In short
Eye On A.I. Podcast Episode #145: Summary and Key Takeaways
Episode Overview
- Host: Craig S. Smith
- Guest: Riley McCormack, CEO of Digimarc
- Focus: The role of digital watermarking in securing digital assets, especially in the context of the evolving AI landscape.
Key Themes
- Digital Watermarking: A technology that embeds machine-readable and covert identifiers within digital content to protect intellectual property and ensure authenticity.
- Impact of AI: As generative AI gains prominence, the importance of watermarking to safeguard content and uphold copyright law becomes crucial.
- Legislative Context: Discussion on the Digital Millennium Copyright Act (DMCA) and its implications for copyright management and protection.
Episode Structure
Introduction
- 00:00: Preview and introduction of sponsor Shopify.
- 03:05: Introduction to digital watermarking and its significance in the digital age.
Digital Watermarking Explained
- 07:01: Definition and evolution of digital watermarking.
- Covert and machine-readable identifiers embedded in digital assets.
- 14:21: Historical context and current applications of Digimarc's technology.
- Long-standing role in protecting various digital media and global currencies.
Protecting Digital Content
- 21:13: How digital watermarking enhances content protection.
- 28:56: Key characteristics of digital watermarking.
- Covert, immutable, ubiquitous, redundant, and secure identifiers.
Applications of Digital Watermarking
- 35:12: Practical implementation across various media types—images, audio, video, and documents.
- 42:46: Role in blockchain and NFTs.
- Addressing the authenticity and ownership of digital assets.
Challenges and Future Directions
- 49:11: Introduction of Digimarc Validate for copyright protection and authenticity.
- 01:03:01: Outro and final thoughts.
Key Takeaways
- Importance of Digital Watermarking
- Digital watermarking serves as a critical tool for protecting the rights of content creators in an AI-driven world. It provides a mechanism to ensure that artists and creators are compensated for their work.
- Legal Framework
- The DMCA 1202B emphasizes the legal implications of altering or removing copyright management information (CMI). Digital watermarking is recognized as a valid form of CMI, providing avenues for legal recourse against unauthorized use.
- Risks and Rewards of AI
- Generative AI poses risks to content creators, including unauthorized use of digital assets. However, it also presents opportunities for enhanced productivity and innovation in content creation.
- Future of Digital Content Security
- There is a growing need for an ecosystem that combines digital watermarking with blockchain technology, enhancing the traceability and authenticity of digital assets. Collaboration with organizations like C2PA is vital for establishing industry standards.
- Practical Applications
- Digimarc Validate helps users track their digital assets and identify unauthorized use, offering immediate copyright protection and a path toward authenticating digital content.
Conclusion Riley McCormack emphasizes the urgent need for digital watermarking to adapt to the fast-evolving digital landscape shaped by AI. The conversation highlights the delicate balance between innovation and protecting the rights of creators, as society navigates the challenges posed by new technologies.
For more information and to explore Digimarc's offerings, visit [Digimarc's website](https://www.digimarc.com).
Written by AI. May contain mistakes. Listen to the episode to check what was said.
Transcript
Automatic transcript. May contain errors.0:00A digital watermark is a machine readable and covert identifier that upon embedding becomes part of the item itself. So a key there is machine readable and covert as opposed to human visible and human readable. If I'm creating content and that content is allowed to be used without my consent or without my compensation, what is my motivation to continue to create content? I'm not an artist. Why should I care about copyright law? Because it's going to impact all of it. We are giving control to a handful of tech companies and state-sponsored bad actors unless we protect people's copyrights. That is a conclusion that is hard to escape.
0:34Hi, I'm Craig Smith, and this is Eye on AI. Welcome to today's podcast where we delve into the fascinating world of digital watermarking with Riley McCormack, CEO of Digimark. From protecting artists' creations on the internet 30 years ago to securing global currencies, as generative AI gains prominence, how does watermarking secure the digital future? Stay tuned to unravel these insights and more. Hi. Good tech solves problems you know about. Great tech solves problems you haven't even thought about. What can the commerce platform trusted by millions of merchants do for you? It's time for Shopify.
1:22The commerce platform revolutionizing millions of businesses worldwide. Whether you're a garage entrepreneur or IPO ready, Shopify is the only tool you need to start, run, and grow your business without the struggle. Shopify puts you in control of every sales channel. So whether you're selling satin sheets from Shopify's in-person point-of-sales system or offering organic olive oil on Shopify's all-in-one e-commerce platform, you're covered. Shopify powers 10 % of all e-commerce in the United States. And Shopify is truly a global force, powering Allbirds, Rothy's, and Brooklyn, and millions of other entrepreneurs of every size across over 170 countries.
2:14Plus, Shopify's award-winning help is there to support your success every step of the way. Sign up for a$1 per month trial period at shopify.com slash ionai. That's Shopify, S-H-O-P-I-F-Y dot com slash IonAI. That's E-Y-E-O-N-A-I all run together. Go to Shopify dot com slash IonAI to take your business to the next level today. Excuse me, sir. I couldn't help but overhear. Did you say Shopify? Shopify? Oh, Shopify.com slash I on AI. Carry on. Thanks for having me on. My name is Riley McCormack. As you mentioned, I am the president and CEO of Digimark. And my route to Digimark was maybe a little bit different.
