AI Style Mimicry and the Limits of Choreographic Protection

Abstract

The ability of generative AI to learn and imitate a choreographer’s unique movement style poses a threat to the copyright principle that “style is free to use. Generative AI’s ability to learn and replicate a choreographer’s unique movement style threatens the principle of copyright that “style is free to use. Indian law is plagued by two issues: an unlicensed set of training movements on the recorded material would seem to violate the Copyright Act, but an output that closely resembles a style without copying specific movements would not seem to infringe the copyright. Current Section 52 defences are without any explicit exception for AI or text and data mining, which poses a liability tension between training and output. Solutions range from the use of moral rights, compulsory licensing for training, and even the introduction of a sui generis “style” right; all are associated with practical and doctrinal issues that warrant rethinking of copyright norms.

Introduction

The ability of generative AI to actually learn the choreographer’s style and create entirely new dances which clearly possesses its own style, without copying a single routine clears the  fundamental rule of copyright law: that style is free for anyone to use. When AI can mimic a creator’s dynamics, rhythm and spatial patterns, the distinction between inspiration and appropriation becomes blurred. Can copyright have anything to say if a machine learns to perform copyrighted choreography and it generates commercially valuable performances in the choreographer’s style? 

The Two-Stage Copyright Problem: Training and Output

The problems of copyright for AI-generated dance can be categorized into two aspects: training and output. The training stage brings up the question: Is using copyrighted choreography for creation of an AI model a copyright violation? The Indian Copyright Act defines recorded choreographic works and the use of such works for training purposes, without an exception, infringes the copyright. Choreographic works are protected under the Copyright Act, 1957, Sections 2(h) (as reduced to writing or otherwise recorded) and 13 as “dramatic works.” This legislation is especially crucial for AI training. Beyond the issue of reproduction, however, the question of whether such use is an infringement of the exclusive rights under Section 14 also arises when recorded choreographic works are reproduced, copied, stored and disseminated without permission in training datasets. The Indian law does not have any specific exception for AI-training or text and data mining, which further supports reasons for potential infringement. 

In contrast to the U.S. where courts have occasionally determined that large-scale copying for technological purposes is fair, India’s exception, under Section 52, is more limited, and there is no express exception for AI training or text-and-data mining. As a result, it may become challenging to justify the unlicensed training on choreographic datasets, given the existing laws. The output stage presents itself to be a different challenge The more a dance is different from any protected sequence, the harder to prove the infringement. It is expression, not ideas, which is protected by copyright. This means that a model could create dances that are steeped in copyrighted material without the copyrighted dances thereby being classified as illegal. The standards for training liability and output liability should thus be different. 

A style-expression dichotomy and its breaking point

For choreographers, the most pressing issue with AI is the principle of copyright that style cannot be protected. This was once a good idea because creativity hinges on the ability to learn from and develop others’ work. Human imitation, through personal interpretation and creative choice, can turn into influence (which may turn into original authorship).

Generative AI is set to alter this trend. Through the training of hundreds of routines, a model can learn to detect and replicate a choreographer’s movement patterns, rhythms and spatial decisions in countless outputs. No particular routine can be reproduced but the dances that are created can be very in line with the choreographers style.

This reveals a lacuna in the copyright doctrine. Courts have always denied protection to a style, because it was and is impractical to imitate the human form. But AI can mimic at scale and on a commercial level. Indian copyright law has not yet considered how far this distinction between a traditional style expression and the new technological reality is sufficient to serve the interests of copyright enforcement.

Towards a Legal Framework: Moral rights, performer rights and doctrinal recalibrations

Whereas copyright law offers some protection against AI style mimicry, two alternatives are worth considering: moral rights and performer rights. The authors have moral right to be attributed and have the right to integrity, as per Section 57 of the Indian Copyright Act. These rights do not cover the protection of style, but AI-generated dances that closely mimic a choreographer’s signature style without attribution can conflict with the spirit of the right to attribution, especially if consumers believe it is a product of that choreographer.

Another potential avenue might be through the rights of the performer. Many choreographers also perform their own pieces, and training AI on recordings of these performances could mean that they are liable under Section 38 as well as copyright to the choreography.

In more fundamental terms, AI can necessitate changes to current laws. A compulsory licensing regime for AI training, thereby guaranteeing compensation for creators whose works are used for training AI sets, is one possibility. The other is a style right, sui generis, protecting the creators who have created a distinctive movement vocabulary that can be objectively evidenced by existing works.

Both methods have their drawbacks, such as defining style in an adequately exacting way, or restricting legitimate creative borrowing. However, the sentiment expressed is resonating: creators’ unique artistic voice must not be used for commercial purposes by an algorithm without their permission, acknowledgment, or remuneration on a large scale.

Conclusion

The AI-choreography controversy reveals that the copyright law’s expression vs. style dichotomy is not watertight. AI alters that reality, which is the reason for that distinction being created for a world of imperfect human imitations. If a model can learn from hundreds of dances that are protected, produce commercially viable dances in the distinct ‘style’ of a choreographer and since no exact sequence was copied, then copyright is undermining its own purpose? Indian law needs to rethink the issue of AI training and style-expression. Choreographers, and other artists, will not be given adequate protection until the doctrine of copyright evolves to protect AI-generated creativity. 

References 

The Copyright Act, 1957 (India), Sections 13, 14, 38, 52, 57.

Authors Guild v. Google, Inc., 804 F.3d 202 (2d Cir. 2015).

Getty Images (US), Inc. v. Stability AI, Ltd., No. 1:23-cv-00135 (D. Del., filed Feb. 2023).

Directive (EU) 2019/790 on Copyright in the Digital Single Market (DSM Directive), Arts. 3–4 (Text and Data Mining Exceptions).

European Parliament, Artificial Intelligence Act (Regulation (EU) 2024/1689), Transparency Provisions for General-Purpose AI Models.

Copyright Office of India, ‘Guidelines on AI-Generated Content and Copyright’ (2023, draft).

Zhu, W. et al., ‘Bailando: 3D Dance Generation by Actor-Critic GPT with Choreographic Memory’, IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2022 — example of AI dance-generation research.

Lemley, M.A. & Casey, B., ‘Fair Learning’, Texas Law Review, Vol. 99, pp. 743–784 (2021).

Ginsburg, J.C., ‘People Not Machines: Authorship and What It Means in the Berne Convention’, IIC – International Review of Intellectual Property and Competition Law, Vol. 49, pp. 131–135 (2018).

Balganesh, S., ‘Foreseeability and Copyright Incentives’, Harvard Law Review, Vol. 122, pp. 1569–1632 (2009).

Drahos, P. & Braithwaite, J., Information Feudalism: Who Owns the Knowledge Economy? (Earthscan, 2002) — background on IP expansion debates.

Ficsor, M., The Law of Copyright and the Internet (Oxford University Press, 2002) — comparative copyright methodology.