Artificial Intelligence Can’t Work Without Our Data. We Should Be Paid for It | Hacker News Reader

Artificial Intelligence Can’t Work Without Our Data. We Should Be Paid for It

politico.com

35 pointsrobg19 comments

> Our proposal is simple, and harkens back to the Alaskan plan. When Big Tech companies produce output from generative AI that was trained on public data, they would pay a tiny licensing fee, by the word or pixel or relevant unit of data. Those fees would go into the AI Dividend fund. Every few months, the Commerce Department would send out the entirety of the fund, split equally, to every resident nationwide. That’s it.

Let's call it something catchy like "Universal Basic Income." And collecting money when companies do stuff? Brilliant. I wonder if it could even be applied more generally... perhaps by taking a percentage of every companies profits? I should write this down...

In all seriousness, I do love UBI, but this seems like a weird way to go about it. Also, AI contributions are worldwide, how's that gonna work?

Andrew Yang's focus groups and a/B testing stated that "freedom dividend" had better reception than UBI or any other phrasing, at least when it came to it being done in the US, it would be very interesting to see the outcomes in other countries

"small, starting at $0.001 per word generated by AI" - this isn't small. GPT3.5 is $0.0015 for 1000 tokens. You're basically making gen AI 750x more expensive. If you write a 1000 word explanation of something that'll be a dollar. Imagine trying to use ChatGPT at this pricing - basically every conversation will be $5-10, so anyone who wants to use it will be coughing up hundreds per month.

Why would it have be cheap in the first place? If you like to splurge on something and surround all your life on ChatGPT workflows then you’d have to pay. For most people with average needs it would not be too expensive.

I would much prefer some effective way to opt out of having my data used for training.

we opted in a very long time ago, the only opt out is pure silence for now

I never opted in. However, you're right about silence being the only defense. It's why I closed all of my websites to the public. But there needs to be a better way to handle this than just removing information from the public web.

One idea I had was seeding my comments and online data with AI generated data as a trap of sorts - I've heard it can degrade the quality, kind of like quicksand on the beach it wants to dig up

Actually no. It should not go to a common fund. The bots read particular things and we know who wrote them. Those people should be able to claim money. No one else.

Large portions of Wikipedia were written by a small group of people. It is not right to take their work and distribute the benefits to anyone anymore than it's right for someone to be able to use it in an ai model for an unlicensed use.

Most likely the best LLM will come out from Hugging Face or a similar community and it will be free.

Nothing works without a reference to something else. Whose data? Who should be paid for this? The original knowledge creator, the thousands of businesses like Politico or blogs that regurgitate and reword their articles from a myriad of sources, the people that pay to host the data that was scraped whether its original or not?

Let's say I actually take the time to write a long form article with months of research and a list of references. I put it online and it, along with a hundred other articles on similar topics gets scraped. That text gets sliced, tokenized, vectorized, and whatnot.

I prompt an LLM that picks a random seed and generates the output. Where did the combination of n words or patterns originate? How can one prove that "this output" was scraped from Politico?

Are there ways to prove this beyond reasonable doubt? And honestly, all discussion of ethics, fairness, enforcement, etc aside; even if some sort of solution came along that somehow blocked bots from scraping data, what is stopping any hacker from pointing a camera at their screen and building a bot that uses advanced OCR to simply pretend to be an end user?

I don't know how to process this future. The one thing that's stood the test of time is that information routes around censorship, for better or worse.

I don't have any answers. Just endless questions about this. I understand content creators want to be paid for the value they produce. I also realize this is a hard problem and like the war on drugs, this war on information is already lost IMO. I can accept what is inconvenient. Like mosquitoes. It was lost the moment the internet was created and every nerd who's hacked since the beginning in the traditional and obscure networks knows this.

Embrace the future and pivot now. What is considered valuable will evolve as it always has. It would be wise for everyone to reevaluate what produces value to other people in the future. Not what produced value in the past. Not what is producing value today, in this weird transition period. But what will produce value and demand in the near future.

We should perform active sabotage.

The Vatican should be able to charge a royalty on every marble sculpture currently produced, because undoubtedly the sculptor has studied The David. Yeah. Totally that.

If individual artists slowly learning from examining individual works were really much like being automated borrowing at scale, we wouldn't be having this conversation.

The scary thing about ai is people willing to compare it to humans. They sound a bit like conspiracy theorists. While it is true that in theory ai can reach human levels of intelligence, it still is software, and software should be treated as such. Pay for the content used or dont use it all unless explicitly permitted by the authors.

Telling software to pay you sounds like you’re comparing it to / treating it as humans

But it is built from something extracted from humans and not only that but that devalues humans afterwards because it’s much cheaper to pay an LLM pennies than others humans a few dollars. The reason LLMs charge pennies is only because they don’t pay anything for the knowledge they are trained on. Seems like the most obvious outcome is for people to feel threatened, close circles and not share, perhaps run private LLMs and harvest by themselves the value they already put in.

> but that devalues humans afterwards

and yet you had no qualms about low pay workers in developing countries doing factory work do you?

> they don’t pay anything for the knowledge they are trained on.

that knowledge was disemminated for free. There cannot be a restriction on knowledge usage, as it is not a right that has been granted to the creator of said knowledge.

The obvious outcome is people get over it and change nothing. Their individual sharing makes no difference.

Reply on news.ycombinator.com