3:14So I've been an investor in technology for the last 25 years. And if you put the last 25 years of technology up against the last 2 ,500 years of almost any other industry, obviously, it's the change, the revolution has been incredible. And so I've had operational roles to a certain degree, but nowhere near to the extent that I have here at Digimark. And I actually came across Digimark in 2014. As I mentioned, I've been in and around tech for 25 years, absolutely love it, live, eat, breathe it. And a friend of mine talked about this company and this technology I'd never heard about called Digital Watermarking and suggested I check it out.
3:52And that began the journey. So I started diving in. Digimark historically had been a licensing company. The big contract with the central banks, which I'm sure will come up later in this conversation, but was really a licensing company and was starting to productize this technology. And so it was really a small community, I guess you could say, in certain ways. It was almost like the homebrew club for computing. So I was going to GS1 conferences in the basement of hotels in Washington, D.C. and just became more and more fascinated. Truly believe this technology can change the world. And so I joined the board in 2020 and became CEO in 2021.
4:33Oh, that's interesting. I didn't realize. Yeah. So that all predates, well, I mean, the transformer algorithms. has been around since 2017, but really predates the gold rush in generative AI. So, yeah, I mean, what has captured a lot of people's imaginations in the last three or four months in particular is the idea of digitally watermarking text outputs from generative ai and i've i've had conversations with people about statistical methods uh and and i'm hoping maybe we can talk about that but but digimark started uh before that uh are are you uh does digimark do digital water marking for uh images only images and text uh yeah maybe you explain what your tech does?
5:36Yeah. So let's take a step back. We've been around for almost 30 years doing digital watermarking. So our founder actually was a astrophysicist as well as a photographer and wanted to find a way to protect his images as he was posting them on this new thing called the internet. And how could he claim those images? How could he attribute them back to him? How could he show his copyright? So Digimark's been around for almost 30 years. We are the pioneer and the widely recognized leader in digital watermarking. Our innovations and inventions in digital watermarking have been deployed at an unprecedented scale many, many times larger than any other digital watermarking company for a much longer duration for such use cases as protecting movies.
6:17So the Blu-ray anti-piracy standard and the digital cinema initiative to measure TV and radio audiences as part of both Nielsen's and Arbitron's audience measurement services. Digimark secured the credentials of the majority of the driver's licenses in the US, as well as secure credentials around the globe. And then as I mentioned, our work with the central banks. So for 24 years, we've been working with a consortium of the central banks using our digital watermarks to protect the world's currency. So yes, to answer your question, we predate Gen AI in terms of digital watermarking. Our digital watermarks can be applied to anything that is digital, anything that is made from something digital or is digitally processed.
7:01So basically the only thing we can't really do is things made by hand. So my kids water colors, actually we could, after they create them, we could digitally process them and watermark them or things made by mother nature. Yeah. And, and, uh, what I mentioned digitally watermarking text, is that something that you do as well? So our focus right now, uh, text is going to be a difficult challenge, uh, for, the whole world, right? Our focus right now is images, is audio, is video, and is documents. Because one of the things, and I would love at some point to get into how we decided, how we stopped looking at Gen.AI as a adopter of it, or not stop, because we are going down that path, but also realizing the incredible productivity that it's going to bring, but as well as the risk, right?
7:49Every revolutionary technology changes the status quo. And so by By changing the status quo, you unlock all this incredible positive, but you also open up potential negatives. And those negatives, those risks have to be addressed.
8:04So we have to, like I mentioned today, we have images and audio. We are working on getting out quickly video and documents because there is no time to waste. I think one of the things that is important for people to understand is that every time you release a digital asset into the world that's unprotected, that's it. It's gone. So take this podcast, right? If you were to release it without Digimark Validate, it's out there. It's going to be copied. It's going to be ingested by Gen AI engines. There will be multiple, multiple copies of it. Your ability to protect it and control it are gone. So we didn't want to wait for text before we came out with our product.
8:48We wanted to get out there what we could because urgency, we felt a real urgency that time matters. Yeah.
8:58So first, explain what digital watermarking means in the context of Digimark or what you've been doing. And then can you talk about how it's relevant to AI? Yeah, that'd be great. And I think it's important not just to talk about what digital watermarking is, but also what it's not, because I think there's a lot of confusion about what digital watermarking is. So let me start off by talking about what digital watermarking is. So very high level, digital watermarking is the science of hiding information about an item in the item itself. And then some terms that important to understand is the act of hiding that information is known as embedding the information and the act of discovering that hidden information is known as detection.
9:44Okay. So that's what digital watermarking is. A digital watermark is a machine readable and covert identifier that upon embedding becomes part of the item itself. So a key there is machine readable and covert as opposed to human visible and human readable. And because it becomes part of the item, the identifier is immutable once it's been embedded. There's some other benefits or some other attributes. Because it's covert, it can be ubiquitous, which means it covers the totality of the item. So if it's an audio clip, it would run through the whole audio. If it's an image, it covers the whole image.
