For 40 years, Alaskans have opened their mailbox to find a check waiting for them, a piece of black gold at their feet. This is Alaska’s Permanent Fund, funded by state oil revenues and paid annually to all Alaskans. We are currently in a resource rush of a different kind, with companies hawking bits instead of oil. It’s generative AI.
Everyone is talking about new AI technologies such as ChatGPT, and AI companies tout their incredible capabilities. But they don’t talk about how that power comes from all of us. You can’t sell anything without our writing and photos that AI companies use to train their models. Big tech companies are now taking the jobs of American citizens and pocketing the proceeds without our knowledge or consent and without licenses.
We have a duty to profit from the data that powers AI today, and we have a way to make it happen. We call this the AI dividend.
Our proposal is simple and reminiscent of Alaska’s plan. When big tech companies generate output from generative AI trained on public data, they pay a small license fee per word, pixel, or related unit of data. These fees will be donated to the AI Dividend Fund. Every few months, the Department of Commerce would divide the entire fund equally and send it to all residents of the country. that’s it.
No reason to make it any more complicated. Generative AI requires a variety of data. This means that we are all valuable, not just those who are professional, prolific, or good writers. Given that even the companies themselves don’t really know how their models work, figuring out who contributed which words the AI outputs will be difficult and invasive. Paying dividends in proportion to the words and images people produce will only incentivize people to generate endless mockery, and worse, use AI to generate that mockery. increase. A key takeaway for big techs is that if their AI models were created using public data, they would have to pay the fund. If you are an American, your salary will be paid from the fund.
The plan waives fees for hobbyists and small businesses in the United States. Only big tech companies with significant revenues will be required to contribute to the fund. And you pay at the point of generated AI output, such as ChatGPT, Bing, Bard, or for embedded use in third-party services via application programming interfaces.
Our proposal also includes a compulsory licensing scheme. By agreeing to pay for the fund, AI companies receive a license to use public data when training AI. Of course, this is not superseded by normal copyright law. It’s another matter if models start creating copyrighted material that goes beyond fair use.
Using today’s numbers, we get: License fees can be as low as $0.001 per AI-generated word. Similar charges apply to other categories of generated AI output, such as images. It’s not a lot, but it adds up. With most of the big tech companies starting to integrate generative AI into their products, these fees represent hundreds of dollars per year in dividend payouts for her.
The idea of paying for data is nothing new, and some companies have tried to do it themselves for opted-in users. And the idea of people being rewarded for their use of resources predates the Alaska Oil Fund. But generative AI is different. AI, like it or not, uses data from all of us, is ubiquitous, and potentially very valuable. It would cost a big tech company from scratch to create synthetic data equivalent to ours, and the synthetic data output would almost certainly be worse. They can’t make good AI without us.
Our plans will apply to generative AI used in the United States. Also, dividends are only issued to Americans. Other countries can create their own versions and apply similar fees to AI used within their borders. Countries can manage their own AI policies, much like an American company collects VAT on services sold in Europe but not in Europe.
Don’t get me wrong. This is not an attempt to stifle this nascent technology. Generative AI has interesting, valuable, and perhaps transformative uses, and this policy is aligned with that future. Even factoring in AI dividend fees, generative AI is going to be cheap, and even cheaper as technology improves. There are also risks (both mundane and esoteric) posed by AI, and governments may need to develop policies to remediate the harm caused.
Our plan does not guarantee that the development of AI will be free of downsides, but we hope that all Americans will share the positives, especially since this new technology would not be possible without our contributions. I guarantee you will.
This essay was written with Bharath Raghavan and was previously published on Politico.com.
*** This is Schneier on Security’s Security Bloggers Network syndicated blog written by Bruce Schneier. Read the original post: https://www.schneier.com/blog/archives/2023/07/the-ai-dividend.html
