Brought to you by:

Type a carefully worded prompt into an AI chatbot, and you might receive an impressive story—sometimes even enhanced with images and music. Ask it to write in the style of Tolkien, Shakespeare, or Thomas Hardy, and it rarely disappoints.

How do these AI systems so seamlessly replicate the voices of renowned writers, painters, and musicians? And why is the debate over their data sources growing more intense as models evolve?

Both questions boil down to how AI companies train their models, a process that directly impacts future performance. To achieve standout results, a chatbot needs a wide variety of training inputs—and not everything can simply be purchased or scraped off the web.

Data: The Building Blocks of AI

Imagine an AI model as an inquisitive toddler. The more experiences it has, the smarter it becomes. This aptitude for learning from existing data makes AI systems voracious consumers of information: the more diverse their training data, the better they tend to perform.

That’s where the complication arises. To deliver high-quality responses, AI chatbots require diverse datasets—often not readily available for purchase or straightforwardly scraped online.

For example, training an AI to recognize human faces demands millions of images of people in various contexts, attire, and settings. To differentiate faces across cultures and regions, the dataset must include a broad array of skin tones and facial features.

But it’s not just about volume. Before data is fed to an AI model, it must be meticulously “cleaned” to remove inaccuracies, duplicates, or bias. Once processed, the data is used to calibrate how well the model responds during training. Simply put, richer and cleaner datasets lead to stronger performance.

Where Does Data Come From?

Now that we understand how an AI model is trained, the next question is where companies find this diverse, accurate, and unbiased data.

Most AI firms start by tapping into publicly available datasets. These can be text, images, and videos sourced from the internet—often at minimal cost, which helps manage overall training expenses. An interesting example is how companies like Apple, Meta, Anthropic, and Nvidia have leveraged movie and TV dialogue (extracted from subtitle files) to train Large Language Models (LLMs).

If you want an AI to write like Shakespeare, for instance, you’d need access to all of his works—just one illustration of how specialized tasks create an ongoing scarcity of data. Beyond publicly available sources, some organizations also pay for licensed, proprietary datasets. This extra expense can be worth it for more niche or high-quality information.

However, friction arises when AI firms use copyrighted content—such as news articles, analyses, and creator-generated material—to train their systems. Publishers often object, arguing that their content is being used without permission or proper compensation.

The four primary sources of AI training data—each with unique benefits, costs, and challenges.

Publishers vs. AI

Ever since ChatGPT made waves in 2023, questions have been swirling about the data used to train OpenAI’s models. Most AI companies remain tight-lipped about these details, citing competitive advantage and proprietary processes.

Nonetheless, the issue has become a flashpoint in legal and ethical discussions, evidenced by the rising number of lawsuits from content creators, publishers, and artists. For instance, Thomson Reuters filed suit against ROSS Intelligence for using Westlaw’s proprietary content to train an AI-driven legal research tool.

In another high-profile case, The New York Times sued OpenAI (and Microsoft by extension) over the alleged misuse of millions of its articles without authorization. Interestingly, News Corp—owner of publications like The Wall Street Journal and, at one point, The New York Times—later struck a deal with OpenAI to share content for AI training.

Meanwhile, media outlets in Canada and India have also taken legal action, accusing OpenAI of scraping their content without consent. The core dispute revolves around defining “fair use.” AI companies maintain that web scraping falls under fair use, but publishers dispute this, emphasizing the resources they’ve devoted to building online archives.

In reaction to these controversies, many publishers have amended their website Terms of Use to curb unauthorized data scraping. Notably, Meta faced public scrutiny for utilizing “shadow libraries”—essentially pirated books—to train its models. Cases like this underscore ongoing questions about what truly constitutes “fair use.”

Is There a Solution?

As AI companies push the boundaries of technology and train ever-larger models, their appetite for diverse and ethically sourced data will continue to escalate. One way forward is through licensing partnerships that ensure legal clarity, fair compensation, and mutually beneficial relationships between data providers and AI developers.

Another promising pathway is leveraging synthetic data—artificially generated datasets designed to safely bridge gaps where real-world information is scarce, costly, or sensitive. As regulatory scrutiny intensifies and data rights debates evolve, synthetic data will increasingly serve as a valuable, ethically sound alternative.

Ultimately, the future of AI won’t simply rely on bigger datasets or more powerful algorithms—it will depend on thoughtful, responsible, and strategic choices about how data is sourced and utilized.

Building Your AI Team

At Athyna, we believe strongly in empowering companies with ethical, world-class talent helping them navigate these critical decisions and build teams capable of shaping an AI-driven future that benefits everyone.

Is your company ready to responsibly navigate the future of AI? Athyna connects you with world-class talent who can help you harness AI ethically and strategically.

Sponsored by Athyna. We have equity in the company.

*Disclaimer: The Bay Area Times is a news publisher. All statements and expressions herein are the sole opinions of the authors or paid advertisers. The information, tools, and material presented are provided for informational purposes only, are not financial advice, and are not to be used or considered as an offer to buy or sell securities; and the publisher does not guarantee their accuracy or reliability. You should do your own research and consult an independent financial adviser before making any investments. Neither the publisher nor any of its affiliates accepts any liability whatsoever for any direct or consequential loss howsoever arising, directly or indirectly, from any use of the information contained herein. Assets mentioned may be owned by members of the Bay Area Times team.

Please read our Terms of Service and our Privacy Policy before using Our Service.*

If you want to learn more about Athyna, especially how they can help your company, click here to receive more information.

—

*Sponsored by Athyna. We have equity in the company.

Disclaimer: The Bay Area Times is a news publisher. All statements and expressions herein are the sole opinions of the authors or paid advertisers. The information, tools, and material presented are provided for informational purposes only, are not financial advice, and are not to be used or considered as an offer to buy or sell securities; and the publisher does not guarantee their accuracy or reliability. You should do your own research and consult an independent financial adviser before making any investments. Neither the publisher nor any of its affiliates accepts any liability whatsoever for any direct or consequential loss howsoever arising, directly or indirectly, from any use of the information contained herein. Assets mentioned may be owned by members of the Bay Area Times team.

Please read our Terms of Service and our Privacy Policy before using Our Service.