I’m on board with most of this but This seems somewhat pointless and may even be harmful if it gives people a false sense of security. In fact, I think the primary beneficiary of this type of...
I’m on board with most of this but
Ensure manufacturers of deepfake technology comply with proposed disclosure and watermark laws
This seems somewhat pointless and may even be harmful if it gives people a false sense of security. In fact, I think the primary beneficiary of this type of watermarking technology would actually be deepfake technology vendors who would be more easily able to filter deepfakes from its training data, helping to mitigate the human centipede issue.
100% agree. AI Watermarking reminds me of the Do Not Track HTTP header, which was well-intentioned but ultimately boiled down to trusting bad actors to do the right thing. And it had the...
This seems somewhat pointless and may even be harmful if it gives people a false sense of security.
100% agree. AI Watermarking reminds me of the Do Not Track HTTP header, which was well-intentioned but ultimately boiled down to trusting bad actors to do the right thing. And it had the unintended effect of adding one more bit of fingerprintable user data for those bad actors to capitalize on.
I don't know exactly how to solve this conundrum but it's trivial to remove every form of watermarking that I'm familiar with. I could imagine some kind of cryptographic signature scheme that proves the provenance and authenticity of an image... it wouldn't guarantee it wasn't AI-generated, but it could verify the creator of the file. Combined with some sort of reputation system we would at least have a better idea of whether a given image is credible. But most non-techie folks are pretty resistant to figuring out this stuff so I'm skeptical it could ever happen.
For all the drama over blockchain technologies, provenance is really the one thing it is good at (because at the end of the day it's just an authenticated ledger). If transactions on the chain...
For all the drama over blockchain technologies, provenance is really the one thing it is good at (because at the end of the day it's just an authenticated ledger). If transactions on the chain weren't so expensive a distributed catalog of real and fake images could be compiled. Unfortunately I expect the overhead and necessary scale to be prohibitive.
And more importantly: It can't have any speculative value outside of being a ledger. Proof of Stake isn't that bad provided you're not competing with an infinite potential userbase expansion with...
And more importantly: It can't have any speculative value outside of being a ledger. Proof of Stake isn't that bad provided you're not competing with an infinite potential userbase expansion with "get rich scheme" as a background theme.
TBH a blockchain with the national governments as the only nodes, using it as a copyright ledger, will work much better than many other things. A closed loop with 200-500ish nodes maximum, clearly identified by owner, will be much more efficient than anything fully zero-trust.
I do generally agree. I think the main thing that it offers is the kind of multimaster replication between disjointed entities that don't explicitly have power over one another. Do you trust a...
I do generally agree. I think the main thing that it offers is the kind of multimaster replication between disjointed entities that don't explicitly have power over one another.
Do you trust a database run exclusively by the US government? What about copyrights in Germany? Whom "owns" an international copyright ledger?
The proper answer is the UN, and having it be hosted by individual members provides more auditability and less able to be manipulated by one dominant member that ignores UN rules whenever it suits them....like the USA.
This all ignores the underlying issue that copyright is determined by the courts. The blockchain doesn’t stop me (or the USA) from lying about copyrights, it simply allows us to record our...
This all ignores the underlying issue that copyright is determined by the courts. The blockchain doesn’t stop me (or the USA) from lying about copyrights, it simply allows us to record our statements. Dispute resolution can’t live entirely inside the blockchain, no matter how we structure the smart contracts, especially because those smart contracts will have to be agreed to by none other than the participants.
Unless I’m missing something, using the current system, most citizens are not willing – nor able – to quickly verify a given digital image’s authenticity; an image which may or may not be cropped,...
Unless I’m missing something, using the current system, most citizens are not willing – nor able – to quickly verify a given digital image’s authenticity; an image which may or may not be cropped, a (re-compressed) screenshot, etc., i.e. it is not a given that there even exists a pixel-perfect image match, even before we’re talking about AI-generated images.
Is that not reason enough for having a “verifiable origin” system for digital images?
Edit: reading further along in the thread, I tend to agree with what you said here – a “records of everything” does sound like unfeasibly high effort, when some form of creator + key = signature would probably work too.
