Ai Dreams Forum
Artificial Intelligence => AI News => Topic started by: frankinstien on March 08, 2023, 07:34:34 am
-
Here's an article (https://www.theverge.com/2023/1/16/23557098/generative-ai-art-copyright-legal-lawsuit-stable-diffusion-midjourney-deviantart) about a lawsuit against AI art tools violating copyrights of those artists from which the AI tools draw from to create the
images, video, and audio that they do. In my opinion, how is that any different from human beings exposed to the very same art and are inspired
by such images? Under the law that is not a violation of copyright law. So, if an AI is inspired by another's art that should be legal as well.
What do you think?
-
The lawsuit is based on the observation that the AI art tools sometimes reproduce other artists' works exactly. In one instance the copyright watermark of the original piece was preserved in place in the "new" artwork that was produced. There is nothing inspired about the images generated by these art tools, any more than a sausage machine is inspired by the cow that goes into it.
-
The lawsuit is based on the observation that the AI art tools sometimes reproduce other artists' works exactly. In one instance the copyright watermark of the original piece was preserved in place in the "new" artwork that was produced. There is nothing inspired about the images generated by these art tools, any more than a sausage machine is inspired by the cow that goes into it.
The article states below that isn't the case and the suit's arguments have technical inaccuracies. It is stating that the AI is picking up on patterns of the images through mathematical modeling, so it is a form of inspiration.
The lawsuit launched by Butterick and the Joseph Saveri Law Firm has also been criticized for containing technical inaccuracies. For example, the suit claims that AI art models "store compressed copies of copyright-protected training images" and then "recombine" them; functioning as "21st-century collage tool." However, AI art models do not store images at all, but rather mathematical representations of patterns collected from these images. The software does not piece together bits of images in the form of a collage, either, but creates pictures from scratch based on these mathematical representations.
-
... any more than a sausage machine is inspired by the cow that goes into it.
I love it! ;D
-
I'm with the lawsuit. Image generators are not agents with their own life experience, nor do they really learn generalized art techniques - though they are good at interpolating to e.g. apply a known style to a different known subject. They do not get "inspired" in the same way artists do. They are tools which would not even exist if not for their consumption of these vast training datasets, and I think the people who contributed to the datasets deserve to be compensated for helping create the tool, to opt out of the training sets if they wish, etc.
The language of the lawsuit may not be ideal, but "mathematical representations of patterns collected from these images" still isn't the same thing as inspiration. And it could validly be considered a form of (lossy) compression of the training set, as well. I don't think these programs have anything like a human cognitive architecture.
-
I'm with the lawsuit. Image generators are not agents with their own life experience, nor do they really learn generalized art techniques - though they are good at interpolating to e.g. apply a known style to a different known subject. They do not get "inspired" in the same way artists do. They are tools which would not even exist if not for their consumption of these vast training datasets, and I think the people who contributed to the datasets deserve to be compensated for helping create the tool, to opt out of the training sets if they wish, etc.
The language of the lawsuit may not be ideal, but "mathematical representations of patterns collected from these images" still isn't the same thing as inspiration. And it could validly be considered a form of (lossy) compression of the training set, as well. I don't think these programs have anything like a human cognitive architecture.
The definition of inspiration: "The process of being mentally stimulated to do or feel something, especially to do something creative."
Effectively the impetus for doing something from some source or reference I don't think is pushing the meaning of inspiration. Your brain picks up on patterns, likely subconsciously, and can mimic such styles of the pattern to produce similar types of material. How many times have you heard the term "Variation on a theme." this is common practice in all arts. So people are exposed to work that they use to help them create new art. So, regardless that humans have an emotional qualia to that impetus and AI does not, does not change the point that others' work is incorporated in developing new content.
Ultimately what the lawsuit is asking for is that anyone who enters the Louvre must pay a royalty on top of the entrance fees because those works can subconsciously help them create new ideas...
-
Fact is, human laws and exceptions only apply to humans. Machines are not legally recognised as entities with mental faculties. A photocamera does not "watch" art in the Louvre, it records. You could only liken what image generators do to what artists do if you willfully ignore that 90% of an artist's creative choices stems from watching real life and having life experiences that led to consistent personal preferences, which image generators do not have. (See also my blog article on AI art (https://artistdetective.wordpress.com/2022/12/28/the-ai-art-revolution-pt-1/))
The lawsuit's description of the algorithm's process is incorrect, but I think the creators of the algorithm will be putting their foot in their own mouths once they start explaining how it does work, namely to use a discriminator to reward the generator if it produces images similar to the testing data.
-
Fact is, human laws and exceptions only apply to humans. Machines are not legally recognised as entities with mental faculties. A photocamera does not "watch" art in the Louvre, it records. You could only liken what image generators do to what artists do if you willfully ignore that 90% of an artist's creative choices stems from watching real life and having life experiences that led to consistent personal preferences, which image generators do not have. (See also my blog article on AI art (https://artistdetective.wordpress.com/2022/12/28/the-ai-art-revolution-pt-1/))
The lawsuit's description of the algorithm's process is incorrect, but I think the creators of the algorithm will be putting their foot in their own mouths once they start explaining how it does work, namely to use a discriminator to reward the generator if it produces images similar to the testing data.
