Dr Leon Furze

In a recent online lunch & learn for the Copyright Agency, I presented on the rapid advances in multimodal generative AI, including audio, video, code, and virtual reality applications. While these technologies offer exciting possibilities for creativity and innovation, they also raise serious concerns about copyright infringement and the potential harm to both consumers and creators.

I’ve written about copyright before as one of the nine areas of ethical concern in my Teaching AI Ethics series, and I’ll be updating each of those areas over the coming months because, basically, the industry hasn’t really gotten any better in the past year. In fact, we’ve seen an increasing number of lawsuits, journalistic exposés, and scandals. On the copyright front, even as I prepared the resources for the session another handful of high profile lawsuits were started, including against Google.

Sounds familiar…

Audio generation platforms like Stable Audio, Udio, and Suno have quickly gained popularity in the past few weeks and months, but other than Stable Audio, the origins of their training data remain murky at best. While Stable Audio claims to use only licensed stock music, it’s parent company Stability.AI is far from squeaky clean. Their biggest open source Stable Diffusion model is currently the subject of multiple lawsuits, and is implicated in enabling the creation of non-consensual and underage explicit deepfake materials. Udio and Suno provide even fewer assurances, and there’s a strong suspicion that they’ve scraped unlicensed copyrighted music to train their models. In fact, that very article exposing Suno’s shortcomings was written by Ed Newton-Rex, the former VP of Audio at the aforementioned Stability.AI, who left because he was so opposed to the company’s approach to copyright and intellectual property. More on him later.

This lack of transparency is deeply problematic. When journalists and industry experts start investigating these platforms and generating music that sounds suspiciously similar to existing artists, it becomes clear that the tech industry’s cavalier attitude towards intellectual property is hurting creators. Musicians and composers whose work has been used to train these models without permission or compensation are left out in the cold, while the AI companies reap the benefits.

Video Killed the Video Star

Video generation is still in its early stages, but platforms like Runway and OpenAI’s Sora are already making waves. While the output quality isn’t perfect yet, with glitchy artefacts and inconsistencies, the potential for misuse is clear: deepfakes, nonconsensual videos, and the reproduction of copyrighted materials are amongst the concerns, alongside the impact on jobs in the creative industries.

Even as these video generation tools become more accessible and integrated into existing software like Adobe Premiere Pro, the question of training data provenance looms large. OpenAI, for example, remains tight-lipped about the sources of Sora’s training data, raising red flags about potential copyright infringement. Did they scrape YouTube? Use synthetic data created in Unreal Engine? GRab as much video content as possible from the open web? At the moment, we have no way of knowing.

The Convergence of AI and Creative Industries

As generative AI continues to develop, we can expect to see a convergence of these technologies across various creative industries. Adobe’s integration of generative AI tools into its 3D creation software, Substance, and the rise of AI-generated game assets and characters are just a few examples.

While these developments may lower barriers to entry and streamline certain processes, they also pose significant risks to creators whose work could be used to train these models without proper licensing or compensation. The notion of “democratizing creativity” rings hollow when the very artists and creators whose work enables these AI systems are left out of the equation.

The copyright issues surrounding generative AI are a minefield, with new lawsuits and proposed legislation emerging on a near-daily basis. From the ongoing legal battles faced by Stability AI and Google to the recent, but incredibly confusing, copyright registration granted for an AI-assisted novel, it’s clear that the legal system is struggling to keep pace with the rapid advancements in AI technology.

Initiatives like the proposed Generative AI Copyright Disclosure Act in the US and the newly formed Copyright and Artificial Intelligence Reference Group (CAIRG) in Australia are steps in the right direction, but there’s still a long way to go in terms of protecting creators’ rights and ensuring fair compensation for the use of their work.

Responsible AI Development

As developers, creators, and regulators try to find a way through these new technologies and the issues the present, it’s crucial that we prioritise responsible AI development that respects intellectual property rights and values the contributions of human creators. Organisations like Ed Newton Rex’s Fairly Trained, which offers certification for AI models trained on licensed data, provide a glimmer of hope that it’s possible to create powerful AI systems without resorting to indiscriminate data scraping.

However, the onus shouldn’t just be on individual developers to act ethically. We need systemic changes, including adjustments to training data laws, disclosure requirements for datasets used in AI models, and clear guidelines on user-generated content and the attribution of AI-generated works. Tech companies must be held accountable for their use of copyrighted material and should work collaboratively with creators, rights holders, and policymakers to develop fair and sustainable licensing models for AI development.

Some developers appear to be coming to the party, including OpenAI. An internal team recently published a paper on arxiv detailing a possible licensing solution based on cooperative game theory. Unfortunately, given OpenAI’s track record and notorious obfuscation over these issues, it’s hard to believe they have creators’ interests at heart.

As this technology unfolds, it’s imperative that we educate about the importance of protecting copyright, advocating for transparency, and ensuring that the benefits of these powerful technologies are distributed equitably. The alternative – a world where artists and creators are left behind while tech companies profit from their work – is pretty bleak.

If you would like to get in touch to discuss generative AI, consulting, advisory services, and professional development, use the form below.