News Valve Allegedly Bans Games on Steam From Using AI Art

There are several lawsuits inflight right now against the OpenAI folks due to them using unauthorized data from millions of people to their their AI. The folks making the complex data modeling software were so focused on whether they could that they never stopped to them if they should. They operated in a "better to ask forgiveness then permission" model and the consequences of that are starting to take effect.
 
interesting. i understand the concern and why they would have such a policy, however...

my question would be how does Valve know what was used to train the AI ?? seems like it would be impossible to know how the AI was "trained" and on what data set.
They don't know. This is just Valve rightly staying out of the legal Line of Fire. Until the question of content rights is settled via the lawsuits Palladin refers to expect more of this from large corporations. They all have money shaped targets on their backs.
 
There are several lawsuits inflight right now against the OpenAI folks due to them using unauthorized data from millions of people to their their AI. The folks making the complex data modeling software were so focused on whether they could that they never stopped to them if they should. They operated in a "better to ask forgiveness then permission" model and the consequences of that are starting to take effect.
From a practical point of view, how would you go about obtaining permission from the millions of data sources in advance?
 
  • Like
Reactions: phenomiix6
From a practical point of view, how would you go about obtaining permission from the millions of data sources in advance?

Practically doesn't matter, only lawfully. Could I walk into your house and take pictures by merely saying it wasn't practical to get your permission? "Seriously your honor, the guy wasn't in town and I really needed those pictures for this cool project I'm working on."

The data sets they used to train ChatGPT was taken from millions of people without their permission. At the absolute minimum it's big time IP theft, at the worst it's some serious invasion of privacy akin to installing cameras into millions of people's homes.

Everyone is jumping away from OpenAI as fast as possible, nobody wants to be caught up in using stolen goods. Other AI's that weren't trained on data gathered from mass surveillance should be fine.
 
How is what AI does any different than what the human brain does? Yes, this applies to art too. Most artists have styles that are OBVIOUSLY influenced directly by other artists. AI just has an audit trail for how it constructs art. Heck, entire art 'types' are essentially the entire market imitating one successful artist's unique style. All Impressionist art is essentially other artists stealing from Monet.
 
Our (human) thought is create through experience and interaction with the world around us. To claim these thoughts are not my own because they were derived from information gathered from others and the world around me is complete and utter nonense. This is exactly what AI is, it samples the world around it (training) and creates something completely new and of its own.
 
"Artificial Intelligence" is a misnomer. It is not an intelligent, thinking entity. It does not take inspiration from anything or have any thoughts of its own. In its current form, it is a tool owned and used by people to transform one thing into another thing, typically using other people's work in the process. When Stable Diffusion generates an image, it is you using SD as a tool to create that image, and that tool was made illegitimately off of millions of stolen images. Valve is right to ban any game which uses art that was created in such a manner.
 
How is what AI does any different than what the human brain does? Yes, this applies to art too. Most artists have styles that are OBVIOUSLY influenced directly by other artists. AI just has an audit trail for how it constructs art. Heck, entire art 'types' are essentially the entire market imitating one successful artist's unique style. All Impressionist art is essentially other artists stealing from Monet.
Artists take as inspiration to create something new.

AI just smash the trained data together based on keyboards and brute force something that resembles what those keywords mean.

You can and already has happened, that AI guys trained to imitate artists with perfect accuracy.
All because 1) They said the artists were egolatric and deserve to be poor and 2) charged too much for slaving away hours drawing.

And of course.. companies salivate the prospect of never needing to draw or pay humans for content. Why? Just make the AI copy the art of an artist that the director wants.
That is the problem.
 
  • Like
Reactions: palladin9479
you're assuming the only data set is from the web. i have thousands of pics on my hdd that i could use to train such a thing and see what it produces.

i understand how data sets work and what is going on as it "learns". but that does not mean valve has any real idea what was used to train the AI.

hence my question. they can do what they want but it seems to me like they have no way to actually prove what data set was used or not used. not like there is any DNA left behind from the learning they can trace back.....

i'm not a fan at all of AI and personally hate the way it is being shoved down everyone's throat. but that does not mean i can't think about it in an open minded way. so just wondering aloud how exactly one could even know what was used and where it came from, if it is even possible to know.
And where did you obtain those images?
Are those produced by you?
Or you bought them? If yes, does the usage for these pictures give you permission to use on AI?


As for valvle, the point is moot. IF they do not want to get in trouble later.. they are smart to 1) do not enrage creators 2) do not break possible future regulations.
Because you like or not.. 99% of the current AI datasets are indeed used by scrapping the web.
That includes DAL-E, midjourney, etc..
And most of the tools that people use are tied to those datasets and trained data.
 
  • Like
Reactions: palladin9479
From a practical point of view, how would you go about obtaining permission from the millions of data sources in advance?
Whomever trained the image generator used to make the art needs to sort out that problem. There are vast archives of stock photos, for instance. Some of those probably have the legal rights to license them for training AI image generators, although I'm going to bet they'll demand royalty payments for the AI art generated by the resulting model.

It's also a fair bet that some of the online photo sharing websites assume ownership of works uploaded to them. Facebook has used photos uploaded to their site for training facial recognition engines, for example.

On a side note: there seem to be some parallels between this and the legal battles that ensued when sampling music became a "thing", back in the early 1980's. Not that I'm quite that old, but the idea of taking bits of other music and crafting them into something new definitely has some similarities.
 
Last edited:
  • Like
Reactions: palladin9479
How is what AI does any different than what the human brain does? Yes, this applies to art too. Most artists have styles that are OBVIOUSLY influenced directly by other artists. AI just has an audit trail for how it constructs art. Heck, entire art 'types' are essentially the entire market imitating one successful artist's unique style. All Impressionist art is essentially other artists stealing from Monet.
This is mostly where I come down on the matter.

The main difference I see is that I'm pretty sure the current AI image generators are designed mostly to "interpolate" what art they've seen, rather than to try and innovate, as a human could do. However, it's probably only a relatively small leap to design an image generator that tries to come up with new styles and content it's never previously seen.
 
"Artificial Intelligence" is a misnomer. It is not an intelligent, thinking entity.
You're confusing "artificial intelligence" with real intelligence. AI merely refers a system that seems to behave intelligently. In the 70+ years the AI field existed, they never set such a threshold for themselves as the one you're constructing.
(AI) The subfield of computer
science concerned with the concepts and methods of symbolic
inference by computer and symbolic knowledge representation
for use in making inferences. AI can be seen as an attempt to
model aspects of human thought on computers. It is also
sometimes defined as trying to solve by computer any problem
that a human can solve faster. The term was coined by
Stanford Professor John McCarthy, a leading AI researcher.

The famous Turing Test established the first gold standard for AI: that it merely be good enough to fool a human into thinking they were chatting with another person.


It does not take inspiration from anything or have any thoughts of its own.
Creativity isn't anything too mystical. It's not hard to approximate by running some random noise through enough filters to reject the rubbish. At some level, that's even what your brain is doing.
 
  • Like
Reactions: tamalero
it seems to me like they have no way to actually prove what data set was used or not used. not like there is any DNA left behind from the learning they can trace back.....

i'm not a fan at all of AI and personally hate the way it is being shoved down everyone's throat. but that does not mean i can't think about it in an open minded way. so just wondering aloud how exactly one could even know what was used and where it came from, if it is even possible to know.
Heh, by using more AI, of course! There are already numerous detectors designed to identify AI-generated images. Sure, it's something of an arms race, but it's not hard to believe there are some telltale signs, especially if you're suspicious whether a photorealistic image is an actual photo.
 
Practically doesn't matter, only lawfully. Could I walk into your house and take pictures by merely saying it wasn't practical to get your permission? "Seriously your honor, the guy wasn't in town and I really needed those pictures for this cool project I'm working on."

The data sets they used to train ChatGPT was taken from millions of people without their permission. At the absolute minimum it's big time IP theft, at the worst it's some serious invasion of privacy akin to installing cameras into millions of people's homes.

Everyone is jumping away from OpenAI as fast as possible, nobody wants to be caught up in using stolen goods. Other AI's that weren't trained on data gathered from mass surveillance should be fine.
No offense, but that's a really dumb analogy. The images they used were publicly available. No one broke in into somebody's house/server/you-name-it and used personal data to train the AI. If you're going to use a house analogy, this is pretty much similar to Google riding through the public streets and photographing your house's facade for Google Maps.

Moreover, even that analogy is incorrect as none of the used data is being redistributed. So to adjust my Google Maps analogy, it's like a Google employee riding through the streets and then telling a story about a house that may or may not look similar to yours.

Frankly, as a professional tech website representative, you should know better than to make blatantly false exaggerations like that, and that's coming from someone who has their own share of concern toward "AI" and its immediate impacts.
 
  • Like
Reactions: bit_user
Our (human) thought is create through experience and interaction with the world around us. To claim these thoughts are not my own because they were derived from information gathered from others and the world around me is complete and utter nonense. This is exactly what AI is, it samples the world around it (training) and creates something completely new and of its own.

It's not that at all, it's not intelligent or sentient or anything thing remotely close to that. It merely has created a massive data model that links similar data to each other. ChatGPTs for example is a 50 thousand element 12 thousand dimension matrix that contains every word in the English language and it's relationship with every other word in that language. By analyzing input text it can sequence the next most likely word and so forth until it gets an output that is then sent out as text. That's it, that's all there is to it, just extremely good data modeling combined with fantastic mimicry.

All SD and other art tools is doing is analyzing the input words, determine what effects, filters or other material those words most likely mean, then fetching them and performing that operation. It still has to fetch source material, and if the creator does not have the legal rights to that source material, then a copyright violation can take place. Now there are rules about fair use and derivative works, which might eventually encompass AI assisted material, but they aren't there yet.

I suspect that eventually legal rulings will place some limits on how things like Stable Diffusion and ChatGPT can treat source material and what qualifies as fair use and derivative works.
 
Last edited:
The main difference I see is that I'm pretty sure the current AI image generators are designed mostly to "interpolate" what art they've seen, rather than to try and innovate, as a human could do. However, it's probably only a relatively small leap to design an image generator that tries to come up with new styles and content it's never previously seen.

It's not mostly, that is exactly what anything "AI" does now. They have a giant matrix with all the appropriate english words given a numeric value and then within that cell is a list of all numeric values associated with that word. You say "Tree" it thinks "word 26284", then looks up the data associated with that word and find a list of things that are associated with 26284, applies the transformation model with those things and viola AI Art. ChatGPT is similar.

Kyle Hill did a very good breakdown in laymen's terms on what is really happening.

View: https://www.youtube.com/watch?v=-4Oso9-9KTQ



The only thing really breakthrough with this is the development of LLM's and how they mimic how our brains process language. This doesn't make a computer smart, it's just better at understanding natural language as spoken by evil hairless monkeys. Think of it as a good translator between humanese and computerese.
 
The guy making the claims discussed in the article didn't seem to provide any examples of what sort of AI-generated images Valve took issue with, aside from describing them as "a few assets that were fairly obviously AI generated". For all we know, they may have generated images that looked like blatant ripoffs of copyrighted characters or redrawn art from existing games.

And what kind of games does this guy even develop? They were careful to avoid providing any clues about what the game was, going so far as to block out the game's name from the screenshot they posted of Valve's message to them. It might just be a repackaged asset flip with some AI images inserted to hide the fact that they didn't put any effort into developing an actual game. According to them...
The trailer was pretty bad though as it was made from entirely generated artwork, as I planned on just using some quick generations for the first draft and to replace it later with better stuff.
That sounds a lot like shovelware to me, if their entire trailer consisted of AI-generated images.
 
Last edited:
  • Like
Reactions: bit_user
How is what AI does any different than what the human brain does? Yes, this applies to art too. Most artists have styles that are OBVIOUSLY influenced directly by other artists. AI just has an audit trail for how it constructs art. Heck, entire art 'types' are essentially the entire market imitating one successful artist's unique style. All Impressionist art is essentially other artists stealing from Monet.
Exactly. There are two types of artists. Those who make money off of it and those who see it as a hobby/passion.
Incidentially, only the former group is angry......(Yes I made my own research).

There is nothing wrong with A.I generated art if it has been trained on public data. If you post something online like this comment, IT'S PUBLIC. How on earth do people not know this is mind baffling to me ........


The art industry won't have an impact, this will sieve the pool and leave behind people who actually/genuinely love art rather than seeing it as a cash cow.
For example, I would much rather get an oil painting by a real human rather than a mechanical machine into the future.

(Real) Artists will be fine.

A.I generated art will only aid people, who aren't art specialists. I was thinking about creating a game with A.I art/textures too but now with this false takedown, I wonder whether Valve is high. Anime Unity Template garbage is being spammed onto Steam, yet A.I generated textures is not allowed?

Should we even respect Steam as a knowledged platform? Seems more like a decade long terrible student software project, that steals 30% off of developers, which is more than some actually criminals charge....
 
Heh, by using more AI, of course! There are already numerous detectors designed to identify AI-generated images. Sure, it's something of an arms race, but it's not hard to believe there are some telltale signs, especially if you're suspicious whether a photorealistic image is an actual photo.
Actually, there is an invisible watermark added to the AI images like those produced by Stable Diffusion. It's a pretty clever algorithm that can survive quite a lot of manipulation. Here's a write up: https://medium.com/@steinsfu/stable...le-watermark-in-generated-images-2d68e2ab1241
 
  • Like
Reactions: bit_user
Practically doesn't matter, only lawfully.
Fair point.
akin to installing cameras into millions of people's homes.
Now that's taking it too far.

I have trained images of my family in to stable diffusion and created pictures of us around the world wearing traditional clothes for the region. It was a great educational experience for the kids. Back to the point... if I generate a pic of me at the beach, I get some generic background with sand and water and maybe a few rocks because the training data likely had 1000's of pics with the tag "beach" in them. Ditto for the pyramids or ankor wat, etc. No harm no foul as far as that goes. Where I draw the line is when the prompt includes things like "in the style of Greg Rutkowski" or Tomer Honuka, etc. as that's a pretty blatant rip off of an artist's hard work. Even still,

Google already scrapes images, modifies them slightly (makes thumbnails), and reposts them on its search results. They were sued and won. I think AI generate art has a more defensible position than Google's use case.
 
  • Like
Reactions: bit_user and BX4096
Reminds me of that game from a decade ago - Limbo of the Lost which stole assets from a number of games like Thief DS, Wolfenstein, Stalker and HL2
I think they got shut down hard. This is a similar scenario.
Its not just about using questionable assets in the game itself but also about using AI "generated" screenshots and in game footage to sell pre-order copies of the game, which on release could be completely different. This is probably where Valve might get in trouble and why they are blocking it preemptively .