The Death of an OpenAI "Whistleblower": Ethical Dilemmas and Copyright Disputes in Generative AI

The Death of an OpenAI "Whistleblower": Ethical Dilemmas and Copyright Disputes in Generative AISuchir Balaji, a 26-year-old former OpenAI engineer who contributed to the development of GPT-4, was found dead in his San Francisco apartment on November 26, 2023. Officially ruled a suicide, Balaji's death has brought intense scrutiny to the burgeoning ethical and copyright controversies surrounding generative AI

The Death of an OpenAI "Whistleblower": Ethical Dilemmas and Copyright Disputes in Generative AI

Suchir Balaji, a 26-year-old former OpenAI engineer who contributed to the development of GPT-4, was found dead in his San Francisco apartment on November 26, 2023. Officially ruled a suicide, Balaji's death has brought intense scrutiny to the burgeoning ethical and copyright controversies surrounding generative AI. He was a vocal critic of OpenAI's practices, arguing that the company's training of large language models involved the unauthorized use of vast quantities of copyrighted internet data, violating the intellectual property rights of countless original creators.

Balaji's concerns were not unfounded. His personal blog detailed the unreasonable nature of OpenAI's data usage. He pointed out that OpenAI collected massive amounts of internet text data, including copyrighted blog posts and books, to train models like ChatGPT. This unauthorized use enabled the AI to generate text highly similar to the original content, infringing on the intellectual property rights of original creators and causing them direct harm. He argued this amounted to creating "substitutes" for original works, stealing the creators' rightful rewards. For instance, if ChatGPT could generate answers comparable to those of top-tier experts on platforms like Zhihu, the platform and the experts' value would be severely challenged, potentially collapsing the entire ecosystem.

OpenAI countered that its use of publicly available data complied with copyright law and was crucial for innovation and American technological competitiveness. However, Balaji was not alone in questioning generative AI's ethics. A growing number of critics point to the serious ethical dilemmas inherent in generative AI's training methods, demanding that regulatory bodies introduce relevant laws to govern the field.

Balaji's death thrust the ethical dilemmas of generative AI into the spotlight. Foreign media commentary noted that his outspokenness on AI ethics garnered both admiration and criticism, highlighting the tension between innovation and responsibility. More importantly, the information he possessed could play a crucial role in future lawsuits against OpenAI.

Indeed, copyright lawsuits against generative AI are already emerging. In late 2023, The New York Times sued OpenAI and its partner Microsoft, alleging the illegal use of millions of original articles to train large language models, arguing this unauthorized use directly harmed their business and predicted ChatGPT would eventually replace traditional journalism. Similarly, in late 2022, three artists collectively sued several generative AI platforms for using their original works without permission to train their models and generate derivative works strikingly similar to the artists' creationsan act considered unauthorized infringement. If courts deem AI-generated works unauthorized derivative works, severe penalties for infringement could be imposed.

In China, similar infringement issues are becoming increasingly prominent. "Magic-modified" videos are a prime example. AI technology allows users to "transplant" characters and scenes from classic films and television shows into entirely new contextsa phenomenon prevalent on major short-video platforms. However, this seemingly entertaining practice involves infringement of actors' portrait rights, reputation rights, and the copyright of the original works. Zhang Bo, deputy director of the Beijing High Courts (Zhengzhou) Law Firm, points out that the face-swapping and script alteration in AI "magic-modified" videos constitute re-creation of the original content, but the pursuit of traffic fosters numerous new types of infringement.

At the 8th China Network Copyright Protection and Development Conference, experts engaged in in-depth discussions on the copyright risks posed by AI. Yi Jiming, director of the Peking University Center for International Intellectual Property Studies, noted that in most cases, AI development companies have not obtained authorization from copyright holders, creating substantial legal risks. Therefore, the industry urgently needs clear legal and public policy guidance on the legality of using data to protect the interests of creators and the public. The legality and reasonableness of data used in AI large model training have become a major issue demanding urgent resolution in the AI era.

However, resolving this issue faces numerous challenges. Firstly, defining AI infringement is difficult, and creating regulatory documents is equally challenging. Copyright law protects works of originality, but whether AI-generated content possesses originality and how to determine it remains complex and contentious. Different countries have different views; for example, the US Copyright Office believes that AI-automatically generated works are not protected by copyright law, whereas Chinese courts have ruled in some cases that if AI-generated content reflects the original intellectual input of a natural person, it should be protected by copyright law.

Secondly, under traditional copyright law, the author of a work is a natural person or legal entity with legal personality, but AI, as the generator of content, has an unclear legal status and cannot directly claim copyright as an author. This requires lawmakers to reconsider the determination of authorship.

Finally, lawmakers need to strike a balance between protecting copyright holders' rights and promoting the development of the AI industry. The dilemma of "regulation leads to stagnation, deregulation leads to chaos" makes regulatory progress extremely challenging.

• Nevertheless, the development of generative AI is unstoppable, and the boundary between AI-generated and human-generated content is becoming increasingly blurred. Perhaps, as predicted by Balaji in his book, The Balaji Prediction: Technology, Truth, and a Guide to Building the Future, blockchain technology, with its immutable characteristics, and incentive mechanisms established through cryptocurrencies and smart contracts, can offer new solutions. Domestic experts also suggest developing more efficient digital fingerprinting technology and applying blockchain technology to give AI-generated works unique digital identities for traceability and protection of the copyright of both human and AI-generated works.

Suchir Balaji's death may only be the "first chapter" in the unfolding drama of AI ethics. The trajectory of technology is not subject to human will. We can only hope that AI will ultimately become a tool for humanity, not the other way around. Achieving this goal requires the collective efforts of society to seek a balance between technological innovation and ethical standards, establishing a sound legal framework and regulatory mechanisms.