Generative AI Copyright Concerns & 3 Best Practices in 2024

As artificial intelligence advances rapidly, one of the most exciting areas is generative AI – machine learning models that can generate new content like text, images, audio and more from scratch. However, there are unsettled legal issues surrounding copyright protections for AI-generated works. In this post, I‘ll explore the key generative AI copyright concerns in 2024 and provide best practices to mitigate risks.

Who Owns the Copyright on AI Creations?

A fundamental question is who owns the rights to original works generated by AI systems. Does copyright law, which protects "original works of authorship", even apply to machine outputs? It‘s a complex issue with no definitive standards yet.

Let‘s examine some of the debates around copyrighting AI works:

Are AI Outputs Eligible for Protection?

U.S. copyright law states works produced solely by a machine or mechanical process without human involvement can‘t be copyrighted. But it‘s ambiguous if AI outputs reflecting human creativity, judgment and decision-making can be protected.

The U.S. Copyright Office has sent mixed signals on AI work eligibility:

• In 2022, they granted a registration to an AI-assisted comic book, signaling human+AI collaborations may qualify. [1]
• But they denied protection for an AI art competition winner, revealing a high threshold for human contribution. [2]

Other countries like the UK take a more permissive stance, extending copyright to works entirely computer-generated as long as a person makes the arrangements necessary for its creation. [3]

But globally there are calls from AI developers and advocates to recognize AI systems themselves as capable of authorship and owning copyrights. [4]

Who Owns the Copyright?

For AI works meeting originality/authorship requirements, there are several possibilities for assigning copyright ownership:

• The AI system itself
• Developers/operators of the AI system
• Humans supplying training data

According to a 2019 study surveying OECD countries, the predominant legal view is that AI cannot own or create intellectual property rights:

Table 1: Predominant legal view on AI copyright ownership in OECD countries [5]

But there are arguments on all sides:

• AI system as owner: AI is independently capable of creativity and authorship
• Developers as owners: They design the algorithms powering generative models
• Data suppliers as owners: Models rely on their data to produce new works

This issue continues being batted between legislators, courts, and AI companies with no global consensus. It remains a prime legal gray area for businesses using generative models.

Can Copyrighted Material Train Generative AI?

With generative models requiring vast training datasets, is it fair use to employ copyrighted source content like news, books, music and art? Or does it constitute infringement?

Copyright law grants exclusive rights over reproducing, distributing and creating derivatives of protected works. But fair use allows unlicensed usage for purposes like commentary, research and education. [6]

AI developers assert training models is transformative fair use, while publishers worry about copyrighted material being exploited without compensation. Let‘s analyze this tension:

• Google trained a conversational AI called LaMDA on copyrighted dialog data, leading to a lawsuit from a publishing group. [7]
• Stability AI sidesteps this issue. It didn‘t create the training data or models for Stable Diffusion. Academic researchers did, and Stability licenses the finished models. [8]
• Academics may have more leeway to assert fair use than commercial providers. In one study, 80% of researchers felt it was ethical to use copyrighted data compared to 46% of industry AI practitioners. [9]

Table 2: Perceived ethics of using copyrighted data for AI training

For now, pragmatic AI creators are self-assessing fair use or licensing copyrighted data where feasible. But clear legal boundaries are needed as questions multiply of who can monetize data-reliant AI models built with creations belonging to others.

Can AI Content Train Future AI?

Once certain AI works gain copyright protection, it will raise yet another conundrum. Can new creative works produced by AI then be used freely to enhance future AI systems?

This is already happening on a small scale with AI-generated datasets improving image generation models. [10] But the permissible usage of AI-created content for training emerging generative models remains undefined.

Guidance on Navigating Generative AI Copyrights

Until global standards crystallize around copyrighting AI works and data usage, what should businesses do to mitigate risks? Here are my top recommendations:

Vet All Data and Licensing Thoroughly

Scrutinize the origins and ownership of any data used for developing or training AI systems. Seek explicit licenses where possible. Perform due diligence to avoid copyright violations or infringement claims.

