5 ms·

These don't seem all that difficult to fix to me. Most of the examples are not really generic, but are shorthand descriptions of well-known entities. "Video gam

These don't seem all that difficult to fix to me. Most of the examples are not really generic, but are shorthand descriptions of well-known entities. "Video game plumber" is practically synonymous with "Mario" and anyone that has the slightest familiarity with the character knows this. Likewise, how difficult is it to just use descriptive tools to describe Mario-like images [1] and then remove these results from anyone prompting for "video game plumber"? 1. The describe command can describe an image in Midjourney. I imagine other AI tools have similar features: https://docs.midjourney.com/docs/describe https://docs.midjourney.com/docs/describe

gchamonlive 3y ago
The thing is that those are really trivial or extreme examples. What we should take from this: 1. Generative AI systems are fully capable of producing materials that infringe on copyright. 2. They do not inform users when they do so. So potentially any output could be infringing copyright source material, even from some obscure but still protected corner of the web, and anyone using that output could be exposed to lawsuit risk without warning. This is very hard to fix.
- keiferski 3y ago
  But how is that any different from creating an image from scratch? If I make a logo and use it for my business, but it turns out to be very similar to one already being used by another company, it’s the same situation. I think the main concern here is with the top 1,000 or so brands/copyrights which seem fairly straightforward to deal with using the method I described.
  - gchamonlive 3y ago
    It's plagiarism (https://www.youtube.com/watch?v=yDp3cB5fHXQ https://www.youtube.com/watch?v=yDp3cB5fHXQ). It's not the same situation. You can't possibly expect someone to be exposed to the entirety of the internet like ChatGPT is. It is a matter of scale. If you still think they are the same thing, the industrial revolution was about scale and had transformative impacts in the society.
    - madamelic 3y ago
      ChatGPT is not intelligent. It's the user's responsibility to ensure they aren't infringing on copyright. If you are producing creatives for pay, you absolutely shouldn't be right-click saving images and using directly off the web whether you got it from DALL-E or a Google search. Who cares about plagarism? Plagarism is a made up boogeyman created by high school English teachers. If the work is sufficiently composed of your own thoughts, the fact you used someone else's structure for part of it is not a problem as long as it isn't your entire work or entirely derivative of one work. If I use a Coca-Cola bottle cap as structure in a sculpture, that doesn't mean I am infringing on Coca-Cola's copyright. I still had to mold my original work to work with it.
      - gchamonlive 3y ago
        > Who cares about plagarism? I linked a 4h video on that which goes into painstaking details, which you skipped. It is in the context of content creators but can be extrapolated to various areas. And about intelligence, what do you mean by intelligence? ChatGPT is certainly not thinking like a human, if that is what you mean, but it can reason really well when given directions. That taxes demo when gpt4 launched is certainly a demonstration of emergent cognition in the predictive system. So depending on how you see intelligence, ChatGPT certainly shows a lot of it, if we only let go of human constraints of cognition.
- taberiand 3y ago
  Why isn't the solution a strong disclaimer when the user generates images along the lines of "beware that the images produced may not comply with copyright laws in your country" etc etc? Any artist can privately draw a picture of Mario, what's so different about having an LLM generate that image?
  - RandomLensman 3y ago
    There is a huge amount of copyrighted material in the world and these tools let you create a lot of content in a short time, that would imply an increased risk vs slower, more manual approaches. What I am also wondering is, if for some things copyrighted material could be somehow dominant in the output (beyond prompting more or less for it).
  - gchamonlive 3y ago