Agreed, having the technology coopted by finance bros was probably the worst thing that could have happened to the tech. There is a lot if really interesting mathematics around proof of stake and...
Agreed, having the technology coopted by finance bros was probably the worst thing that could have happened to the tech. There is a lot if really interesting mathematics around proof of stake and zero knowledge proof systems.
Except it’s not really all that good at provenance. If I want to claim that I made something, what I should do is give that object a signature with my key. Now if you have that object and...
Except it’s not really all that good at provenance. If I want to claim that I made something, what I should do is give that object a signature with my key. Now if you have that object and signature you can verify it, without me first publishing a database of every thing in the universe that I ever made.
Gotta say, if I was gonna list off which company I expected to make me aware of this act, I wouldn't have guessed Steak-umms in a billion years. Your marketing department is genius Steak-umms.
Gotta say, if I was gonna list off which company I expected to make me aware of this act, I wouldn't have guessed Steak-umms in a billion years.
I honestly only every knew about them due a throw away line in an early episode of Family Guy. I wasn't even sure if it was a brand or just some Rhode Island region way of cooking steak. So yea,...
I honestly only every knew about them due a throw away line in an early episode of Family Guy. I wasn't even sure if it was a brand or just some Rhode Island region way of cooking steak. So yea, very surprised by this one.
I wonder what a transformative justice approach to deepfakes and other AI harm risks are. If we pursue a legislative response to these threats, is there a risk that we concentrate control of AI in...
I wonder what a transformative justice approach to deepfakes and other AI harm risks are. If we pursue a legislative response to these threats, is there a risk that we concentrate control of AI in the hands of states and megacorps? The balance of power is already skewed towards those actors, so how can various other types of communities develop the resources they need to be able to exercise consent over the use of AI on them, and be able to prevent and heal from harms arising from violations of that consent?
It's not an unreasonable take. If the balance of power was tipped in favor of the government, and it mostly issued fines for bad behavior, why would the company developing AI/deepfake technology...
It's not an unreasonable take. If the balance of power was tipped in favor of the government, and it mostly issued fines for bad behavior, why would the company developing AI/deepfake technology care about complaints from the citizens? They could always revert to the corporate excuse: "we are sorry for the behavior and promise to improve the product". They would also not have an incentive to be transparent, so the entire thing would become the classic game of cat-and-mouse that we have seen so many times: bad actors do bad things in secret, eventually they do a very bad thing that goes public, new punitive legislation is implemented, rinse and repeat.
But a transformative justice approach would push them to go beyond such superficial excuses, and would imply that community-based systems for accountability (committees, organizations) would have an officially recognized power. This could be achieved by: government funding, laws that demand transparency from both government and companies, policy frameworks that encourage public participation in AI governance, and government funded education programs.
I don't quite see how this take on transformative justice approach differs from a legislative response. Or, what kind of powers would this grant to community based systems, and how does this...
I don't quite see how this take on transformative justice approach differs from a legislative response. Or, what kind of powers would this grant to community based systems, and how does this differ from a regulator?
And more generally, what does transformative justice mean in terms of preventing and healing from harms of the actual risks of deepfakes and other AI, such as election interference, phishing attacks, disinformation campaigns, revenge porn, copyright issues, and so on.
A good example would be the Truth and Reconciliation Commission which received its mandate from the president, and could take concrete actions (e.g. grant amnesty). I think that the main takeaway...
A good example would be the Truth and Reconciliation Commission which received its mandate from the president, and could take concrete actions (e.g. grant amnesty).
I think that the main takeaway is, if the focus is on implementing regulations and punitive measures, it will have two effects.
First, only entities as rich and powerful as the leading corporations will be able to participate in this domain, from all points of view: economically, and also as members of a dialogue that aims to change how these technologies work. But a transformative justice approach could democratize the conversation around AI, ensuring that not just the economically powerful have a say. This could enrich the debate with a wider array of experiences and concerns.
Second, by focusing only on the legislative response, we could miss the opportunity of endorsing this sort of civil organization with actual laws & frameworks, and giving them the voice to push for change.
Think of the US government body. Most politicians who have a say are old and out of touch with technology. It is easy for them to pass punitive laws & regulations, especially since their livelihood isn't directly affected by these technologies.
