Have they removed copyrighted "training" material or are they still relying on people's hard work which they "learned from" without consent and are selling without permission?
From the end of this announcement (emphasis mine):
"DALL·E 3 is designed to decline requests that ask for an image in the style of a living artist. Creators can now also opt their images out from training of our future image generation models."
Very carefully-worded statement. So...still relying on people's hard work, but on the upside, you get to opt out of having your work be fodder for DALL·E4." </s>
Well OpenAI is consitent in that it consistently tries to monetise other people's work without paying for it and in doing so it leverages the gullibility of the masses to defend their actions. Clever.
An artist using an artbook as reference material = good
An artist with one billion arms, using every artbook ever published, creating millions of images per second in every possible style for fractions of a penny per image, outcompeting every other artist forever = bad
I think I'm sympathetic to that idea, but the ethical consideration there isn't that it violates copyright. It's that it's disruptive to society/the economy.
If the artist stole the artbook from a shop, the problem is not that they are using it as a reference; the problem is that they stole it. Likewise if they download the pdf from a pirate site. It is a separate, different problem. You can tell because there would still be a problem even if they didn't then go on to produce art, or read it at all.
If, however, the rights holder chooses to just give them away; or, y'know, puts them on their own website for anyone to look at - there is then no license fee to be paid for looking.
Note that this still does not mean someone can make copies and sell them. That's a separate right. But using such materials as a reference is just fine, and people do this all the time.
Legally this is likely a non-issue. It depends on if they successfully make the case that their AI learns like a human, and thus any outputs that aren't direct copies of existing art are new creations.
This even applies if the AI copies an artists art style (in the same vein as a human looking at one artists art over a weekend and then being commissioned to paint something in the same style, which is completely legal since you can't copyright an art style; although Adobe would love that[0]).
How would you turn that into legislation though? I can simply draw a few pieces of art in the style of Naoki Urasawa, train a model on it, and claim that the outputs of the model are non-infringing. An artistic style is either copyrightable or not - I don't think a blurry middle ground helps anyone.
Corporations are not literally people in a legal sense. Corporations have some of the same rights as people in certain legal transactions because treating them as entities rather than each individual within a corporation separately is more convenient, and because to do otherwise would in some cases infringe upon the rights of the individual humans who make up the corporation (for example, corporations have a right to free speech because the individuals within that corporation have a right to free speech, and it would be impossible to deny free speech to corporations without also denying it to individuals.)
But systems of law are still capable of recognizing the distinction between the personhood of corporations and of people, just as they can recognize the difference between humans and AI even if AIs can be demonstrated to "learn" the way humans do. As always, context and nuance matter. Laws aren't written or decided upon based on pure logic or calculus but on what human beings want and consider to be in their self-interest.
They haven't. Probably one reason Adobe's AI will beat them out long term.
Also another thing that's been on my mind is I wonder if all this AI generation stuff could cause a Games Industry style crash where due to such a over saturation of highly advertised but meaningless/worthless AI content consumers lose interest and stop spending money in different respective industries (books, ganes, films, digital art, music, etc.) and then they crash.
If there's an AI crash it will be due to the vast number of AI companies with insufficiently differentiated products and subsequent race to the bottom, not due to the ubiquity of the output.
I am not aware of a games industry crash, it would appear that gaming is an industry larger than all other forms of art combined. But indeed niches that are saturated by enshitified content have almost crashed and I get your point. I suppose the average will turn even more average and indeed people will stop spending money on it. AI being a statistical machine it will excell at making whatever is common and plenty and as such those industries will suffer even more. Average music, content writing, drawings, etc, will drop to near zero value, that's guranteed.
The comment you're replying to is referring to the video game industry crash in North America in the early 1980s. Basically the market was flooded with games of poor quality due to a lot of factors, including Atari's complete lack of quality control on games they put on their 2600 console. Nintendo ended up redesigning their Famicom console as the Nintendo Entertainment System with an emphasis on it looking like a VCR as opposed to a cheap game console like NA audiences were used to (the Famicom itself is fairly small and plasticky with permanently attached controllers). Additionally they were strict about licensing development on the system with the goal of fostering a crop of family friendly, high quality games. It was a couple years after the fact (iirc) but Nintendo's efforts to differentiate themselves in the wake of the crash obviously payed off and led to a long period of Japanese ascendancy in the games markets. So the crash cleared out a lot of the market and led to a huge opportunity for Nintendo.
No, this has been directly confirmed by Nintendo developers at various points. For people to even be able to try SMB, they'd have to be convinced that the NES wouldn't suffer the fate of the Atari 2600 and the like. It wasn't just about the VCR-looking design - the marketing materials had very careful wording to avoid associations with the failed game consoles, and accessories like R.O.B. (that never even existed in Japan!) were mostly made to make the NES look like a complex electronic toy, and not a game console.
Truly, I have always hated how human artists have "trained" by looking at other people's art without permission, downloaded those without permission into their meat-brains, and trained their organic neural networks on this art.
You can't do that. It's copyright-maximalist copyright infringement.
The fundamental differences in scale between manual recreation by a human and automated replication by a machine are what led to the creation of copyright law in the first place.
No, that isn't what led to the creation of copyright in the first place. The concept predates Gutenberg's printing press.
Copyright in the United States was drafted into the Constitution as a way of rewarding creators so they could create more.
They're already being rewarded, perhaps too handsomely, there is no need to extend it further. If they persist in trying to take more than they're given, then the public will just need to revoke the privilege. It's not a human right.
Proto-copyright goes back centuries before that. I think around 600AD, King Dermott said "to every cow its calf, and to every book its copy". Then it really picks up in the Middle Ages.
> Prior to the invention of movable type in the West in the mid-15th century, texts were copied by hand and the small number of texts generated few occasions for these rights to be tested. During the Roman Empire, a period of prosperous book trade, no copyright or similar regulations existed, copying by those other than professional booksellers was rare. This is because books were, typically, copied by literate slaves, who were expensive to buy and maintain. Thus, any copier would have had to pay much the same expense as a professional publisher. Roman book sellers would sometimes pay a well-regarded author for first access to a text for copying, but they had no exclusive rights to a work and authors were not normally paid anything for their work. Martial, in his Epigrams, complains about receiving no profit despite the popularity of his poetry throughout the Roman Empire.
> The printing press came into use in Europe in the 1400s and 1500s, and made it much cheaper to produce books. As there was initially no copyright law, anyone could buy or rent a press and print any text. Popular new works were immediately re-set and re-published by competitors, so printers needed a constant stream of new material. Fees paid to authors for new works were high, and significantly supplemented the incomes of many academics.
Incidentally, if you click the link about King Dermott and get to https://en.wikipedia.org/wiki/Battle_of_C%C3%BAl_Dreimhne, it says that's "an account that first appears... nearly a thousand years after the alleged events supposedly took place, and therefore a highly unreliable source".
That ship sailed long, long ago with ImageNet, I'm afraid. All that theft is part of "the economy" now, which means it ain't comin' back. Best we can hope for is a legal decision that says, "AI doesn't make shit. It's all public".
Funny how people complained about chinese factories stealing IP and wanting protection, but now they defend OpenAI and others, and even rejoice at the idea that this will "free society". Not clear to what it will be free to do since many that were left without jobs due to said factories have switched to white collar work, which is now to be stolen by ... the people that complained about china stealing their IP. It's hilarious to watch this mass histeria kickstarted by one single corporation. People are literarily like cattle - you can steer them any direction you wish if you know how.
Funny how there's so many well paid IP lawyers around, and they focused so hard lobbying to extend the copyright to 100+ years for simple copying, but they never had the imagination and creativity that copyright is supposed to be all about protecting, to extend copyright beyond that.
Most of the times this argument is fielded, including this one, it is formulated in the shape of an appeal to a general moral principle. I don't see what this principle is supposed to be: as my analogy shows, there is clearly no general moral principle against learning from copyrighted material and people's hard work without their explicit permission. The more narrow interpretation, in which the claimed principle is that a machine must not learn from copyrighted material (...), is also implausible: since we have no real history of machines learning from copyrighted material in any way that is recognizable as learning, it stands to reason that a principle addressing that scenario can not yet have become general.
The appeal is thus to a completely novel principle that you have come up with for yourself; and it seems that rather than presenting arguments for why others should adopt this principle, you are trying to present it in such a way that someone not paying close attention would be fooled into believing that it is common sense and widely accepted. An analogy with the classic "you wouldn't download a car" comes to mind.
Humans can indeed be taken apart piece by piece and put back together again. We just don't have that level of technology yet. There's nothing physically stopping it from happening though.
You can melt down a lathe and it is quite hard to reassemble it, and even if you did reform the entire thing people would doubt if it is the same lathe.
Humans have had parts removed and reattached. With transplants components have been replaced entirely. There is a point at which you can destruct a machine from which it is impossible to reconstruct without getting into ship of Theseus issues. That point is different for different things.
Sometimes debating the issue with ai fans is like arguing with children. They come up with all sorts of what they think are "clever comebacks" but really all they do is a reduction to absurdity. It only proves the fact that they are indeed children and fail to understand the topic alltogether. In such scenarios is best to leave them be.
Hmm. I don't think so. Whether humans are machines or not is really a matter of faith, not dictionary definitions, I would think.
It was a pithy one-liner about categories in response to a pithy one-liner about categories.
But I'd say the underlying question I'm trying to ask is philosophical: what property do humans have and machines lack that makes the first's learning from copyrighted works acceptable, and the second's unacceptable? (eastof suggested a property below).