10:20And because of this ubiquity, as well as the signal itself, it's redundant, which means it can survive a lot of damage and still deliver its message. And then finally, it's secure. So that's what a digital watermark is. If you don't mind, I'd love to spend a couple minutes on things that are not digital watermarks, but I think a lot of people confuse them as digital watermarks. Yeah, absolutely. So first of all, the idea of simply watermarking an item. Now, notice I didn't use the word digital. This is hundreds of years old, and this refers to the actual literal process of adding water to an item to leave a mark, a watermark.
10:58mark. So that's one thing where people get confused a little bit is in the physical world, if you have a really nice piece of stationery that has something you can read behind it, that's hundreds of years old technology. Now, the idea of marking an item to put in a human readable, a human visible code has been applied to the digital world. For example, the images, is the marks you will see in like stock photography vendors' photos, right? So this is simply what I said, though. This is a mark applied to a digital item. It is not a digital watermark. So let's go through our tests. These marks are not covert.
11:39They're meant to be read, right? That's why they're there. They're not immutable. In fact, when you pay a stock photography vendor for the right to use that photograph, they'll remove the watermark. They're not ubiquitous. they're not redundant, and they're not secure. And then recently and confusingly, some companies have started calling the metadata attached to a digital item a digital watermark. This just isn't accurate. And when it comes to proving authenticity, which is where digital watermark, one of the things that digital watermarks can do in the world of Gen.AI, it's actually quite dangerous.
12:14So the metadata attached to a digital item, it is covert, it is machine readable. So it passes those two tests. But really think about the metadata as almost a set of printed digital instructions that come along with an item in a box. So when you open medicine, you get your instructions. That's basically what metadata is. So it's not ubiquitous and it's not redundant. There's only one copy of these digital instructions and they can't survive any damage. Moreover, that metadata can and often is severed from the digital item. So when you post an image to X, for example, X splits out that metadata.
12:54They just don't want to store it. There's nothing nefarious there. So it's not immutable. It actually is often severed. But this is where it gets really dangerous. And all of those issues, the lack of immutability, redundancy, ubiquity, those are concerning. The worst one is it's not secure. You can actually change the metadata or a nefarious actor could change it. And one of the truisms, and again, we've been building massive multinational, multi-stakeholder, mission-critical systems of trust and authenticity for nearly 30 years. One of the truisms that runs through authenticity repeatedly is that false certainty is worse than uncertainty.
13:35And so the fact that using metadata as a way to give authenticity, but the metadata is easily changeable, it's not secure, that's worse. That could give false certainty. False certainty is worse than uncertainty. Yeah. Well, on the digital watermarking that you guys do, yeah, can you describe, presumably you introduce some sort of a pattern in the pixels or in the histogram or something that can be read after the fact. But yeah, can you talk about how you do this? Sure. Yeah, and it depends on the medium, right? So in audio, we're doing something in the audio channel. For images, you're right.
14:31It's a modulation of a pixel and chrominance and luminance on a pixel by pixel basis. um you know uh so what we're doing is we're we're we're putting an immutable covert machine readable redundant ubiquitous and secure identifier um and it you know we we like i mentioned we can work one of the one of the ways we're different from other digital watermarking and there's multiple uh but one of them is we can be applied as i mentioned to audio to images to video and documents. Yeah. And, and how is that applied? Is yeah, maybe you can take us through the, the tech. Magic of software. Yeah. So, so, so multiple ways, but it comes down to just a software processing of, of the, of the item, right?
15:21So audio, we can do live real time. We could be watermarking the audio in this conversation right now, just software. And it's not even that compute or processing intensive or memory intensive. Same with images. It's just a processing of a digital asset. Right. And it's not having to do with the codec or, you know, how the data is compressed or something. I mean, can you think of an analogy, yeah, maybe processing a photograph on a certain medium that's copyrighted so that that physical photograph cannot be separated from the medium on which it's printed? or is it is it yeah if you can try and explain to readers uh how it's done i'm specifically on images yeah we can start with images yeah sure so our software will ingest a image uh and then we will look at the way and i'm i'm probably going to get yelled at by my product and engineering people for for for um misspeaking or in in in attempts of simplification misspeaking a little bit.
16:53It's a little more complicated than this. But we will look at that image. Our software will look at this image. And obviously, there's no human interaction here. This is software to software. And then decide the best way to modulate the pixels, individual pixels, again, on both the chrominance and luminance, so both brightness as well as color, in the way that will make it the most imperceptible to the human eye, but still be as clear as black and white to a machine. Yeah. So pixel by pixel, meaning every pixel in an image is manipulated? No, it doesn't need to be. Again, this is really smart software that we've spent 30 years looking at the ways to make it the most robust, the least visually impactful.
