New Startup Analyzing AI Outputs to Find Out Where They're Stolen From

Credit's Due

Are generative AI models mass plagiarism machines? Many would argue that they are. For creating products that regurgitate other people's content, AI companies lock down billions of dollars in investment, while the creators whose works were purloined by the machines get nada.

That's the way tech entrepreneur Bill Gross sees it, and he says he has an answer. His new startup, called ProRata, claims it will launch its own chatbot-slash-search engine that will use a patented algorithm to identify and attribute the work used by AI models, and through revenue-sharing deals, make sure that everyone involved gets compensated.

"We can take the output of generative AI, whether it's text or an image or music or a movie, and break it down into the components, to figure out where they came from, and then give a percentage attribution to each copyright holder, and then pay them accordingly," Gross told Wired.

Healthy Diet

ProRata hasn't released its search chatbot yet, but it has already raised $25 million.

Notably, it's also partnered with major media entities including Universal Music Group, The Atlantic, The Financial Times, and Axel Springer, which are all probably feeling a little burned that they didn't see any money from all the content they own being used as training data for chatbots like ChatGPT, not to mention AI image generators like Midjourney.

"It's stealing," Gross told Wired. "They're shoplifting and laundering the world's knowledge to their benefit."

ProRata, by contrast, is the Goody-Two-shoes's AI. The chatbot will only use licensed data, rather than stuff scraped from the web en masse. Not only would this be more ethical — and make content easier to attribute — but Gross argues it would also produce higher quality AI outputs.

"I'm claiming that 70 million good documents is actually superior to 70 billion bad documents," Gross said. "It's going to lead to better answers."

Licensible Solution

ProRata aims to release its chatbot this October. One way it plans to make money is through subscription fees to use its service, generating revenue that will be split fifty-fity with content owners. The long game, though, according to Gross, is that its ethical attribution capabilities will entice big AI firms like OpenAI to adopt the tech into their own AI models, too.

"I'll license the system to anyone who wants to use it," Gross told Wired. "I want to make it so cheap that it's like a Visa or Mastercard fee."

A win for everyone, in theory, because with more AI leaders onboard, the more work that can be fairly credited, and the more money that will go to artists and creators. That is, if the idea takes off.

More on AI: MIT Economist Blasts AI Hype, Says It's Too Dumb to Really Impact Jobs

The post New Startup Analyzing AI Outputs to Find Out Where They're Stolen From appeared first on Futurism.