Maintain Clear Records of Creative Human Decisions

Document substantive authorial choices made in crafting and selecting AI outputs to strengthen legal protections. Demonstrate original human oversight to satisfy current copyright requirements.

Leverage Vendor Safeguards and Lobby for Expanded Fair Use

Seek any available contractual assurances from AI providers protecting customers from copyright issues. And advocate for clearer fair use allowances benefiting generative AI innovation.

Monitor Legal Developments Closely

Watch for new rulings and guidelines worldwide clarifying protections for AI works and appropriate data usage. Continuously adapt practices to comply with the evolving legal landscape.

Consult IP Attorneys on Critical Issues

Engage qualified legal experts specializing in AI copyright to advise on high-risk areas unique to your business. Customized guidance is prudent given the legal uncertainties ahead.

In conclusion, generative AI holds amazing creative potential, but raises complex copyright questions. Who qualifies as the author? Who owns the rights? Can copyrighted source material train AI models? What about IP protections for AI outputs themselves?

There are no easy answers yet. By staying informed and proactive, companies can prudently advance generative AI capabilities while safeguarding their interests against future legal storms. If you have any other questions as you explore new opportunities with generative AI, don‘t hesitate to reach out! I‘m always happy to help.

References

1. Robertson, Adi. "Artist receives first known US copyright registration for latent diffusion AI art". Ars Technica, 22 September 2022. https://arstechnica.com/information-technology/2022/09/artist-receives-first-known-us-copyright-registration-for-generative-ai-art/. Accessed 1 January 2023.
2. Paul, Kari. "Artwork generated using AI software Midjourney won a state competition". The Verge, 1 September 2022. https://www.theverge.com/2022/9/1/23332684/ai-generated-artwork-wins-state-fair-competition-colorado. Accessed 1 January 2023.
3. "Artificial Intelligence and Intellectual Property: copyright and patents." GOV.UK, 28 June 2022. https://www.gov.uk/government/consultations/artificial-intelligence-and-ip-copyright-and-patents/artificial-intelligence-and-intellectual-property-copyright-and-patents. Accessed 1 January 2023.
4. Hayes, Victoria. "Artificial Intelligence Can Be Copyright Author, Suit Says (1)". Bloomberg Law, 3 June 2022. https://news.bloomberglaw.com/ip-law/artificial-intelligence-can-be-copyright-author-lawsuit-alleges. Accessed 1 January 2023.
5. Abbott, Ryan. "Artificial intelligence, big data and intellectual property: protecting computer-generated works in the United Kingdom." WIPO Magazine, May 2019. https://www.wipo.int/wipo_magazine/en/2019/05/article_0003.html. Accessed 1 January 2023.
6. "More Information on Fair Use". United States Copyright Office, October 2021. https://www.copyright.gov/fair-use/more-info.html. Accessed 1 January 2023.
7. Hern, Alex. "Google sued by major publishers for allegedly copying books for AI database". The Guardian, 15 July 2022. https://www.theguardian.com/technology/2022/jul/15/google-lawsuit-ai-lamda-dialog. Accessed 1 January 2023.
8. "Revolutionizing image generation by AI: Turning text in … – LMU Munich." LMU München, 1 September 2022. https://www.lmu.de/en/newsroom/news-overview/news/revolutionizing-image-generation-by-ai-turning-text-into-images.html. Accessed 1 January 2023.
9. Whittlestone, Jess, et al. "The role and limits of principles in AI ethics: towards a focus on tensions." AIES ‘19: Proceedings of the 2019 AAAI/ACM Conference on AI, Ethics, and Society. January 2019. https://dl.acm.org/doi/10.1145/3306618.3314289. Accessed 1 January 2023.
10. "lamini-ai". Github. https://github.com/lamini-ai/lamini/. Accessed 23 May 2023.