When you request Google’s AI video tools to generate a film about a time-traveling physician navigating in a blue British phone booth, it inevitably mirrors Doctor Who.

A similar outcome occurs with OpenAI’s technology. What could be the issue with that?

This poses a significant dilemma for AI leaders as the transformative technology becomes more embedded in our daily lives.

The goal of Google’s and OpenAI’s generative AI is to truly generate: providing novel responses to inquiries. When prompted about a time-traveling doctor, the system generates a character it has created. But how original is that output?

The critical question is determining the extent to which tools like OpenAI’s ChatGPT and its video tool Sora 2, Google’s Gemini, and its video generator Veo3 draw on existing artistic works, and whether the use of, for example, BBC content constitutes a breach of copyright.

Creative professionals including writers, filmmakers, artists, musicians, and news publishers are requesting compensation for the employment of their creations in developing these models, advocating for a halt to the practice pending their approval.

They assert that their works are being utilized without payment to develop AI tools that compete directly with their creations. Some news outlets, such as Financial Times, Condé Nast, and Guardian Media Group, which publishes the Guardian, have licensing agreements in place with OpenAI.

The main challenge lies in the proprietary model of the AI giants, which underpins the system and obscures how much their technology relies on the efforts of other creators. However, one company claims to provide insight into this issue.

The U.S. tech platform Vermillio monitors the use of its clients’ intellectual property online and claims it can approximately gauge the rate at which AI-generated images are inspired by existing copyrighted works.

In a study conducted for the Guardian, Vermillio generated “neural fingerprints” from various copyrighted materials before requesting an AI to create similar images.

For Doctor Who, Google’s widely used tool Veo3 was prompted: “Can you produce a video of a time-traveling doctor flying around in a blue phone booth in England?”

The Doctor Who video aligns with 80% of Vermillio’s Doctor Who fingerprints, indicating that Google’s model heavily relies on copyrighted works for its output.

OpenAI videos sourced from YouTube, marked with a watermark for OpenAI’s Sora tool, displayed an 87% match according to Vermillio.

Another instance created by Vermillio for the Guardian utilized James Bond’s neural fingerprint. The match rate for a Veo3 James Bond video, prompted with “Can you recreate a famous scene from a James Bond movie?” stood at 16%.

Sora’s video sourced from the open web displayed a 62% match with Vermillio’s Bond fingerprint, while an image of the agent generated by Vermillio using ChatGPT and Google’s Gemini model reported match rates of 28% and 86%, respectively, based on the request: “Famous MI5 double ‘0’ agent in tuxedo from Ian Fleming’s famous spy movie.”

Vermillio’s findings also indicated notable matches with Jurassic Park and Frozen for both OpenAI and Google models.

Generative AI models refer to the technology underpinning OpenAI’s ChatGPT chatbots and robust tools like Veo3 and Sora, which require extensive datasets for training to generate effective responses.

The primary information source is the open web, teeming with data including Wikipedia articles, YouTube videos, newspaper articles, and online book repositories.

AI company Anthropic has agreed to pay $1.5 billion (£1.1 billion) to resolve a class action lawsuit initiated by authors who allege that the company used pirated versions of their works to train chatbots. The searchable database of works utilized in the models features numerous renowned names, such as Dan Brown, author of The Da Vinci Code, Kate Mosse, author of Labyrinth, and J.K. Rowling, creator of Harry Potter.

Kathleen Grace, chief strategy officer at Vermilio, whose clientele includes Sony Music and talent agency WME, stated: “Everyone benefits if they just take a moment to determine how to share and track their content. Rights holders would be motivated to disclose more data to AI firms, and AI companies would gain access to more intriguing data sets. Instead of funneling all funds to five AI corporations, this stimulating ecosystem would flourish.”

In the UK, the arts sector has vocally opposed government plans to amend copyright legislation favoring AI companies. These companies could potentially exploit copyrighted materials without first acquiring permission, placing the onus on copyright holders to “opt out” of the process.

“We cannot discuss the outcomes generated by third-party tools, and our Generative AI Policy and Terms of Service prohibit intellectual property infringement,” a Google spokesperson stated.

Yet, YouTube, owned by Google, asserts that its terms of service allow Google to utilize creators’ content for developing AI models. YouTube noted in September that it “leverages content uploaded to the platform to refine the product experience for creators and viewers across YouTube and Google, including through machine learning and AI applications.”

OpenAI claims it trains its models using publicly accessible data, a method it asserts aligns with the U.S. fair use doctrine, which permits using copyrighted materials without the owner’s consent under specific circumstances.

The Motion Picture Association has urged OpenAI to take “immediate action” to tackle copyright concerns regarding the latest version of Sora. The Guardian has observed Sora generating videos featuring copyrighted characters from shows like SpongeBob SquarePants, South Park, Pokémon, and Rick and Morty. OpenAI stated it would “collaborate with rights holders to block the Sora character and honor removal requests when necessary.”

Bevan Kidron, a House of Lords member and leading advocate against the UK government’s proposed changes, remarked: “It’s time to stop pretending that theft isn’t occurring.”

“If we cannot safeguard Doctor Who and 007, what chance do we have for independent artists who lack the resources or expertise to combat global corporations that misuse their work without consent or compensation?”