US Congressman Adam Schiff introduced a bill proposing that technology companies train artificial intelligence models to reveal the employ of any copyrighted data. This follows criticism from artists and creators that their work is being unfairly exploited by tech companies raking in billions. Could this stifle innovation? The Mint explains.
What does the American bill propose?
The bill, titled Generative AI Copyright Disclosure Act of 2024, proposes that technology companies that create artificial intelligence models disclose the source of their data. So far, companies like OpenAI, Microsoft, Google and Meta have built enormous AI models trained on trillions of data points. However, revealing them was a problem. The bill requires companies to disclose the employ of any copyrighted data when training artificial intelligence models in a centralized “register” 30 days before introducing the relevant model or product. The bill proposes a fine of “not less than $5,000,” which has so far been welcomed.
So you can’t employ any copyrighted work?
The Act does not prohibit the employ of copyrighted works to train artificial intelligence models, but instead aims to compensate original creators for the employ of their works in large-scale commercial artificial intelligence products. Creators around the world, including The Novel York Times, have filed lawsuits against companies such as OpenAI, alleging employ of copyrighted works. Experts say the bill is a positive move and could assist establish a single trading model that everyone could emulate around the world. This could provide companies with greater transparency regarding the commercial employ of AI models, which could raise AI adoption in the future.
If this becomes law, will it set a precedent?
The bill will now be voted on at two levels of the US Congress. The bill will then become law once it receives the U.S. President’s signature. However, even as a bill, policymakers in India believe that the proposals set a robust precedent that can assist India shape its regulatory model for AI development in various fields.
How will this impact AI innovation?
Most stakeholders in the AI ecosystem do not see this bill as stifling innovation. Instead of creating approval bottlenecks for tech companies, experts say the proposed bill could create a benchmarking process that would reveal how reliably an AI model has been trained. For now, training AI models remains a “black box” – one that offers little in the way of definition and disclosure of details. It could also allow creators to be fairly compensated for their work and allow Large Tech to train models without having to worry about lawsuits.
Are there alternatives to copyrighted data?
Adobe only used data it owns in its Firefly generative AI platform. While this shows that there are alternatives to using copyrighted data to train AI, general-purpose AI models require broader datasets. As models get larger, it will become increasingly complex for technology companies to avoid copyrighted data when training their models, which in turn will be used in future commercial AI products. Copyright issues have so far hampered enterprise adoption of generative AI on a circumscribed scale, which is now poised to open up.
