AI companies often argue that licensing training data is impractical, so they shouldn't have to. But the number of training data deals, and the strength of the AI training data market in general, suggests otherwise.
I wrote a short paper on this. Key points:
- There are at least 21 agreements between AI companies and major media providers that are known to cover AI training (I provide a list)
- There are lots more that are suspected to cover training, but the companies are guarded about the details
- This is just the public deals; as Reuters says, "companies often don’t disclose agreements". We should assume the number is *much* higher than we know.
- There are a number of data marketplaces and other companies making licensing even easier
- There would be even more agreements in place if AI companies hadn't decided to argue training is fair use
- Some projections predict it will be a ~$30B market by 2032
AI companies like to portray the difficulty of licensing as a key reason they shouldn't be required to pay for training data. But licensing is perfectly possible, as is proved by the strength of the current market. And this market will continue to grow - as long as governments don't acquiesce to AI companies' requests to be gifted one of their business' key resources for free.
ed.newtonrex.com/ai-licensing-m
Read 6 replies