17:43So without getting into the magic of all of our software, no, it doesn't need to be. But what we do say is that it is covert. It is machine-renewable. It is immutable. It is ubiquitous, and it's redundant. Yeah. So. Yeah. And how does this relate to registering digital assets on the blockchain? Well, so just, it doesn't need to have anything to do with distributed ledger, right? But it absolutely could, right? So again, what we are doing is we are putting an identifier into a digital asset via a digital watermark. So our digital watermark is an identifier of that digital asset. It can be read by machines.
18:26It is immutable. It is covert. Once you've identified anything, there are so many things you could do. So we have what we call our digital twins, which can convey authenticity, can convey provenance, can convey ownership. list is it's extensible it's an extensible digital twin so the list is quite literally endless we can also record all these interactions on distributed ledger and we actually have in other sides of other parts of our business do work with a couple of distributed ledger technology so think of it as less um there are different parts of the stack right what digital watermarking is is doing is identifying something then it is using that identifier to convey information two ways, right?
19:12To get information from the digital twin to deliver it to whatever device just identified that item using a digital watermark. It can also send back, hey, I was just detected here. That's an option, which I think is important in a lot of the use cases we can talk about. How those interactions are recorded, whether it's a non-blockchain database or whether it's a chiseled in stone on distributed ledger, that's just a matter of where people want to send the data yeah yeah uh i'm thinking this through as we're talking so the blockchain the uh the advantage is that it's an immutable ledger of uh that that demonstrates ownership right so there's other ways to do it so absolutely there's so so uh yes i'm sorry finish your The answer is yes, but the answer is it's not the only way to do it.
20:13Right, but just the NFT craze, for example, you come up with a digital asset, you register ownership or the purchase of it on the blockchain. and until there is another transaction involving that digital asset that's registered on the blockchain, that represents your title to the asset, right? Yes. That has nothing to do with watermarking. So you're taking, if we're referring to the NFT, that same digital asset, and you're processing so that it's watermarked. So that if anyone comes across or there's any use of that asset, you can show that it is the asset that you own. And then you can, if you have it registered on the blockchain, you can refer to that ledger.
21:24Yeah, it's actually an interesting example because stay tuned. Everything you're saying is right, except for the fact that while the blockchain records an immutable transaction or an ownership of that asset, how do you know that NFT that you're linking to is the actual NFT? it's a huge gap in the NFT business. It's one that we can address. It's one that we are addressing. I would say stay tuned, but it's exactly that. The blockchain is an immutable ledger of transactions, but how do you make sure the item in the digital world that you are claiming as a transaction or the transaction that you think is being recorded on the distributor is the right one?
22:13It's a very it's a wonderful insight, Craig, and stay tuned because there's a lot we can do with NFTs. Yeah, I mean, it's one of the reasons why I could never understand the NFT craze. I know. Well, I don't want to get into the NFT craze because there's so many divisive issues in this world and so I don't want to wade into that one. But I will say that the technology is not as secure as people think. It's not a problem with the DLT. It's a problem of how do you make sure that that NFT is the one that is being recorded? And again, we have a solve for that. Right. So I have an image or an audio clip and you watermark it for me.
23:02Then what can I do with that information or with that watermark? Yeah. So today, the immediate solve that we're offering through a product called Digimark Validate is protection of IP, protection of copyright, protection of brand equity. As I'm sure you're aware, and I'm sure your audience is aware intimately, the way these Gen.I. engines are trained is by ingesting an incredible and fathomable amount of data. Right. And there is a lot of lawsuits concern. And these are really, really valid. I do want to spend some time on the risks. What the Gen AI companies are claiming is it's fair use under copyright law.
23:49Right. That is or out. Well, in 1997, there was an update to the copyright law called DMCA, the Digital Millennium Copyright Act. And it was a recognition of Congress that the copyright laws on the books left some gaps when it came to the Internet. right? This brand new thing called the internet, how do we update them? One of the updates to the copyright law through DMCA 1202B is the section, is the idea of CMI, of copyright management information. And it makes it illegal to alter or to remove CMI. This is separate. This is outside of whether there is an infringement. This is if there is CMI in a digital asset and you alter it or you remove it, you're in violation of DMCA 1202B.
24:40And that's the US law. There is similar law in multiple other countries. So as this debate, it is going to be a costly war. This billions and billion dollars of spend, it's gonna probably reach the Supreme Court 20 times because on the fair use in terms of, well, is it a fair use in this case? Is it a fair use in this case? that's irrelevant because if you apply Digimark Validate to your digital assets today, you can avail yourself of the protection of 1202B of the DMCA and similar laws in other countries, which have nothing to do with infringement. It's a separate redress from infringement. And there's actually case law that says digital watermarking is a valid, recognized conveyor of CMI.
25:27So that's where we're starting today with this. There's also, again, think about what we're doing. So we're putting an identifier. And in that case, that identifier is conveying management, copyright management information. It can also be used for authenticity. So we're a member of C2PA. I think it's a great organization. And because you can say, okay, okay, this is, yes, Craig did put this asset out. This is Craig's asset. Going back to our conversation, for example, on metadata, without a digital watermark, you could put something out. I could go into the metadata, literally just delete, delete, delete, put in Riley.
26:07Now all of a sudden, hey, this is my asset, right? Or I could make something up and say, yeah, Craig put this out. So that's authenticity is the next step we're going to take on that. And then beyond that, it's really unlimited. We're in conversations with multiple partners, big tech companies and the like, because once a digital asset is identified, there are so many other things you can do with it. So it's immediate copyright protection today. It's authenticity. And then let's see where our partners take this technology. We're really excited to begin that journey. Yeah. Well, for example, on the generative AI training sets, if you had watermarked your audio file or your images and you wanted to keep them out of or search to see whether they are in the training set or exclude them from the training set, training set, does the watermark help in some way?
27:21I mean, yeah, go ahead. So first of all, when you say if, we are. We are a Digimark Validate customer. Part of the reason we raced to get this product out is because we wanted it for our own use to protect our digital assets. So no, if we are a customer. But absolutely. So one of the things, again, regardless of whether or not the Gen.AI engines take our SAFE, that is the name for our digital watermarking technology, our SAFE watermark detection software, they're protected under DMCA 1202B and similar other laws in other countries. But there's so many benefits to the Gen.AI companies for taking our detection software, which we offer to them for free.
28:04because not only going through the multiple benefits, first of all, if they're using digital assets that have CMI, including digital watermarking, including Digimark validate digital watermarks, they're in violation of the law. So that's reason enough that they should be doing it. Taken to the extreme, there's absolutely on the fair use side, we talked about those lawsuits earlier. All of these Gen AI engines are going to be spending a lot of money on lawsuits. And the outcome of this could be untold trillions, probably an aggregate of damages, plus model retraining, if they're found guilty of all these things.
28:39So there's a lot of reasons to do it around complying with the existing law, DMCA 1202B, not playing a game of chicken with model retraining under fair use, which is an outcome for these companies. but there's other benefits so one of the things is as more and more content that they're scraping is itself gen ai uh created without any human interaction and it starts to get a little bit silly um they face model collapse you know the example i give is you know are you a family guy fan by any chance yeah yeah yeah so you know how stewie's head is shaped like a football yeah right well if over time with iterations if a gen ai starts to make the human head more elongated in, I was going to say in three years, but the way Gen.ai is moving so quickly, it could be in three seconds.
29:27Gen.ai is going to think humans all look like Stewie, right? So they face model collapse unless they're actually having some way to automated and efficiently read a tag on a digital asset they're adjusting. And then there's a lot of other benefits. Again, the identifier that we're putting into a digital asset for the content creators benefit can also be used as a way, a tag for training Gen.ai, right? So maybe a Gen.ai company wants to train on a, I don't know, they have permission to train on a company's advertising collateral, holiday collateral in a Spanish speaking world. How are they really going to understand how to do that?
30:10Unless there is a digital identifier, they can say, okay, find me all of these images. So there's benefits to both the content creators, but also to the Gen.ai. And that's key to what we're doing. We're doing this. We believe regulation is coming. We believe there is some really powerful regulation that will be helpful. But along the way, the more we can do to make this a win-win-win for everybody, the faster it happens and the better off everybody is. Yeah. Just going back, you said that if a training data set contains a digital watermark uh watermarked uh asset uh it's in violation explain that again yeah so what i'd really suggest is is that um there's a lot on our website digimark.com so d-i-g-i-m-a-r-c uh dot com on the dmca 1202b it is a like all laws i mean there's there's a lot to it um and we're not um lawyers i i'm gonna make my chief legal officer happy by putting that disclaimer or verbally in here.
Read the full transcript
31:16It's on our website too. Although we have a lot of lawyers in-house and we have some excellent legal scholars that we've consulted with. But for those really interested in understanding the ramifications of DMCA 1202B, we encourage people to go to our site because there's a lot of great information there. Yeah. Yeah. Okay. So can you give us some of the the big use cases that you guys have handled in the digital domain or digital across all of our business yeah so we launched yeah so we launched digimark validate uh a week ago we could go today um and again are the initial uh value profit that our customers are getting immediately is is copyright protection by availing themselves of the protection available under uh dmca 1202b and similar laws.
32:12There is some authenticity benefits. So we also offer a web crawler. So one can go out, a customer can go out, use our web crawler and find where their digital assets are. We have extensions, web extensions, browser extensions, so they can see whether it's theirs. But again, looking for that identifier. The authenticity, the widespread authenticity conveyance is going to come with a partnership. So I've mentioned C2PA. We're talking to other, we call them nodes, as we're thinking about where the detection sits out in the internet. Other big node providers, whether they're social media companies, whether they're e-com providers, whether they're device manufacturers.
32:59The way we look at this is there is a real crisis on this copyright. And it's not just the content creators themselves. I mean, taken to the extreme, and I'm going to say this, but I need to preface this. We are big believers in the power of Gen AI. We think this is going to be the most important productivity tool since the internet. So we are fans of Gen AI, but it's just important. Whenever there is a change in technology, you can't just look at the benefits. You got to look at the risks. And the risks of not respecting copyrights. It goes beyond just the copyright owners themselves. It impacts all of us.
