In October, I saw someone share on Twitter that they had trained an AI model to generate art in the style of renowned artist Kim Jung Gi, who had passed away only days before. Naturally, the tweet received a lot of backlash from the Twitter art community for disrespecting the memory of Kim by essentially trying to replicate all of his talent and hard work through an impersonal machine.
I think that this incident raises some interesting questions about copyright and AI art. In particular, I’m wondering whether it is copyright infringement to use copyrighted works to generate art depicting a completely different scene but in a similar style. On the one hand, one may argue that this constitutes reproduction under Copyright Act s.3(1), with the use of the copyright owner’s style amounting to a reproduction of a substantial part of their skill or judgement (Cinar v Robinson). On the other hand, is it really a reproduction of a copyrighted work if the allegedly infringing art depicts a scene that the copyrighted work never depicted? To accept this as a binding rule would mean that human artists too may be infringing copyright simply by taking inspiration from an influential artist’s style to create an otherwise original work.
Overall, this issue really highlights the competing policy tensions in Théberge between protecting the public interest and protecting copyright owners. While it would be unfair to allow people to use AI to crank out tons of artwork in a style that an artist may have spent their entire life developing and polishing, it also doesn’t seem fair to prevent the public from copying art styles that they admire to create new works, as is consistent with the connector model.
Communications Law
Hi Chloe, this is a topic that I’m really interested in! I think another question would be, should the owner of the works that AI used to train itself with (i.e. the input data) also have some copyright to the final outcome produced by the AI? If only one particular artist’s works are used by the AI to create the new work, then it’ll be easy to attribute some credit to that artist. However, if countless people’s artworks are used by the AI, then it would be more difficult. But this still doesn’t answer the question of whether the original artist of the input data ‘should’ have copyright to the AI’s work. Just some thought to play around with!