    > what's so different about having an LLM generate that image? You don't have to pay 20 bucks a month for a private company to reproduce Mario manually at home for a start... you also can't reproduce Mario in many poses at an industrial scale solo. The list can go on, and we have seen OpenAI change their license agreement, so the risk can change over time.
- numpad0 3y ago
  AIUI/IIUC/IANAL, Generative AI systems stack up tuned butterfly effects to create meaningful outputs. Which means, data is not merely stored inside a butterfly but spread across the system, and potentially every output from the butterfly cage is infringing everything, just almost negligibly. But no one can prove that or otherwise to be the case with current technology. I think it's just unfixable as far as fixing goes. The "best" way is to nuke all GPTs and DALL-Es and bury the technology, and the next best is to mark them all un-copyrightable and un-publishable as a compromise. This second option should be an all-around win that also encourages edge and on-prem deployments, IMO.
  - bbor 3y ago
    The third option would be to change our copyright laws to reflect the new technology, rather than neutering the technology or restricting use of its outputs. Even if that’s the wrong way, it’s important to remember that we’re not inescapably trapped in the status quo
mrweasel 3y ago
It's going to be hard to remove every single "shorthand descriptions of well-known entities" or other prompts that can be used to generate copyrighted or trademarked content. Sure, if you're not deliberately trying to generate infringing content, you can probably remove or discard those results, the trouble is the people who will try to trick the AI to generate this content, blocking those people is going to be impossible, without excluding any copyrighted or trademarked training material. Another issue for generative AI is mentioned in the article: "Systems like DALL-E and ChatGPT are essentially black boxes." What happens when an AI is used to make decisions where the user/victim is entitled to know exactly why the AI did what it did? From a business and legal perspective I think the current AI solutions are dangerous and should be used very sparsely, exactly because even the creators can't point to the exact pieces of information that caused the AI to make the choices it did.
- keiferski 3y ago
  But does this actually matter if the people are only generating images for their own use? Does Photoshop prevent people from making drawings that look like Mickey Mouse? Of course not. I think it will be easy to prevent the obvious copyrighted stuff via the method I mentioned. People going around those restrictions are subject to the same rules as someone drawing the copyrighted image from scratch.
  - mrweasel 3y ago
    > But does this actually matter if the people are only generating images for their own use? Arguably that might actually be a very small issue, but what happens when it happens on a larger scale? Disney might not care, they can easily fight you in court if your DALL-E generated comics looks like Mickey Mouse and Nintendo will make sure that your video game about a electrician named Marvin from the Bronx, but who looks a lot like Mario is never going to get featured on Steam. The issue is the smaller artists that might not have the resources to fight AI content in court. There's also the issue with using LLMs to "white-wash" articles and books. There will be people who will just run articles through ChatGPT and claim that it's AI generated content that was in no way stolen from The Barents Observer. The absolute massive volume, lose in revenue and cost to fight this in court could make running an investigating newspaper impossible and leave us without any actual reporting. Not thinking ahead and having a plan for copyrighted material was an oversight by the current AI companies, but they are arrogant and just assumed that it would be a detail and anyway "disruption" so screw it. There has been zero consideration to the fact that their product is useless without the previous work of millions of people. My concern is that AI takes over content generation to the point where we actually run out of human generated content to train future AIs on. We need to be incredibly careful about implementing AI and ensure that we do not pollute future training data, but people don't care, because they want profit now.
