This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.
Jump in the discussion.
No email address required.
Notes -
Define copypasta. Anime-girl, like any "object", can be described as a projection in the plane of some three-dimensional model with superimposed textures and lighting. The girl, as a model, cannot be patented (unless it's a branded object like hello-kitty or Mickey Mouse). But even a branded object is assembled from geometric primitives – spheres, cylinders, triangulated surfaces – covered with commonly available textures and painted in standard colors.
OK, you'll do what they do now - make some verbal description of the style and a set of examples of that style in the form of pictures proprietary. But style is a point in a huge "style space." Some dots are now occupied by authors, some by corporations, some owned by no one (like tumblr noses, etc.) – but the vast majority of style dots are free. Neurobros will simply go and create new styles – or generate new ones as unlike the existing ones as possible by interpolating the old ones. And what do you do? Train a neural network that will assess the degree of coincidence of the author's and copycat's manners of writing? Well, it will start to register all the usual artists as thieves, most of whom will not even be aware that their way of drawing is similar to someone else's patented style – because there are a fuckin' million artists and it's impossible to know them all. As soon as the person starts to draw and posts the first picture, he'll get sued for stealing the identity of some obscure patented artist – because the picture will be similar to this artist's "naive_relism_№9714443" or "primitivism_№540235".
You can't patent a sphere. You can't patent the color blue. You can't patent a girl with a dick. Can't patent the slap. Can't patent the grass, the sky and the sun. The point of Transformers as ML models is that they have no ceiling on the level of abstraction and depth of image analysis (which you, for example, do have: below is receptor resolution, above is a ceiling on the complexity of the patterns being analyzed). It doesn't steal fragments of other people's images, it stupidly parses the world into ontologically real objects like people and trees, or geometric primitives, textures, shaders and lighting, and then mixes them.
The only thing you can do is to forbid using images of authors to train models. But then they would be trained on open datasets – and "styles" would be pulled from publicly available classics or the same interpolation of existing styles. Or, you can just set a reference point in the shape of a model trained on real photos, and define "style" as the sum of the deviations from photorealism, generate a million random deviations and set up a focus group to evaluate them and select a few of the best tens of thousands of them. And all of these styles cannot be considered "stolen" because they will have been created as transparently and honestly as possible.
You hope in vain that Uncle Styopa The Militsioner will come and protect you from nefarious thieves. Or that the Top Skills will raise a stink and show the arrogant technicians their place. But here's the thing: there was no theft. [...]
It's not the picture that's "stolen", it's the idea. How are you going to codify an idea? Here you have Beksinski, for example. You have a collection of his paintings - they are copyrighted not as sets of ideas, but as some instances. Each painting is protected as a unique object, but inside it are thousands of ideas, assembled into an original construction of universal yoba-blocks – both created by Beksinski himself and stolen by him from someone else. In theory, you can break down each picture into blocks, brands, logos and labels – that is, you can select a set of objects as "Beksinski's mickeymouses", select the original "Beksinski textures", the color palette "Zd_Beks_001", etc., and catalog it all in albums as prototypes. Then you can set metrics to estimate the similarity of other people's mickeymouses to prototypes in similar albums. And that's it, that's no more you can do.
Except that there are billions of such prototypes, and most of them are unoccupied and have never been used. A neural network will generate a bunch of mickeymouses and textures and palettes, check that they are far enough away from the prototypes (far enough away that you can't sue them) – and send them to production. Or it may prove that Beksinski himself looks like a lot of other dudes and stole half of his original style from them. And the shit will go down the tubes like never before in the history of art.
Copyright was created accounting for dishonest people and with technical means of editing in mind – but it's powerless against the technical means of the human level, which doesn't screw anyone over but just finds solutions similar to the ones people have the skill to find. And in addition to these already found solutions, it can find a million others.
The problem is, after all, not with the model itself, but only with a questionable dataset in terms of copyright. In a year it will all be the same, but trained not on garbage, but on open and clean data. And no lawyers can do anything about it.
Generally speaking, the main question every artist should ask himself is: what in my profession is really creative and what can be attributed exclusively to technical skills? The sad truth of life is that in the profession of artist, almost EVERYTHING can be attributed to technical skills and tricks around human cognitive bases and visual illusions (aka "academy", "shadier in the shades, lightier in the lights", "work not from the line, work from the blot" and so on), while the real "creativity" comes down to a dozen elementary combinatorial methods. And that the complexity of their production (in the sense of the length of the description of the structure of cultural artifacts – paintings, statues, etc.) is incredibly low and meets a clearly distinguishable upper ceiling, ceiling which generative models do not have at all.
That is, most artists are not creators, but designers, ordinary outsourced corporate servants, aesthetic engineers who «make it pretty» for the masses of workers. And in the long run all this stoned and drunk public will go straight to the streets, because a neural network can "make it pretty" much faster and better than all of them put together. And on top of that, it doesn't drink, it doesn't get high, and it doesn't fuck up deadlines. Unlike them.
Stop hiding behind semantics, wanker. The words «think» and «invent» mean nothing and function solely as mental plugs in everyday psychological discourse. SD works as a semantic archiver – something like a winrar, which effectively compresses data without the possibility of unambiguous retrieval. In terms of statistical learning theory, it reconstructs a hierarchy of probability distributions over a finite sample – that is, it solves the general problem of induction. From the point of view of information theory, it reduces the entropy of the data. From the point of view of computational theory, it splits the original data into data structures and algorithms. From the point of view of algebra – into a basis of initial elements and a basis of elementary operations. The same is done by neural network structures in your own head and social network structures consisting of creators, gatekeepers, fans and patrons analyzing, generating and filtering cultural content.
A neural net doesn't pilfer art. It fundamentally can't generate a single original object from the training dataset because its compression during neural net training was done with IRREVERSIBLE data loss. If you compare the prototypes from the dataset and the generated pictures pixel by pixel, there is nothing in common between them. At the level of complex textures and brushstroke and line techniques, the similarity will already be visible. At the level of gestures and compositions, the similarity will be 50-70%. And at the level of color schemes, it would be as much as 95%. Do you know why? Because artists have been using the same basic set of poses, angles, compositional techniques, color schemes and light sources for centuries. That squealing animal on Twitter didn't invent new poses, angles and lighting methods – he took them from catalogs, from the mainstream or from «how to draw bullshit X» type pulp manuals. And to prove this, just run a semantic search on the original dataset and you'll find tens of millions of supposedly «original» and «copyrighted» images, which in reality are virtually indistinguishable from each other. Because the Internet has long been choked with copypastes and generics. But it's not the neural network to blame, it's the artists who, instead of doing creative search and active experimentation with new styles, have sold their souls for likes and are generating monotonous mainstream millet around the clock.
All of your so-called «anime», from the neural network perspective, is nothing more than a dense clump of pixel distributions, a single
object(Anime)- a limited, dumb, primitive set of visual schemes that activate dopamine receptors. It's not art, it's kawaii engineering. You don't solve creative problems and you certainly don't set them – you're just doing calculus and graphics work at the picture-building university in the department of anime glamour. Real artists are not threatened by anything – their product is authentic, material and forever fixed in real time and space, merging human tragedy, history and destiny. But you are not artists, you are decorators. The «derivativeness» of the neural network is just a mirror reflecting your own derivativeness. And don't blame the mirror if your face is crooked.More options
Context Copy link