33:39So we're seeing this with the strikes today and with some of these lawsuits where if I'm a content creator, and this could be musician, this could be a movie producer, this could be a podcast host, um you know it could be a company's creative um whatever if i'm creating content and that content is allowed to be used without my consent or without my compensation what is my motivation to continue to create content um and you know what is the single probably if you were to put a a thematic bar chart of all the sci-fi dystopian horror movies, it's machines taking over. And that's exactly the path we are on when it comes to content creation.
34:37If human content creators aren't incentivized to create content, but machines can do it untethered with any human contact just based on the exponentially increasing processing power and algorithmic power that's being thrown here, we're going to cede content creation to a handful of big tech companies and some state-sponsored bad actors. That's the path we're on. Now, again, I didn't mean to get dark there. I am a wildly optimistic human being. We believe in the power of Gen AI. We think this is going to be a wonderful productivity tool. We just got to address the risk. And so this this idea of, well, I'm not a, I'm not a artist.
35:18Why should I care about copyright law? Cause it's going to impact all of it. We are giving control to a handful of tech companies and state sponsored bad actors, unless we protect people's copyrights. That, that is the, that is a conclusion that it's hard to escape. Fortunately, there is a solution and, and, and, you know, going back to my natural sunny positive side, because it, you know, we, I get excited when I talk about this stuff. I get excited when I talk about tech. What's really cool about this is the reason we're having this conversation today is because of the internet. Why did the internet came around?
35:50How was it invented, right? The 1960s, during the Cold War, we needed a bomb-proof way to communicate. That was the crisis. How can we, with this specter of nuclear global war, how can we make sure we have communications up? the outcome of solving that crisis led to the 973 billion ways we use the internet today that we couldn't have fathomed back then. Same thing with Digimark Validate. The crisis of solving for copyright protection, and it is a crisis, is going to lead to something that the internet's needed for a while, and this is the catalyst, which is an ecosystem of trust, authenticity, and security.
36:30Because again, by solving for this copyright issue, now we can start to address the issues of authenticity that also need to be addressed. Yeah. Well, that was going to be my next question. But before that, if you have this watermarked asset and it's in a Gen AI training data set. Does any of the water markings survive the ingestion by the large model? And the reason I ask is, and I'm sure you're very familiar with this, but if you use a lot of the image generation models, mid-journey or stable diffusion or these, oftentimes uh the output has uh the artifact of an artist signature in the corner uh i mean it's no longer rendered in as a recognizable signature but it was in the training data so uh does any of the water markings survive uh in in uh in a large model Yeah.
37:49So the short answer is every engine is different. Every engine is constantly changing. It's impossible to give a universal truth to that, but I would go back to this, which is if they're altering, which I go back to digital watermark is a form of CMI, of copyright management information. If it's altered or removed, it's violation. So it's not the fact that a digitally watermarked asset is in the training data that's a breach of law. It's the generation. Alteration or removal. It's the alteration or removal of CMI. Copyright management information of which digital watermarking is a legally recognized case, you know, precedent example of CMI.
38:40Right. But I would guess then that deciding whether or not a generative AI image is altering or removing is a question for the courts, right? Altering or removing the CMI? Well, let's say I have a song that I've digitally watermarked or a painting that I've digitally watermarked. And to make it more direct, a piece of digital art that I created that I've digitally watermarked. It's included in the training data for a big generative AI image generator. And then that image generator generates an image that has been influenced or includes elements of my original asset. Does that constitute a removal or mutation of the CMI?
39:53Yeah. So, again, I think maybe you're – this is a new avenue, so I get the confusion. I think what you're thinking about is like, I'm forgetting, blanking on the term now, but transformative use of, does it change? It doesn't matter what the output is. Digital watermark is CMI. Digimark validate gives you CMI, copyright management information. If that copyright management information is altered or removed, then it's a violation of DMCA 1202B and similar law. So it doesn't matter what the output of the engine is. If that watermark, and again, without, it's impossible to give a universal answer because there are a ton of engines and the algorithms change almost every day.
40:42So, but, you know, it would be, I will give you a very high level. It is very hard to imagine an ingestion of a bunch of content with the purpose of creating other content where our digital watermarks would survive that whole process in totality. all they need to be is altered and they're definitely going to be altered right so um that doesn't mean they can't read i mean this is one of the wonderful things about our digital watermarks if you go back to the benefits right because of both the ubiquity meaning that they cover the totality of a of a digital asset uh but also the redundancy brought by that ubiquity as well as the code itself it can survive a lot of damage and still read but if that that damage if it's done um uh by a machine is an alteration of the watermark it's still damaged so it's it's um it it almost doesn't matter if it comes out at the end and reads or not it's the alteration or removal of it okay uh then on authentication uh which as you said is becoming a huge issue.