rco8786 3y ago
> Likewise, how difficult is it to just use descriptive tools to describe Mario-like images [1] and then remove these results from anyone prompting for "video game plumber"? This approaches impossibility at scale.
- keiferski 3y ago
  Trademarks already include text descriptions and images of the item being trademarked. This is already in the USPTO database.
  - throwoutway 3y ago
    Using generic text will end poorly. I predict a future where 99 of 100 requests result in Photoshop AI saying "I'm sorry, I can't do that". Google for silly trademarks. Facebook trademarked the word "Home". Star wars trademarked breathing under a mask.
  - rco8786 3y ago
    Exactly. An enormous database of generic, unmoderated text descriptions. Basically any question you ask will be “covered” by some trademark description somewhere. Not to mention that the cost and scale of checking every query against that entire database is … not approachable.
bnralt 3y ago
It seems like a somewhat dystopian thing to fix. Imagine a scenario where Photoshop would scan images you uploaded for copyright material and then refuse to work if it determined image contained any copyrighted material or characters (even if it was just a fan drawing you did). This reminds me of the early days of the internet where people wanted to remove free fanfiction for violations of copyright laws. Trying to apply copyright laws to personal use cases where the creator isn't trying to sell the material is pretty terrible, in my view. Imagine a scenario 50 years from now - "Robot, can you cut out this picture I drew for a school diorama." "Certainly." "And this one as well." "Error: Your picture seems like it might contain some copyrighted materials, and as such I am unable to interact with it."
- keiferski 3y ago
  I doubt that the media companies would be happy about this; but maybe a compromise is a “copyright infringement filter” that can be enabled or disabled, with a flashing notification that you’re responsible legally if you turn off the filter and have issues.
  - atq2119 3y ago
    Sure, there are legitimate but opposing interests here. The solution doesn't have to be technical though. The key part is making sure that the copyright owners still have some recourse, but one that isn't punitive for unknowing infringement. For example, make it legally impossible to impose punitive penalties for unknowing infringement without commercial interest, but make it possible for the rights owner to demand the relevant material be removed etc. Also keep in mind that the Mario examples from the article are perhaps not the best guide here. Mario is sufficiently pervasive in our culture that you can't reasonably claim unknowing infringement. It's the somewhat more obscure cases that I'd be worried about.
- atq2119 3y ago
  It is dystopian, and it already exists, e.g. printers refusing to print anything that looks sufficiently like money. Like many things, I suspect this will end up getting worse before there's a chance for it to get better.
TheRoque 3y ago
How do you know you are inputing "well known entities" if you don't know it beforehand ? If I type "columbian coffee logo" and end up with logos of brands that existed before hand, should I just reverse engineer the whole internet to find out if these logos existed already or not ? The AI should show its inspiration. A human who takes inspiration of something else for its creation knows precisely what it used, and if he crossed the line of plagiarism or not, but the way AI work are too opaque for that. I think the thing it needs to do is reveal its sources, nothing more, but it also means for the AI companies to reveal their dataset, and maybe information they shouldn't have, nor disclose.
bbor 3y ago
Seems insane to try to prevent the model from reproducing content with a blocklist like this - to say the least, it’s more than just Mario. Plus, how would you possibly code common-sense fair use into the model? What’s the difference between a cartoon mouse and Mickey Mouse? What if it’s parody? This seems beyond ridiculous to try to enforce on the tool level.