But if some hypothetical civil organization were officially endorsed to advice on the impacts, solutions and real-life harms of these technologies, we could have a more comprehensive picture and more democratic solutions to it. This organization could work on the ground to provide immediate support to those affected by AI misuse, which regulators might not be positioned to do. This work on the ground could be done in many capacities: prevent and heal from the harms caused by deepfakes and AI, like creating local support groups for victims, awareness campaigns, promoting digital literacy and legislative & regulatory advisory.
If legally endorsed, they could also bring fresh perspectives to the table, which might be more current and technologically informed than those of some legislators which is useful in the context of rapid innovation in this space, which outpaces the legislative understanding and response.
Don't get me wrong though, I don't think that the legislative approach and transformative justice are mutually exclusive. But there is a risk that, by focusing on the former, we'll never get the chance to work on the latter.
I’m on board with most of this but
This seems somewhat pointless and may even be harmful if it gives people a false sense of security. In fact, I think the primary beneficiary of this type of watermarking technology would actually be deepfake technology vendors who would be more easily able to filter deepfakes from its training data, helping to mitigate the human centipede issue.
100% agree. AI Watermarking reminds me of the Do Not Track HTTP header, which was well-intentioned but ultimately boiled down to trusting bad actors to do the right thing. And it had the unintended effect of adding one more bit of fingerprintable user data for those bad actors to capitalize on.
I don't know exactly how to solve this conundrum but it's trivial to remove every form of watermarking that I'm familiar with. I could imagine some kind of cryptographic signature scheme that proves the provenance and authenticity of an image... it wouldn't guarantee it wasn't AI-generated, but it could verify the creator of the file. Combined with some sort of reputation system we would at least have a better idea of whether a given image is credible. But most non-techie folks are pretty resistant to figuring out this stuff so I'm skeptical it could ever happen.
For all the drama over blockchain technologies, provenance is really the one thing it is good at (because at the end of the day it's just an authenticated ledger). If transactions on the chain weren't so expensive a distributed catalog of real and fake images could be compiled. Unfortunately I expect the overhead and necessary scale to be prohibitive.
And more importantly: It can't have any speculative value outside of being a ledger. Proof of Stake isn't that bad provided you're not competing with an infinite potential userbase expansion with "get rich scheme" as a background theme.
TBH a blockchain with the national governments as the only nodes, using it as a copyright ledger, will work much better than many other things. A closed loop with 200-500ish nodes maximum, clearly identified by owner, will be much more efficient than anything fully zero-trust.
What I fail to see is how this isn’t better done by, oh, say, a database. Like we currently do with copyrights, trademarks, patents etc.
I do generally agree. I think the main thing that it offers is the kind of multimaster replication between disjointed entities that don't explicitly have power over one another.
Do you trust a database run exclusively by the US government? What about copyrights in Germany? Whom "owns" an international copyright ledger?
The proper answer is the UN, and having it be hosted by individual members provides more auditability and less able to be manipulated by one dominant member that ignores UN rules whenever it suits them....like the USA.
This all ignores the underlying issue that copyright is determined by the courts. The blockchain doesn’t stop me (or the USA) from lying about copyrights, it simply allows us to record our statements. Dispute resolution can’t live entirely inside the blockchain, no matter how we structure the smart contracts, especially because those smart contracts will have to be agreed to by none other than the participants.
Unless I’m missing something, using the current system, most citizens are not willing – nor able – to quickly verify a given digital image’s authenticity; an image which may or may not be cropped, a (re-compressed) screenshot, etc., i.e. it is not a given that there even exists a pixel-perfect image match, even before we’re talking about AI-generated images.
Is that not reason enough for having a “verifiable origin” system for digital images?
Edit: reading further along in the thread, I tend to agree with what you said here – a “records of everything” does sound like unfeasibly high effort, when some form of creator + key = signature would probably work too.
Agreed, having the technology coopted by finance bros was probably the worst thing that could have happened to the tech. There is a lot if really interesting mathematics around proof of stake and zero knowledge proof systems.
As we saw in the video, proof of stake can be falsified with a tofu-based stake alternative smothered in vegan cheese sauce.
