AI systems and large-scale language models need to be trained on vast amounts of data to be accurate, but they shouldn't be trained on data they don't have the rights to use. Licensing agreements signed by OpenAI with The Atlantic and Vox last week indicate that both sides are interested in licensing AI training content.
Human Native AI is a London-based startup that is building a marketplace to broker such deals between the many companies building LLM projects and companies willing to license their data to them.
Its goal is to help AI companies find data to train their models while ensuring that rights holders opt in and get paid. Rights holders upload content for free, connect with AI companies, and enter into revenue share or subscription agreements. Human Native AI also helps rights holders prepare and price their content, and monitor for piracy. Human Native AI takes a fee from each transaction and charges AI companies for the transaction and monitoring services.
CEO and co-founder James Smith told TechCrunch that the idea for Human Native AI came from his experience working on Google's DeepMind project, which also ran into problems with a lack of good data to properly train its systems. Smith then saw other AI companies facing the same problem.
“I feel like we're entering the Napster era of generative AI,” Smith said. “Could it be better? Can we make content easier to acquire? Can we give creators some control and compensation? I've always wondered why there isn't a marketplace.”
Smith pitched the idea to his friend, Jack Galilee, an engineer at GRAIL, during a walk in the park with their kids, the way he pitches many other potential startup ideas. But unlike before, Galilee said they should give it a go.
The company was founded in April and is currently running in beta. Smith said that demand from both sides has been very strong and that the company has already signed several partnership deals, which will be announced in the near future. Human Native AI announced a £2.8 million seed round this week, led by two UK micro-venture capital firms, LocalGlobe and Mercuri. Smith said the company plans to use the money to build its team.
“I'm the CEO of a two-month-old company, and I've had the opportunity to meet with CEOs of 160-year-old publishing companies,” Smith said. “That suggests there's high demand in the publishing industry. Similarly, all of my conversations with the big AI companies are moving in exactly the same direction.”
Though it's still early days, Human Native AI appears to be building a piece of infrastructure that's missing in the growing AI industry: Big AI companies need loads of data for training, and giving rights holders an easy way to work with AI companies while still giving them full control over how their content is used seems like a great win-win approach.
“Sony Music has written to 700 AI companies, asking them to stop,” Smith says. “That's the size of the market and potential customers that could potentially get data from. The number of publishers and rights holders could be in the thousands, maybe tens of thousands. That's why we think the infrastructure is needed.”
“I also think this could be even more beneficial for smaller AI systems that don't necessarily have the resources to strike a deal with Vox or The Atlantic and get access to their training data. They would love that, too, Smith said, and all of the notable licensing deals to date have involved larger AI players. He hopes Human Native AI can help level the playing field.”
“One of the big challenges with licensing content is that it costs a lot of money up front and it severely limits who you can work with,” Smith says. “How do you increase the number of people buying your content and lower the barrier to entry? I think that's really interesting.”
Another interesting thing here is the future potential of the data Human Native AI collects: Smith said that in the future, it could give rights holders more clarity on how to price their content based on historical transaction data on the platform.
It's also a good time to launch human-native AI. As AI laws in the European Union evolve and future U.S. AI regulations may be introduced, it will become increasingly important for AI companies to source data ethically and have the receipts to prove it, Smith said.
“We're optimistic about the future of AI and what it will bring, but we have to be responsible as an industry and not destroy the industry that got us here,” Smith said. “That's not good for human society. We have to find the right way to get people involved. We are AI optimists on the side of humanity.”