News for hackers. Stuff that matters.
Re:We can't all use AI. Someone has to generate the training... by donpark on Tuesday March 14, 2023 @10:33PM (#35163001)
Derivative training data is already part of the training regime. Earliest example produced rotated images for more robust image recognition.

IP laws are legal constructs and subject to changes like everything else depending on our needs. When and if enough of us see benefits from AI hugely outweighing benefits from maintaining IP framework, I'm pretty sure laws will be passed to make needed adjustments.

Re:We can't all use AI. Someone has to generate the training... by chitowneats on Tuesday March 14, 2023 @11:51PM (#35163626)
Literal shape-rotating?? I thought it was only the wordcels who were in trouble.
Re:We can't all use AI. Someone has to generate the training... by srcreigh on Wednesday March 15, 2023 @12:51AM (#35164029)
Any automated process for generating training data is part of the model, not part of the training data.

Authentic human generated information has irreducible newness.

Re:We can't all use AI. Someone has to generate the training... by adrianN on Wednesday March 15, 2023 @01:08AM (#35164157)
>Authentic human generated information has irreducible newness

I believe that point is open to debate.

Re:We can't all use AI. Someone has to generate the training... by srcreigh on Wednesday March 15, 2023 @01:30AM (#35164297)
Yep. Another commenter said that likes and clicks is enough for AI to generate new relavant information, not new content.

Think language change every 1000 years. Would AI adapt to that? What new training data would it require? Surely obviously it would require some new inputs. It surely requires internet connected speakers of the developing language.

I think the “surely it would require new inputs” is enough to favour the irreducible newness point.

“Irreducible newness” may be just overhyped “changing relevancy” though, that’s what I’m stuck on.

Re:We can't all use AI. Someone has to generate the training... by xingyzt on Wednesday March 15, 2023 @01:25AM (#35164266)
AI couldn't have started any of the artistic and literary movements, which don't just blend between traditional styles but reject some aspect to forge something new. You can't get AI to generate cubism from training on Victorian art.
Re:We can't all use AI. Someone has to generate the training... by adrianN on Wednesday March 15, 2023 @02:26AM (#35164627)
So do you believe that human brains are somehow magic and don't follow the laws of physics and can't be simulated by Turing machines?
Re:We can't all use AI. Someone has to generate the training... by meheleventyone on Wednesday March 15, 2023 @03:10AM (#35164909)
They’re talking about AI now not some notional future where we can model a real nervous system.
Re:We can't all use AI. Someone has to generate the training... by srcreigh on Wednesday March 15, 2023 @05:44AM (#35165821)
I do. It’s not a very controversial belief except if you only train using Internet comments.
Re:We can't all use AI. Someone has to generate the training... by adrianN on Wednesday March 15, 2023 @09:13AM (#35167828)
It's quite controversial since any magic element in the universe undermines fundamental assumptions.
Re:We can't all use AI. Someone has to generate the training... by srcreigh on Wednesday March 15, 2023 @04:33PM (#35174333)
Physics has plenty of holes. It’s a changing model of measurable quantities in our world. It doesn’t take a lot of brainpower to fit other systems either into the gaps in physics, or independent unmeasurable areas
Re:We can't all use AI. Someone has to generate the training... by adrianN on Thursday March 16, 2023 @02:18AM (#35178774)
Magic violates fundamental principles of our understanding of the world like causality or various conservation laws. You can't just squeeze that into the gaps.
Re:We can't all use AI. Someone has to generate the training... by atlantic on Wednesday March 15, 2023 @10:54AM (#35169255)
You've set up a false dichotomy.

"Human brain is magic." <> "Human brain is not simulatable by Turing machines."

There are other possibilities. I'm sure you can think of a few.

Re:We can't all use AI. Someone has to generate the training... by adrianN on Wednesday March 15, 2023 @11:27AM (#35169796)
I can't think of any. Can you help me?
Re:We can't all use AI. Someone has to generate the training... by Aerroon on Wednesday March 15, 2023 @06:31AM (#35166130)
Why not?

Sure, a style like cubism is unlikely to appear from current AI art generation, but ultimately there's a human still in the loop. The human picks the direction the generation goes in and you can definitely see "styles" or even "characters" appear in certain types of prompts.

If you hang around people that do this a lot you'll see that some of them end up generating hundreds, if not thousands, of images of what seems to be the same character in varying situations. That indicates to me that you could potentially find a type of prompt that would lead to an art movement.

Hell, AI art itself might constitute an art movement.

Re:We can't all use AI. Someone has to generate the training... by salawat on Wednesday March 15, 2023 @01:06AM (#35164130)
See, until then though, you don't get to ignore the law as it is.

You see, you're part of the system, not outside it. If it were not for my deep, deep hatred of the farce we call IP law, I must admit, I'd seriously consider walking over to the other side; because I look at these AI folks scouring training data without so much as asking the original creator for consent, or even sparing a couple cycles to think about whether they should, and it really rankles me.

In fact, the entire damn computing industry is filled with people with some seriously flawed ideas around the concept and mechanics of the act of obtaining consent.

Re:We can't all use AI. Someone has to generate the training... by malwrar on Wednesday March 15, 2023 @01:51AM (#35164408)
> because I look at these AI folks scouring training data without so much as asking the original creator for consent

Consent for… what exactly? Why is anyone entitled to some notion of ownership & power over who may and may not make use of information?

I understand to some degree practical arguments around recouping cost of development and distribution of information, but I don’t think anyone should really concern themselves ethically with what some random person wants them to do with the information. This isn’t like ensuring a sexual partner is an enthusiastic participant, this feels to me more like I’m not going to ask mommy for permission every time I want to play outside (let alone care if I’m told “no”). What practical defense is that anyways?