Except it’s not really all that good at provenance. If I want to claim that I made something, what I should do is give that object a signature with my key. Now if you have that object and signature you can verify it, without me first publishing a database of every thing in the universe that I ever made.
Kinda a baked in assumption that its not exactly backwards compatible with older works.
How do you mean?
Gotta say, if I was gonna list off which company I expected to make me aware of this act, I wouldn't have guessed Steak-umms in a billion years.
Your marketing department is genius Steak-umms.
I honestly only every knew about them due a throw away line in an early episode of Family Guy. I wasn't even sure if it was a brand or just some Rhode Island region way of cooking steak. So yea, very surprised by this one.
I wonder what a transformative justice approach to deepfakes and other AI harm risks are. If we pursue a legislative response to these threats, is there a risk that we concentrate control of AI in the hands of states and megacorps? The balance of power is already skewed towards those actors, so how can various other types of communities develop the resources they need to be able to exercise consent over the use of AI on them, and be able to prevent and heal from harms arising from violations of that consent?
What does this mean? I’ve read the wiki page on transformative justice, and I don’t see how this intersects at all with the problem.
It's not an unreasonable take. If the balance of power was tipped in favor of the government, and it mostly issued fines for bad behavior, why would the company developing AI/deepfake technology care about complaints from the citizens? They could always revert to the corporate excuse: "we are sorry for the behavior and promise to improve the product". They would also not have an incentive to be transparent, so the entire thing would become the classic game of cat-and-mouse that we have seen so many times: bad actors do bad things in secret, eventually they do a very bad thing that goes public, new punitive legislation is implemented, rinse and repeat.
But a transformative justice approach would push them to go beyond such superficial excuses, and would imply that community-based systems for accountability (committees, organizations) would have an officially recognized power. This could be achieved by: government funding, laws that demand transparency from both government and companies, policy frameworks that encourage public participation in AI governance, and government funded education programs.
I don't quite see how this take on transformative justice approach differs from a legislative response. Or, what kind of powers would this grant to community based systems, and how does this differ from a regulator?
And more generally, what does transformative justice mean in terms of preventing and healing from harms of the actual risks of deepfakes and other AI, such as election interference, phishing attacks, disinformation campaigns, revenge porn, copyright issues, and so on.
A good example would be the Truth and Reconciliation Commission which received its mandate from the president, and could take concrete actions (e.g. grant amnesty).
I think that the main takeaway is, if the focus is on implementing regulations and punitive measures, it will have two effects.
First, only entities as rich and powerful as the leading corporations will be able to participate in this domain, from all points of view: economically, and also as members of a dialogue that aims to change how these technologies work. But a transformative justice approach could democratize the conversation around AI, ensuring that not just the economically powerful have a say. This could enrich the debate with a wider array of experiences and concerns.
Second, by focusing only on the legislative response, we could miss the opportunity of endorsing this sort of civil organization with actual laws & frameworks, and giving them the voice to push for change.
Think of the US government body. Most politicians who have a say are old and out of touch with technology. It is easy for them to pass punitive laws & regulations, especially since their livelihood isn't directly affected by these technologies.
But if some hypothetical civil organization were officially endorsed to advice on the impacts, solutions and real-life harms of these technologies, we could have a more comprehensive picture and more democratic solutions to it. This organization could work on the ground to provide immediate support to those affected by AI misuse, which regulators might not be positioned to do. This work on the ground could be done in many capacities: prevent and heal from the harms caused by deepfakes and AI, like creating local support groups for victims, awareness campaigns, promoting digital literacy and legislative & regulatory advisory.
If legally endorsed, they could also bring fresh perspectives to the table, which might be more current and technologically informed than those of some legislators which is useful in the context of rapid innovation in this space, which outpaces the legislative understanding and response.
Don't get me wrong though, I don't think that the legislative approach and transformative justice are mutually exclusive. But there is a risk that, by focusing on the former, we'll never get the chance to work on the latter.
In related news, Meta has announced that they will be requiring labelling of political ads that use AI.
https://apnews.com/article/meta-facebook-instagram-political-ads-deepfakes-2024-c4aec653d5043a09b1c78b4fb5dcd79b
My concerns here regarding the implementation include: