California bill (AB 412) would effectively ban open-source generative AI

[Argyle King]

I don't really understand why you would bother running thought experiments about a bill and the impact it might have without reading it.
It's pretty short, take a look at it.

https://legiscan.com/CA/text/AB412/id/3100490

Because

1) doing that thought excercise first would allow for better forethought into how something ought to be written so as to avoid potential issues and abuses of the law's intent. Writing a law without thought toward what it will actually do once implemented seems a poor way to govern.

and

2) My previous post were done while multitasking and unable to read the law.

That being said, some of the thoughts I had do appear to be reflected in the wording. How "AI" is defined within the law is somewhat vague in Secton 1, 3115, definition 'a'.

"(a) 'Artificial intelligence' or 'AI' means an engineered or machine-based system that varies in its level of autonomy and that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments..."


A 'machine-based system' (which can mean a wide variety of things) that can use inputed information (of any type of input) to generate an output that can Influence physical or virtual environments covers a lot of territory as a definition.

Later in the law, the definition of "generative AI" provides more guidance.


"(d) 'Generative artificial intelligence' or 'GenAI' means an artificial intelligence system that can generate derived synthetic content, including text, images, video, and audio, that emulates the structure and characteristics of the system’s training data."


That narrows it down some, but I can still imagine ways in which that could be construed to mean it covers things that I don't believe this law is intended to cover.

 

[dbolack]

This is about generative AI. AI for research works differently and is trained on raw data -pure "facts"- which isn't copyrightable. With this on the books no big company can stop a small competitor any more than they can already -through the very broken patent system-.

But sometimes it is. I used to work for a place before this AI boom that was on the front end of using large models for recognition of patterns in images for sciency purposes. The training data was largely highly proprietary digital imagery owned by the client. That is very copyrightable.

 

[dragoner]

"AI" is great as tool for sifting databases such as medical x rays looking for cancer, a marvel of advanced programing. However when it is misapplied to areas where it should be not, that is detrimental to the general welfare. Such as facebook wanted to open a datacenter for AI south of here, and build a pipeline that would take millions of gallons of water from the river, an absurd amount, we voted no, and lately it has been found the project was filled with corruption, including involvement by state officials. Sometimes the simple wisdom is to say no, especially if all it is, is enriching a billionaire at the expense of common people, for no actual good reason.

 

[Epic Meepo]

Thank you for linking to the actual proposed law!

So basically it's saying that if you train an AI on copyrighted materials, you have to list the copyrighted materials you used. That seems like a reasonable thing to do.

The proposed law also creates an obligation for any company that makes GenAI available to Californians. That company must respond any time it receives a written request from a registered copyright owner, even if the company never used any of that individual's copyrighted material. And each time the company fails to respond to one of these written requests within 30 days, the copyright holder can sue them for $1,000.

That seems a bit farcical. It sounds as though someone nefarious could register a bunch of copyrights on original material no one cares about and then send a crate of 30,000 printed, written requests to every AI company doing business in California. Any company which lacks the resources to process 1,000 printed requests per day is on the hook for $1,000 for each request they fall short.

I would think a better law requiring disclosure of copyrighted training data would instead require each company to maintain an accurate, searchable database of the copyrighted material they used as training data, with no mention of written requests from copyright holders. The company is already hosting a GenAI service, so asking it to also host a database is less intrusive in asking it to set up a mailroom to handle a potential deluge of written requests.

 

[Bohandas]

Oh I tried, but I had to stop at "A.I. Training Is Like Reading" because that's just false. But I'll try again...

They're right about it being a gift for big tech. But then again: isn't the cat already out of the bag and et cetera?

So the only thing this will accomplish is screwing over the common man. That is the one and only difference it will make. Why would you want to take something that could be a boon to everyone and turn it into something that only benefits the 1%?

 

[Bohandas]

The proposed law also creates an obligation for any company that makes GenAI available to Californians. That company must respond any time it receives a written request from a registered copyright owner, even if the company never used any of that individual's copyrighted material. And each time the company fails to respond to one of these written requests within 30 days, the copyright holder can sue them for $1,000.

That seems a bit farcical. It sounds as though someone nefarious could register a bunch of copyrights on original material no one cares about and then send a crate of 30,000 printed, written requests to every AI company doing business in California. Any company which lacks the resources to process 1,000 printed requests per day is on the hook for $1,000 for each request they fall short.

Yeah that could drive the price up so high it could become as overpriced as human writers and artists, completely completing the cost-cutting benefit of this technology and making bespoke writing and artwork once again the exclusive purview of the rich elite who can afford to throw their money away.

I can't afford to comission an artist! I have car payments and medical bills and a crappy low wage job! But I can easily afford a subscription to artbreeder and NovelAI. Why should I be denied that just because I'm not rich

 

[Bohandas]

You're talking as if it was a bad thing. You know, there's always public domain stuff like Guttenberg library?

Most of the stuff that's copyrighted now should be public domain but the big media companies keep bribing congress to extend the duration of copyrights. If it lasted a reasonable amount of time like patents do I'd be with you, but not under the farcical current laws where copyrights can easily last two entire lifetimes

 

[Jfdlsjfd]

So the only thing this will accomplish is screwing over the common man. That is the one and only difference it will make. Why would you want to take something that could be a boon to everyone and turn it into something that only benefits the 1%?

Increased dividends for shareholders. Restricting open-source AI effort can lead to narrowing the field and let only a few commercial providers survive, allowing them to ensure the price stays high enough for "fair" benefits to be reaped, even if otherwise anyone could get the same result for free. The 1% makes people dream of being part of that income bracket, so it's not surprising it is widely supported. High, 19th century inequalities is a desirable goal if you think you'll be closer to a mine or factory owner than a mine or factory worker.

 

[Bohandas]

Increased dividends for shareholders. Restricting open-source AI effort can lead to narrowing the field and let only a few commercial providers survive, allowing them to ensure the price stays high enough for "fair" benefits to be reaped, even if otherwise anyone could get the same result for free.

How is artificially inflating the price and eliminating competition fair? That sounds almost as bad as where we are now.

While we're at it, how could having to pay ANYTHING AT ALL for something that otherwise anyone could get for free ever possibly be fair or just?

My ideal scenario for this tech* is that home computing processors catch up to it, everyone runs it locally on their own machine, and the entire supply side of the equation just vanishes into oblivion. No capitalists, no workers, no businesses, ONLY the consumer, autonomous and supreme

*and ultimately for all technologies and industrial sectors, but this is the only one where it's currently within reach

 

[Jfdlsjfd]

How is artificially inflating the price and eliminating competition fair? That sounds almost as bad as where we are now.

It's not fair. I said it can be a goal. If you want to get rich, you don't want competition. Erasing competition through legal barriers of entry is a way to achieve that. Fair? Probably not. Possible? As long as a majority people support bills enacting those barriers. Effective? Sure.

While we're at it, how could having to pay ANYTHING AT ALL for something that otherwise anyone could get for free ever possibly be fair or just?

Honestly, providing a clicky webservice instead of having to go through the complexity of installing a piece of software at home is convenient enough that several companies exist doing just that, including running the exact same models than anyone can run at home. Nobody is having to pay, but many chooses to, for convenience.

My ideal scenario for this tech* is that home computing processors catch up to it, everyone runs it locally on their own machine, and the entire supply side of the equation just vanishes into oblivion. No capitalists, no workers, no businesses, ONLY the consumer, autonomous and supreme

My guess is that even if your watch was able to run generative video models, there would still be people who'd be customers for having it available without effort. Anyone can cook, yet ubereats exist. How much is anyone accepting to pay for something they can do at home is the big question.