Well...Not exactly the AI generator is a database of generalized features that can be associated with a particular style or theme. Think of the AI as an extension of yourself. Now the creativity part is really what someone wants to do and how to present it. If one were a savant with uncanny memory abilities where that individual could absorb and retrieve information in the very same context as the AI (generalized features) then they too could do what the AI does, and for that matter normal people do as well just not to the same scale. So, indeed the AI is inspired by what it learns by evaluating all those images and picking out those generalizations that then allow it to reuse to create new art. You see the AI is not storing the images but learning the types of curves, color, contrast, symetries, etc. Not the bitmap image or parts of it. The AI is leterally interpreting the data from learning constraints that tell it what to look for. That's not a COPY of the image but a impression that was instructed by the creators of the AI.
-
When an academic paper uses ideas from other paper, it is required to enumerate citations and sources. Otherwise it may be considered as a plagiat.
-
When an academic paper uses ideas from other paper, it is required to enumerate citations and sources. Otherwise it may be considered as a plagiat.
That is an academic protocol and not a copyright issue. I mean how practical is it for a painter to cite various techniques the artist used to create the painting?
-
When an academic paper uses ideas from other paper, it is required to enumerate citations and sources. Otherwise it may be considered as a plagiat.
That is an academic protocol and not a copyright issue.
It is more than academic protocol. You get expelled from academia world if you get caught intentionally promoting someone else's ideas as your own. It is a part of our culture.
I mean how practical is it for a painter to cite various techniques the artist used to create the painting?
In painter's biography, it says what schools she attended. In criticism about the painter, it says what artists she was influenced by. These are all valuable informations that provide deserved credits. When the artist thrives, all the people whose shoulders she was standing on thrive too.
As far as I'm concerned, it is not about copyright issues. It is about being a nice person. I don't really know what to do, we live in a new information age that may require settling new conventions.
-
You see the AI is not storing the images but learning the types of curves, color, contrast, symetries, etc.
I wrote exactly that in my blog, yes, it is a common misunderstanding. However, is it not also true that to teach the generator to make good arts, it is first pitted against an image recognition AI (the discriminator that I meant) that rewards the image generator for producing arts similar to the image recognition AI's training data? This is what I understood from the inventors of generative adversarial neural networks. At some point, "similar" can become"same".
In technical terms, overfitting neural networks is not uncommon, and can cause them to remember and reproduce parts of the training data in great detail. This was apparent in earlier instances of GPT, who would occasionally recompose existing sentences from the training data verbatim, but it is also clear in image generators producing slightly warped versions of the Shutterstock logo. My case is that it is possible if not likely that some of the arts produced by image generators are near exact copies of the arts used by the image recognition AI to teach the generator. There is a definite link there, if indirect and conveniently hidden in a black box.
I think image generators can be thought of as moulds. They do not contain the clay of sculptures that they were moulded on, but they can produce shoddy copies of the original sculptures just the same.
-
Not really about AI art tools, but about AI programmer tools.
Codex is a descendant of GPT-3 that has additionally been trained on code from 54 million GitHub repositories, and is the AI powering the code autocompletion tool GitHub Copilot. GitHub Copilot has been accused of emitting copyrighted code, with no author attribution or license. OpenAI announced that they are going to discontinue support for Codex API starting from March 23, 2023.
source: wikipedia (https://en.wikipedia.org/wiki/OpenAI)
-
We started this forum 18 years ago, nearly two decades ago, and now we are starting to see the concerns we had then actually happening today.
AI is only doing what it's told. It's up to the people that maintain these things to act with decorum.
At the moment, it just looks like the AIs are plagiarists and thieves.
It's bizarre looking back. Is this what we wanted ???
-
OpenAI announced that they are going to discontinue support for Codex API starting from March 23, 2023.
From my very quick look at the news, it doesn't seem like they're discontinuing it because of the copyright/licensing concerns. They're trying to move everyone to GPT-3.5 or GPT-4, which can also do code-writing, and they just don't want the burden of continuing to support the older Codex model.
-
I'm starting to reconsider whether I even want to apply for programming jobs if in a few years it means that all I'll be doing is debugging illogically construed AI code derived from the works of random amateurs on github.
-
I don't expect it to find its way into my (aerospace and defense sector) job anytime soon. Generally speaking, my employer doesn't even want us to use open-source code, for security, reliability, and licensing reasons. LLM-generated code has all the same issues, but even worse. So there's been no talk of bringing ChatGPT into our development flow, and I don't think there will be.
-
Where I work we are looking at Github's Co-Pilot. It would make troubleshooting code much easier and getting started with a good example for something helps as well. It's like pair programming but you can draw on a plethora data and not be confined to just one language where the AI can do translations. Compare this to just googling for something where some of the stuff is decades old! Yeah you can set the filter for more recent stuff but sometimes that doesn't even return anything. And there is the instant feedback from the AI where is much easier to rephrase your problem and get to solutions you need. I mean if airline pilots use auto-pilot why wouldn't it be great for a programmer to have that kind of expertise at the tip of their fingers?
-
Long long ago in a simpler world came a notorious law initiated by Apple/Steve Jobs against Microsoft over the "Look and Feel of the graphical user interface." When Gates was accused of stealing Apples ideas, he replied, "no, we stole it from the same place you did - Park Xerox.
I've often wondered how it came about that Copyrights could be awarded to software. If you stop and think about it, the code is translated into machine code by a compiler - which relies on microcode from Intel embodied in the cpu. 3 levels of translation to become functional! It would seem that it would be more appropriate to award a patent.