gchamonlive 3y ago

The thing is that those are really trivial or extreme examples. What we should take from this: 1. Generative AI systems are fully capable of producing materials that infringe on copyright. 2. They do not inform users when they do so. So potentially any output could be infringing copyright source material, even from some obscure but still protected corner of the web, and anyone using that output could be exposed to lawsuit risk without warning. This is very hard to fix.

keiferski 3y ago
But how is that any different from creating an image from scratch? If I make a logo and use it for my business, but it turns out to be very similar to one already being used by another company, it’s the same situation. I think the main concern here is with the top 1,000 or so brands/copyrights which seem fairly straightforward to deal with using the method I described.
- gchamonlive 3y ago
  It's plagiarism (https://www.youtube.com/watch?v=yDp3cB5fHXQ https://www.youtube.com/watch?v=yDp3cB5fHXQ). It's not the same situation. You can't possibly expect someone to be exposed to the entirety of the internet like ChatGPT is. It is a matter of scale. If you still think they are the same thing, the industrial revolution was about scale and had transformative impacts in the society.
  - madamelic 3y ago
    ChatGPT is not intelligent. It's the user's responsibility to ensure they aren't infringing on copyright. If you are producing creatives for pay, you absolutely shouldn't be right-click saving images and using directly off the web whether you got it from DALL-E or a Google search. Who cares about plagarism? Plagarism is a made up boogeyman created by high school English teachers. If the work is sufficiently composed of your own thoughts, the fact you used someone else's structure for part of it is not a problem as long as it isn't your entire work or entirely derivative of one work. If I use a Coca-Cola bottle cap as structure in a sculpture, that doesn't mean I am infringing on Coca-Cola's copyright. I still had to mold my original work to work with it.
    - gchamonlive 3y ago
      > Who cares about plagarism? I linked a 4h video on that which goes into painstaking details, which you skipped. It is in the context of content creators but can be extrapolated to various areas. And about intelligence, what do you mean by intelligence? ChatGPT is certainly not thinking like a human, if that is what you mean, but it can reason really well when given directions. That taxes demo when gpt4 launched is certainly a demonstration of emergent cognition in the predictive system. So depending on how you see intelligence, ChatGPT certainly shows a lot of it, if we only let go of human constraints of cognition.
taberiand 3y ago
Why isn't the solution a strong disclaimer when the user generates images along the lines of "beware that the images produced may not comply with copyright laws in your country" etc etc? Any artist can privately draw a picture of Mario, what's so different about having an LLM generate that image?
- RandomLensman 3y ago
  There is a huge amount of copyrighted material in the world and these tools let you create a lot of content in a short time, that would imply an increased risk vs slower, more manual approaches. What I am also wondering is, if for some things copyrighted material could be somehow dominant in the output (beyond prompting more or less for it).
- gchamonlive 3y ago
  > what's so different about having an LLM generate that image? You don't have to pay 20 bucks a month for a private company to reproduce Mario manually at home for a start... you also can't reproduce Mario in many poses at an industrial scale solo. The list can go on, and we have seen OpenAI change their license agreement, so the risk can change over time.
numpad0 3y ago
AIUI/IIUC/IANAL, Generative AI systems stack up tuned butterfly effects to create meaningful outputs. Which means, data is not merely stored inside a butterfly but spread across the system, and potentially every output from the butterfly cage is infringing everything, just almost negligibly. But no one can prove that or otherwise to be the case with current technology. I think it's just unfixable as far as fixing goes. The "best" way is to nuke all GPTs and DALL-Es and bury the technology, and the next best is to mark them all un-copyrightable and un-publishable as a compromise. This second option should be an all-around win that also encourages edge and on-prem deployments, IMO.
- bbor 3y ago
  The third option would be to change our copyright laws to reflect the new technology, rather than neutering the technology or restricting use of its outputs. Even if that’s the wrong way, it’s important to remember that we’re not inescapably trapped in the status quo

mrweasel 3y ago

It's going to be hard to remove every single "shorthand descriptions of well-known entities" or other prompts that can be used to generate copyrighted or trademarked content. Sure, if you're not deliberately trying to generate infringing content, you can probably remove or discard those results, the trouble is the people who will try to trick the AI to generate this content, blocking those people is going to be impossible, without excluding any copyrighted or trademarked training material. Another issue for generative AI is mentioned in the article: "Systems like DALL-E and ChatGPT are essentially black boxes." What happens when an AI is used to make decisions where the user/victim is entitled to know exactly why the AI did what it did? From a business and legal perspective I think the current AI solutions are dangerous and should be used very sparsely, exactly because even the creators can't point to the exact pieces of information that caused the AI to make the choices it did.

keiferski 3y ago
But does this actually matter if the people are only generating images for their own use? Does Photoshop prevent people from making drawings that look like Mickey Mouse? Of course not. I think it will be easy to prevent the obvious copyrighted stuff via the method I mentioned. People going around those restrictions are subject to the same rules as someone drawing the copyrighted image from scratch.
- mrweasel 3y ago
  > But does this actually matter if the people are only generating images for their own use? Arguably that might actually be a very small issue, but what happens when it happens on a larger scale? Disney might not care, they can easily fight you in court if your DALL-E generated comics looks like Mickey Mouse and Nintendo will make sure that your video game about a electrician named Marvin from the Bronx, but who looks a lot like Mario is never going to get featured on Steam. The issue is the smaller artists that might not have the resources to fight AI content in court. There's also the issue with using LLMs to "white-wash" articles and books. There will be people who will just run articles through ChatGPT and claim that it's AI generated content that was in no way stolen from The Barents Observer. The absolute massive volume, lose in revenue and cost to fight this in court could make running an investigating newspaper impossible and leave us without any actual reporting. Not thinking ahead and having a plan for copyrighted material was an oversight by the current AI companies, but they are arrogant and just assumed that it would be a detail and anyway "disruption" so screw it. There has been zero consideration to the fact that their product is useless without the previous work of millions of people. My concern is that AI takes over content generation to the point where we actually run out of human generated content to train future AIs on. We need to be incredibly careful about implementing AI and ensure that we do not pollute future training data, but people don't care, because they want profit now.

