As most in the IP world would’ve noticed over the last two months, there’s been much discussion around the Working Paper on Generative AI and Copyright (Part 1), with its ambitious subtitle “One Nation One License One Payment – Balancing AI Innovation and Copyright”. The 125 page report (available here) is focused on ‘the legal issues relating to the use of copyright-protected works as training data for GenAI systems’, while part 2 (yet to be published) will focus on ‘the copyrightability and authorship of GenAI-generated outputs, including the applicability of moral rights and attribution of liability for infringing outputs’.
For those who need a quick refresher – the central crux of the working paper is the suggestion to implement a mandatory (no opt-out) licensing scheme for AI companies to use any copyrighted content, for a (to be determined) royalty rate collected and then distributed by a collecting entity. As is evident to anyone who’s been following our posts, several of us find different parts of the Working Paper quite concerning. Four of us came together and collated our concerns in the form of a submission to the DPIIT and submitted them yesterday. They are available here (PDF).
Our expressed concerns stretch from questioning the foundational premise for the working paper, the legal implications that spillover from their treatment of copyright law, as well as the practical problems that would stem should there be an attempt to implement the idea proposed in the working paper.
As a quick thought exercise – even if all the problems disappear or are ignored, as per the currently provided numbers, what would this break down to?
- As one of the largest AI companies – OpenAI’s projected annualised revenues for 2025 are $20billion;
- The working paper estimates there are likely over a billion different copyright owners, and
- A very generous, hyper-unrealistic revenue rate of 10%
Assuming magically ownership traces of all scraped content is found and sorted (how? By finding and processing copyright ToS for each bit of scraped content?), and payment details are then attached to them, what comes next? A back of envelope calculations would thus give approximately $2 per copyright owner, if everyone only has one work! Once you account for the mega corps ownership patterns, this leaves literal pennies, at best, for the rest. IF the concern regarding creators is what they’ve described – about labour displacement, loss of jobs, etc, how does this help anything? And meanwhile, for most AI companies, yet to figure out how to become profitable, giving away a chunk of revenue takes them closer to a death knell. If they are to survive, it’ll likely be by being absorbed within the big few players (as usual). So – what exactly is the outcome being envisaged here? A lowering tide, lowers all ships!
As quick links for interested readers, I’m also leaving a list of posts that have been carried on the blog discussing different aspects of the working paper below:
- Bulldozer Justice comes to the Digital Sphere: Looking at the GenAI-Copyright Report – Swaraj Paul Barooah
- One Nation, Forced Licenses, Multiple Payments: (Un)Balancing AI Innovation & Copyright – Ambika Aggarwal
- Part 1: Copyright is the Wrong Answer – Akshat Agrawal
- Part 2: Copyright is the Wrong Answer – Akshat Agrawal
- One Nation, One License, One Big Shortcut: Doctrinal Stagnation in the DPIIT AI Working Paper – Shivam Kaushik
- Copyright Maximalism by Design? Rethinking DPIIT’s Licensing Centric Approach to AI Training – Vishno Sudheendra
- DPIIT Working Report: Lawful Access while ignoring Copyright Law – Yogesh Byadwal