41:48First of all, does this apply to video as well? Can you digitally watermark video? We can. We haven't commercialized it yet to the point where we can have our customers do it. So stay tuned on that. So again, what we have today is image protection, audio protection, coming soon, video protection, and document protection. Right. So in an ideal world from your point of view, certainly every piece of digital content produced would be processed with a digital watermark. So if I, you know, gave an interview, or let's say that President Biden gave an interview and that audio was digitally watermarked, then you could say that every place you find that audio, that it's authentic.
42:46And if that audio is altered, you'd be able to see that it's been altered. Is that right? Yeah. Yeah. You know, we talked about the, again, the two risks we see from Gen AI. And I really want to stress this because we are big fans of it. So there's a lot of benefits. You know, that's actually our journey. Quick, humor me with 30 second journey of how we got here. Sure. Which was about a year ago, like most companies, I hope. If not, I hope they're doing it now. Most companies, our R &D team had been aware of Gen AI and looking at it and doing some stuff with it. But as a business, we sat down about a year ago when it really became clear it was a technology that could transform businesses.
43:31And so we took a look, what benefits could we get as a company from adoption? And what were the risks that would come from our adopting? And it's interesting, right? We found numerous and various ways we would benefit, but the risks really boil down to two. And we've touched on this, but I think it's important to go back to it to get to where to answer your question. The two risks all boil down to is how do we protect our brand equity, our IP, our copyright in a world where our digital assets are being consumed without our consent to create new content without our knowledge? So that was one bucket of risk.
44:11And the other one is how do we prove our authenticity? How do we let people know this is a Digimark asset as opposed to somebody pretending to be Digimark, right? We immediately realized the next step, which was, oh my gosh, this wasn't a risk of adoption. This was a risk of existence. Even people who weren't going to be aware of Gen AI, as quaint as that sounds now, we were actually having this conversation. what if people aren't aware of Gen.ai? What is their risk? I think now everybody is aware of Gen.ai. But it was going to affect all content creators, whether or not they opted in for the Gen.ai revolution.
44:49And because it was going to impact all content creators, whether or not they opted in for the Gen.ai revolution, it was going to impact all of us. And that's when we got to work because we knew that based on our history of being the pioneer and widely recognized leader in digital watermarking, our technology had a role to play. What's been wonderful since then is to see that become a universal acknowledgement. Case law, governments, big tech all realized that digital watermarking has a role to play. So that has been great. The second realization, the reason we got to work with urgency, as I was talking about earlier, was there isn't a moment to waste.
45:29Every time a new digital asset is released, you know, distributed or published without protection, that digital asset's gone forever. So we really wanted, I said, we raced to get it done, A, because we wanted to build this for ourselves, for our own digital assets. And we are a happy Digimark Validate customer ourselves. But also because we wanted to offer this to other people because of the risks. we've talked about the risk of copyright and how it doesn't just impact content producers it's going to impact all of us in this potentially dystopian computers take over the creation of content uh defined history uh way um but the other one was authenticity which i think is equally as scary we are spending more and more time in a digital world i mean i don't think um you probably don't need to tell a podcaster this that uh how much time people are spending in the digital world listening, consuming digital media.
46:25If there isn't a way to convey authenticity in this world, we're going to head down to one or two paths. Either people are going to believe everything they see, which is going to cause companies, news organizations, governments, individuals, societies, meaningful and irreparable harm because content creation is now fast, cheap, and easy. And anybody can do it and people will do it and they'll continue to do it for nefarious reasons. So that's one path. We're all going to believe everything we see and it's going to be really divisive. The other one, I think just as scary, people are just going to stop believing anything they see.
47:02People aren't going to tune into your podcast or be like, I don't know if this was Craig and Riley or it could have just been two machines talking. I don't care. And the Internet and all digital media, it's not just entertainment. It's being used to convey really important and legitimate information. Now, I think, I don't know which one of those is worse. Those are the two paths we're going down. To be clear, Jen and I didn't create this problem. Three years ago, we would have been talking about deep fakes. Right. So so this is a problem. This is a path we're going down anyway. What's really cool about this is that, again, for solving for the copyright protection issue, we can create we can solve for the authenticity.
47:41The solving of that crisis is going to be one step back with the crisis, one step forward solving the crisis, 500 steps forward by delivering on the promise of the internet, which is a safe, authentic, and trusted place. Digimark Validate is going to make it as difficult to counterfeit people, companies, and news as we've made it to counterfeit the world's currency. That's exciting. Yeah, absolutely. Another question. And so you watermark your assets, your digital assets, and then you were saying that there are ways to scan the Internet for instances of that asset. And if you find an unapproved use of that asset, then you get your lawyers involved, right?
48:34and that there's an infringement case or something like that. Is there a way or is there technology or are you considering this that nobody can copy a digitally watermarked asset or create a new instance of it without some permission on the owner's part. And if they try to, it won't work. It's a little bit like those PDFs that you get that are set so you can't block and copy the text. Yeah, so that would involve more than just a change of Internet protocol to fully containerize it. However, we can provide a lot of protection along the way. You know, Digimark Validate is multi-layer. What we mean by that is it's not just about digital watermarks.