Maybe my understanding of your framing is off?

Re:We can't all use AI. Someone has to generate the training... by mold_aid on Wednesday March 15, 2023 @07:35AM (#35166651)
>Why is anyone entitled to some notion of ownership & power over who may and may not make use of information?

You mean like copyright holders? Patent holders? Somebody with a trade secret?

>but I don’t think anyone should really concern themselves ethically with what some random person wants them to do with the information. This isn’t like ensuring a sexual partner is an enthusiastic participant, this feels to me more like I’m not going to ask mommy for permission every time I want to play outside (let alone care if I’m told “no”). What practical defense is that anyways?

But yes, anyone should. You're just saying "I don't think people should disagree with me" not "information distribution and consent aren't ethical concerns" - of course they are! Why wouldn't you propose as counterexample a model like medicine's "informed consent" instead of sexual consent or "asking mommy"?

Re:We can't all use AI. Someone has to generate the training... by malwrar on Wednesday March 15, 2023 @03:42PM (#35173671)
> You mean like copyright holders? Patent holders? Somebody with a trade secret?

These are legal fictions that exist because the state’s monopoly on violence. Like I said, I understand the motivation to enable e.g. book writers to seek profit for their work. That’s about where my sympathy ends, especially when those mechanisms are abused for petty control by entities making billions. I hardly view them as ethical constructs, in fact I think the vast majority of restrictions based on these are unethical and immoral.

> Why wouldn't you propose as counterexample a model like medicine's "informed consent"

I think there’s a disconnect here. “Consent” in a sexual or medical context is ethically necessary due to the fact that the asker wants to perform acts directly upon/with the concerned party. Consent to play outside or make use of information does not involve the intimate invasion of one’s person, and really does not need to involve this other arbitrary party at all. This is what I was trying to express in my previous reply.

I don’t see why anyone is ethically entitled under these conditions to insert themselves as a governing authority of how people use information. It’s not even a realistic expectation to hold, information wants to be free and people will do what they want with it regardless.

> You're just saying "I don't think people should disagree with me"

I mean, we are talking about ethics here after all. The best I can do is disagree with your conception of consent and make my own arbitrary value judgement.

Re:We can't all use AI. Someone has to generate the training... by mold_aid on Wednesday March 15, 2023 @05:05PM (#35174703)
>These are legal fictions that exist because the state’s monopoly on violence.

Well, ok I get you there. But they're not really "fictions" when backed by the force of that violence. This critique makes sense, I just don't know if it really can ignore the fact that the legal regime applies in most cases.

Re:We can't all use AI. Someone has to generate the training... by salawat on Wednesday March 15, 2023 @02:43PM (#35172837)
>I don’t think anyone should really concern themselves ethically with what some random person wants them to do with the information.

Did you pay for the product? See, that's what you get under Rights of First Sale. If you pay for it, it's yours. Free and clear, no entanglements whatsoever. You still generally have to ask and negotiate a sale.

>This isn’t like ensuring a sexual partner is an enthusiastic participant.

I wasn't even going to bring it up, but since you opened the door... Someone has to bear significant cost to bring into being something (this is the part you claim to understand, do I'll hold you to your given); you come along and enter a business relationship with them (they are a supplier), set their compensation at non-negotiable 0, use their output as your input, to enable you to do things that they might not be at all comfortable contributing to, even indirectly, which again, you don't even know either way, because you didn't ask. Just saying, you've got more similarity here than you may be comfortable with on further deconstruction. Can't blame you for not wanting to do it, but as someone who has, maybe you should reevaluate?

Besides which, why does everyone think "consent" only applies to sex anyway? Consent is a fundamental underpinning of a liberal government of, by, and for the people.

>This feels to me more like I’m not going to ask mommy for permission every time I want to play outside (let alone care if I’m told “no”).

Live somewhere in the midst of a warzone, political upheaval, or a high crime area, and this could get you shot. In the case you were a child, where your mother has custodianship of you, she bears responsibility for your actions. That you, in that situation, would not care if she said no, just sort of proves my point.

Stakes may be low, but there is a flow to the interaction. If it were not an issue in our industry "it is easier to ask forgiveness than permission" wouldn't be such a sticking point of our industry.

It's reflected everywhere. Even the innocuous places. Our biggest industry actors have made their fortunes on tracking people whether they wanted to be tracked or not. We have browser fingerprinting, which is the equivalent of "hold on a minute, need to figure out what your living room looks like so I can forward the description to all my buddies so we'll know who you are when we next see you." Except that isn't ever outright said or explained to the end user, because if it was, no one would ever consent. Which is why nobody asks and just does, which proves our industry has a problem with consent.

These are exactly the types of thing to get tied in ethical knots over, because damnit, we won't be here forever, and we're setting the example for down the line. While it'd be nice for our practices around IP to be less rigid, there is still a line to be drawn such that it is not okay to just run off with digital info because digital.

Every vice we maintain for ourselves individually, will be abused collectively; likely to a far more destructive degree. Asking first is not a huge ordeal. Lord knows. We might get so tired of being asked we finally figure out some default terms so that people can stop asking. That never happens if no one bothers to get the ball rolling though. You have to ask to get the iteration started.

Re:We can't all use AI. Someone has to generate the training... by 6510 on Wednesday March 15, 2023 @01:01PM (#35171278)
If learning is not allowed the robot overlords can move to countries not on the IP bandwagon like Algeria, Argentina, Chile, China, India, Indonesia, Kuwait, Russia, Saudi Arabia, Ukraine or Venezuela. Those are on the priority watch list, there is also a normal watch list. Those are probably lax enough to allow a lot.

https://ustr.gov/sites/default/files/2019_Special_301_Report...