rco8786 3y ago

> Likewise, how difficult is it to just use descriptive tools to describe Mario-like images [1] and then remove these results from anyone prompting for "video game plumber"? This approaches impossibility at scale.

keiferski 3y ago
Trademarks already include text descriptions and images of the item being trademarked. This is already in the USPTO database.
- throwoutway 3y ago
  Using generic text will end poorly. I predict a future where 99 of 100 requests result in Photoshop AI saying "I'm sorry, I can't do that". Google for silly trademarks. Facebook trademarked the word "Home". Star wars trademarked breathing under a mask.
- rco8786 3y ago
  Exactly. An enormous database of generic, unmoderated text descriptions. Basically any question you ask will be “covered” by some trademark description somewhere. Not to mention that the cost and scale of checking every query against that entire database is … not approachable.

bnralt 3y ago

It seems like a somewhat dystopian thing to fix. Imagine a scenario where Photoshop would scan images you uploaded for copyright material and then refuse to work if it determined image contained any copyrighted material or characters (even if it was just a fan drawing you did). This reminds me of the early days of the internet where people wanted to remove free fanfiction for violations of copyright laws. Trying to apply copyright laws to personal use cases where the creator isn't trying to sell the material is pretty terrible, in my view. Imagine a scenario 50 years from now - "Robot, can you cut out this picture I drew for a school diorama." "Certainly." "And this one as well." "Error: Your picture seems like it might contain some copyrighted materials, and as such I am unable to interact with it."

keiferski 3y ago
I doubt that the media companies would be happy about this; but maybe a compromise is a “copyright infringement filter” that can be enabled or disabled, with a flashing notification that you’re responsible legally if you turn off the filter and have issues.
- atq2119 3y ago
  Sure, there are legitimate but opposing interests here. The solution doesn't have to be technical though. The key part is making sure that the copyright owners still have some recourse, but one that isn't punitive for unknowing infringement. For example, make it legally impossible to impose punitive penalties for unknowing infringement without commercial interest, but make it possible for the rights owner to demand the relevant material be removed etc. Also keep in mind that the Mario examples from the article are perhaps not the best guide here. Mario is sufficiently pervasive in our culture that you can't reasonably claim unknowing infringement. It's the somewhat more obscure cases that I'd be worried about.
atq2119 3y ago
It is dystopian, and it already exists, e.g. printers refusing to print anything that looks sufficiently like money. Like many things, I suspect this will end up getting worse before there's a chance for it to get better.

TheRoque 3y ago

How do you know you are inputing "well known entities" if you don't know it beforehand ? If I type "columbian coffee logo" and end up with logos of brands that existed before hand, should I just reverse engineer the whole internet to find out if these logos existed already or not ? The AI should show its inspiration. A human who takes inspiration of something else for its creation knows precisely what it used, and if he crossed the line of plagiarism or not, but the way AI work are too opaque for that. I think the thing it needs to do is reveal its sources, nothing more, but it also means for the AI companies to reveal their dataset, and maybe information they shouldn't have, nor disclose.

bbor 3y ago

Seems insane to try to prevent the model from reproducing content with a blocklist like this - to say the least, it’s more than just Mario. Plus, how would you possibly code common-sense fair use into the model? What’s the difference between a cartoon mouse and Mickey Mouse? What if it’s parody? This seems beyond ridiculous to try to enforce on the tool level.