AI for Artists: Part IV of Swaptik Chowdhury’s Series

The entertainment industry raises a red flag.

Recently, two of the largest names in artificial intelligence, OpenAI and Google, submitted proposals to the U.S. government requesting permission to train their AI models on copyrighted material. The reason? They argue that restricting access to such content could slow down American innovation in AI, especially compared to countries like China, which recently launched a powerful language model called DeepSeek.

These proposals were part of a broader request for input from the White House, which had sought guidance on AI development across various sectors. While the companies cite national security and economic competitiveness, the entertainment industry is raising a red flag. More than 420 actors, directors, and other creatives recently signed an open letter urging the government to maintain strong copyright protections and not to weaken them for the sake of AI development.

This is a crucial moment, and it’s essential to understand why OpenAI and Google are requesting an exemption from copyright and intellectual protection regulations.

Large language models (LLMs), the AI systems behind tools like OpenAI’s ChatGPT, Google’s Gemini, and Anthropic’s Claude, require vast amounts of data to train on. The LLMs learn to “generate” by analyzing how words, phrases, and ideas are structured across massive text datasets.

Over time, the demand for training data, the information AI models learn from, has grown significantly. When OpenAI released its first model, GPT-1, in 2018, it was trained on about 4.5 GB of text from 7,000 unpublished books. In contrast, GPT-4, launched in 2023, is believed to have been trained on around 13 trillion tokens. That is roughly 10 trillion words!

To better understand these terms, let’s break them down: 

Training Data: This is the information that AI models use to learn, much like how actors learn their lines from a script. For example, if an AI is trained to write screenplays, it would be fed an enormous collection of movie scripts. The more diverse the scripts, the better the AI can understand different writing styles and genres.

Tokens: A token is a small chunk of text, such as a word or part of a word. For instance, the famous movie quote “May the Force be with you” would be broken down into five tokens: “May,” “the,” “Force,” “be,” and “with.” The AI processes these tokens one by one to create meaningful output.

Parameters: These are the internal settings of the AI model that guide how it processes tokens and generates responses. Think of them like a director’s instructions to actors. If the director wants a more dramatic performance, the actors adjust their delivery. Similarly, AI parameters can change the tone or style of the output based on their settings, such as creating more formal or casual dialogue. The AI model “learns” or calculates this parameter during the training phase.

AI requires human-created content.

This means that today’s AI systems are not only significantly larger but also require substantially more human-created content to function effectively.

This is especially concerning for artists, who are already in a precarious situation due to recent downturns in Hollywood and the changing macroeconomic environment that is impacting artistic investments. Creators are justifiably concerned that their work is being used without permission and that these AI models could generate content that mimics their voice, style, or likeness, potentially competing for their audience.

Protecting creatives.

The open letter, signed by the actors, directors, and other creatives, makes it clear: The entertainment industry supports over 2.3 million jobs and generates $229 billion in wages in the United States alone. Weakening copyright protections could undermine this cultural and economic foundation.

Fortunately, there have been some promising successes in recent months. In 2024, the SAG-AFTRA union negotiated rules around consent and compensation for the use of AI in film and TV. In California, two new laws protect actors from unauthorized AI replicas. One law ensures that contracts must state whether an actor’s digital likeness will be used. The other bans the commercial use of digital replicas of deceased performers without permission from their estates. These are crucial steps toward acknowledging that AI should not supersede the rights of creators (living or deceased).

As AI companies continue to push the boundaries of what is considered “fair use,” we can expect increased pressure on lawmakers to relax copyright rules. That’s why the open letter matters—and why artists must stay informed and engaged.

At its core, this is not just a legal issue; it’s a creative one. The very material that makes AI “smart” is our collective culture: stories, images, voices, and ideas created by people like you. As tech companies seek more data, the creative industry is becoming a battleground over what content can be used to train AI and how the costs and benefits are shared across the community.

There is good news…

While the challenges are real, it’s not all bad news for artists. AI tools can also offer significant benefits, particularly in automating administrative tasks. From writing grant applications to applying for funding, AI can save valuable time and reduce the stress of paperwork. It can also be a powerful tool for brainstorming new ideas and helping artists explore creative possibilities. In our next piece, we’ll delve into prompt engineering and examine some of the most promising AI tools available to assist artists. Stay tuned!

I’d love to hear your thoughts. This column is meant to be a conversation where we explore AI and its impact on artists together. Feel free to contact me at swaptikchowdhury16@gmail.com with your comments, questions, or suggestions for future topics. I look forward to reviewing your feedback and creating content that is helpful. – Swaptik