U.S. experts suggest setting up an AI fund: collect data usage fees from large technology companies and distribute them to the public

Source: The Paper

Reporter Fang Xiao Intern Chen Xiaorui

When big tech companies generate output from generative artificial intelligence trained on public data, they pay a small licensing fee per word or pixel-related unit of data. These fees will go into the AI Dividend Fund. Every few months, the Commerce Department distributes the funds equally to every resident of the country.

Other countries could create their own versions, charging similar fees for AI used within their borders. Each country can independently manage their AI policies.

Image source: Generated by Unbounded AI‌

Without the words and photos netizens provide for AI companies to train their models, they have nothing to sell.

**[Editor's note] Last weekend, "Twitter crashed" became a hot search term that attracted global attention, because the company's owner Elon Musk took temporary emergency action to limit the number of tweets that users can read. Musk said Twitter was grappling with "extreme levels of data scraping" and "system manipulation," but he did not reveal who was scraping Twitter's data or explain in detail how the system was being manipulated. Not long ago, the overseas social networking site Reddit also complained about similar problems. **

**The explosion of artificial intelligence has made data scraping a growing problem for platforms. In April, Musk accused Microsoft of "illegal" use of Twitter data, alluding to Microsoft's partnership with artificial intelligence company OpenAI to train artificial intelligence models on "large and diverse text datasets from the Internet." **

** On June 29, Barath Raghavan, a professor of computer science at the University of Southern California, and Bruce Schneier, a researcher at the Harvard Kennedy School and author of "The Hacker's Mind," appeared on U.S. Political News The POLITICO magazine section of the website wrote an article saying that large technology companies should pay them corresponding fees when using Internet users' data to train large language models. The two experts suggested that an artificial intelligence fund can be set up with reference to the Alaska Petroleum Fund, and charge data usage fees from large technology companies, and then distribute it to every citizen in the country. **

澎湃科技（translated and organized this article, slightly deleted.

Alaskans have been finding checks in their mailboxes every year for 40 years, thanks to the black gold (oil) beneath their feet. This is Alaska's permanent fund, funded by the state's oil revenues and paid annually to every Alaskan. Now we're in the midst of another resource boom where companies are marketing bits instead of oil: generative artificial intelligence.

Everyone is talking about these new AI technologies — like ChatGPT — and AI companies are touting their power. But they don't mention how this power comes from all of us. Without all of our words and photos that AI companies use to train their models, they have nothing to sell. Big Tech is taking the labor of the American people and pocketing the proceeds without our knowledge, consent, and authorization.

Your data powers today's artificial intelligence, so you deserve profit, and we have a way to make it happen. We call it "the AI Dividend".

Our proposal is simple, reminiscent of Project Alaska, where big tech companies pay a small licensing fee per word- or pixel-related unit of data when generating output from generative AI trained on public data. These fees will go into the AI Dividend Fund. Every few months, the Commerce Department distributes the funds equally to every resident of the country.

There's no reason to make things more complicated. Generative AI requires a variety of data, which means that all of us are valuable—not just those who are professional, prolific, or well-written. Figuring out who contributed to the text that an AI outputs can be both challenging and invasive because even the companies themselves don't quite know how their models work. Paying dividends in proportion to the words or images people create only incentivizes them to create endless crap, or worse, use AI to create that crap. The bottom line for big tech companies is that they have to pay the fund if their AI models are created with public data. If you are American, you can get paid from the fund.

Hobbyists and small U.S. businesses will be waived under the program. Only big tech companies -- those with significant revenue -- are required to contribute to the fund. They pay for generative AI output, such as from ChatGPT, Bing, Bard, or when embedded in third-party services via APIs.

Our proposals also include a compulsory licensing scheme. By agreeing to pay to this fund, AI companies will receive a license that allows them to use public data when training their AI. Of course, this does not replace normal copyright law. If a model starts producing copyrighted material that goes beyond fair use, that's another issue.

With today's numbers, the situation is that licensing fees can be minimal, starting at $0.001 per AI-generated word. Similar fees would apply to other categories of generative AI output, such as images. That's not much, but it adds up. Since most of the big tech companies have already started integrating generative AI into their products, these fees will mean a few hundred dollars per year in dividends per person.

The idea of paying for data is not new, and some companies have tried to do so for voluntary users. The idea that the public gets rewarded for using its resources predates Alaska's oil fund. But generative AI is different: it uses all of our data whether we like it or not, it’s ubiquitous, and it can be very valuable. If Big Tech were to create a synthetic equivalent of our data from scratch, it would cost a fortune, and the synthetic data would almost certainly result in a worse output. They can't create good AI without us.

Our plan will apply to generative artificial intelligence used in the United States. It also pays dividends only to Americans. Other countries could create their own versions, charging similar fees for AI used within their borders. Just as US companies charge VAT for services sold in Europe, each country can manage their AI policies independently.

Don't get us wrong; this isn't an attempt to kill this nascent technology. Generative AI has interesting, valuable, and possibly transformative uses, and this policy is in line with that future. Even with the cost of the AI dividend, generative AI is cheap, and it will only get cheaper as the technology improves. AI also poses risks—both day-to-day and deep-seated—that governments may need to develop policies to remedy any damage that arises.

Our plan won't ensure that AI advances without downsides, but it will ensure that all Americans share in the benefits—especially because this new technology would not be possible without our contributions.