyright is that power law.” (from Pahwa’s piece). This argument ends up treating a quantitative difference as a qualitative difference in terms of the applicability of law.
A film student watching hundreds of movies to learn what makes scenes work has done nothing wrong, even if they are streamed or downloaded through torrents, it is the torrent hosting service that is infringing, not the user of the service (see J.Patel’s order here). A musician listening to thousands of songs to develop melodic instinct has not violated copyright. Extracting patterns about how expressions function is categorically outside copyright’s domain, regardless of volume. If this is not wrongful when humans do it, why should it become wrongful when machines do it faster? The activity is identical. Only efficiency differs.
Efficiency gains have never been grounds for copyright liability, especially because copyright is, by choice, a market based transaction oriented tool, always committed to economic efficiency(See, for eg Posner on Copyright). If it isn’t efficiency, then bloated exclusionary rights (or the “billowing white goo”) don’t make any sense. Simply put, we don’t say the printing press violates copyright because it copies faster than a scribe or because scribe jobs are going away. Copyright protects you from being undercut by copies of your work. It doesn’t protect you from being outcompeted by someone who learned from your work at a faster pace, or learned from multiple works at a larger scale, and then created something different that competes with you.
If a power law is what we need, we ought to search for one or make a new one, as against using existing copyright, and expanding it further, when it fails to help resolve the core of the issue. For instance, if scale-based market power (and hence foreclosure or price discrimination) is the concern, competition policy is the appropriate tool. If labour displacement is the concern, employment and welfare policy reforms are the appropriate tools. (David Autor’s work is interesting) [more on this in part-2].
The Incentive Paradox
Some say AI threatens to collapse human creation. If machines produce competing content at near-zero cost, why would humans create expression as a matter of their life’s vocation, in a market society, where economic accumulation is essential to survive? Thus, expand copyright to include meta-information claims.
Engage seriously, and this argument refutes itself. Copyright exists because creating is expensive and copying is cheap. When creation becomes cheap, the gap between producing and copying, the very gap copyright bridges, shrinks. If I produce a first draft in an hour that once took a month, my investment requiring protection is radically smaller. The problem justifying copyright is diminishing, not growing. The demand for expansion is intellectually incoherent on its own terms.
An objection: training inputs, the copyrighted works AI learns from, were expensive to produce. This may be true. But the expense of creating a work has always justified protection against copies that substitute for that work in the market – against someone distributing your novel instead of you. It has never justified protection against uses that extract patterns and knowledge from the work. An expensive medical textbook doesn’t expand its scope of copyright simply because it costs a lot of funds to produce – students can still learn everything from it, extract all its knowledge, and use that knowledge to compete with the author, without owing compensation for patterns absorbed.
Proponents are not asking what policy makes sense where creation has become efficient. They (think of who “they” really are) are demanding stronger protection when the rationale is weakening. That is rent-seeking dressed as principle.
Push further. If the concern is efficiency – if we’re genuinely worried about producing creative works at the lowest cost – then AI is the solution, not the problem. AI produces content that satisfies demand more cheaply than human labour can. That’s efficiency. If the social goal is ensuring creative works exist for people to enjoy, and AI achieves that goal at lower cost, then the economic framework has nothing to complain about. The incentive to produce that particular content through expensive human labour was never something the efficiency calculus valued for its own sake. It was a means to an end. If the end is now achievable through cheaper means, demanding protection for the expensive means contradicts the framework’s own logic.
Strip away the economic language – perhaps the concern is not incentives or efficiency at all. What’s actually at stake is human creativity itself – not as economic production or means of livelihood, but as practice, as communication, as meaning-making, as a way of being in the world. Disruptive creativity – unpatterned, intrinsic, emotional expression. What becomes of the poet, the musician, the artist – not as economic actors but as people engaged in one of the most meaningful activities human beings can undertake? If that’s the genuine worry, then the entire market-based framework is beside the point.
Expressive creativity has inherent value beyond satisfying market demand. Writing a poem, composing a melody, crafting a story – these activities matter because they’re fulfilling, not because they’re monetizable. This value cannot be reduced to a cost/benefit calculus measured by market price. If it could, the economic analyst could simply shrug: if human creativity has value exceeding its cost, rational creators will keep creating; if not, why should we care?
Socially as well, the concern is about retaining the human potential for creativity, as well as the production of breakthrough creativity beyond established patterns. Human creativity matters – for its own sake, for human development, as culture is the source of the self. Copyright, however, is a market mechanism. Its tools are exclusionary rights that function through market exchange. The concerns about human creativity are extra-market concerns – they challenge or question the outcomes dictated by pure market efficiency. You cannot track extra-market values by mobilizing the market via property rights.
These concerns are extra-market because market logic actually celebrates what we’re worried about – AI producing content more cheaply and efficiently. What we specifically care about is not captured by willingness and ability to pay: the intrinsic fulfillment of creative practice, the meaning it gives to human lives, and the social fabric of human culture. These values exist prior to and independent of market exchange. Copyright can adjust the price of outputs; it cannot preserve the conditions that make human creative life possible or meaningful. (More on this in Part 2)
Who Actually Benefits?
Let’s take it that both proponents and the DPIIT’s working report are right. Even if AI companies paid licensing fees for every training work, and even if creators received compensation (we all know how much will go to creators, actually – remember sold for a song?) The fundamental competitive dynamic would remain unhinged. AI would still process millions of works, generate content at near-zero marginal cost, and scale output infinitely, and in the garb of licensing (collective and hence meagre per creator fee) even produce verbatim output. A licensing fee doesn’t restore competitive parity. It just means AI companies pay a tax to content industries (and not creators) in the short run while still displacing human creators in the long run. The creator getting a meagre licensing share doesn’t help if their ability to earn a living from creating new works is undermined by AI that generates similar content instantly and cheaply. The prescription is short-sighted!
What then is the way to go? See Part 2