49:35That is what people are most aware of. That is the conversation people are having because, again, governments and big tech realize digital watermarks have a role to play. One of the advantages we have of having done this for 30 years, we can see the next step a little bit. You know, we knew digital watermarks were going to have a role to play. the world is caught up to that truism. We understand the urgency to get this done now, because once a digital asset is released out into the wild for the first time unprotected, it's gone forever. We're starting to see that realization dawn on people. Maybe they're a little bit still behind on that.
50:12But the other one is that this has to be done through a multi-layer system. And that's what we have because digital watermarks are an important component of it, But it's the same reason, you know, you don't have one lock on your door. You probably have floodlights in front of the door. You got multiple locks on the door. And maybe you have a dog who barks if the door is open and you're in bed. Right. So it's the same thing here. It's going to be a multi-layer solve. Digimark Validate is multi-layer. If you go to our website, I don't want to sound too promotional, but because it's not a net, because it's Digimark with a C, I feel like I need to, if people are interested didn't go into the website, I got to spell it out, which is D-I-G-I-M-A-R-C.com.
50:54You'll see that we're launching early 2024 registry and they can click on there to be updated that registry launch. And that's going to be a big part of what you're talking about, Craig. Yeah. That's a distributed ledger registry? Doesn't have to be distributed ledger. We are interfaced with distributed ledgers. It doesn't have to be there's other ways to get security and um comfort besides the lt yeah uh and and so this idea of a mechanism that would prevent uh the reproduction of a digitally watermarked asset without some key or some permission well yeah so you would see without getting too far into the future and stuff we're not quite ready to talk about yet because we don't want to tip our hand too much.
51:45But there's a couple of layers, not to use layer in a different way. But for example, if there is CMI in an image and you alter it or remove it, it's illegal. It's not just for the Gen AI engines, that's for anybody. So that was not a law that was written in 1997, hey, Gen AI engines that won't be invented for 20 years that don't do this, right? Secondly, it will be very clear with Digimark Validate that an image has been altered or changed. Even, you know, let's say there's a picture of you and I, and I put a, I don't know, a knife in your hand. That would be, that would show up as an alteration.
52:31So one is there is protection under DMCA from people changing it. Two, we can make, we can show that that image has been altered, that it has been changed in multiple different ways. And so going back to the point of, you know, whether we, whether Digimark by itself can make an, any digital asset on multiple, that's a internet foundational technology change that we can do. What we can do is make it so that people know it's changed. They can see that this image is not what was intended by when Craig put this out, there was not a knife in his hand. It has been changed. And then it becomes, this is inauthentic.
53:14And so we can provide a layer for consumers to decide for themselves what's authentic. Well, here's a picture of Craig with a knife in his hand. And it's not from Craig. I can see that it's been changed from what Craig originally posted. It's probably not authentic. Right. But in order to, other than having seen the original image and seeing that it's been altered, you need this detection system, Digimart's detection system. Is that right? To see whether or not something's been altered? Yeah, and we're working with C2PA to get, there needs to be standards. This is exactly what C2PA, this is a wonderful organization.
53:58They came together for this reason. We're not, you know, we're having conversations with other, like I mentioned, important nodes. You are absolutely right in that there needs to be, going back to the definition of digital watermarking, it's hiding information and detecting it, discovering it. So there has to be that detection event. So this is where we started with the copyright management with Digimarkality. That's available immediate. There isn't a required network effect. It's going to be a journey of interested parties. And what's wonderful is the reception, the urgency that people are getting to have this.
54:31So that, yes, I'm not entirely sure if your question meant I would need to have seen that photo the first time. No, what it should be is you go on, pick your social media site or web browsers, right? W3C has a big role to play in this too, which is, I know this image has been altered. I know this wasn't what's intended. This is not an authentic image. Yeah. And that detection software would presumably eventually be embedded if it becomes the standard, if Digimark were the standard, would become embedded in search engines and maybe your operating system or I don't know. but so that you would immediately see when, as you scan digital assets.
55:18Yeah. I think, I think C2PA is going to set the standard, right? We're not, we're not going to set the standard. It's just not realistic, nor is it what we want. We want an ecosystem, an open, safe, authentic ecosystem. And we're, there's a real urgency to get there. We really, these risks are real. And so we're not, we just want to get there as a society because we're benefits you know my i have two kids they're on the digital world all the time i have to remind them that the www does not employ a fact checker um and and and uh they get sick of me saying that um but anyway the uh um so we're racing to get there with the group of with the uh you know we have multiple conversations going on with governments to provide uh our input to the uh regulation that's coming um in multiple geographies um we're working with c2pa um but yes there does need to be detection ubiquity um in order to get broad authenticity conveyance yeah and uh can you give me uh we're coming up to an hour uh a few of the big use cases that that you've worked on uh specifically with digital watermarking?
56:33Sure. So, you know, our work with the World Central Banks, unfortunately, I can't say that much about it because we're at Secret Service. We probably joined this call before it were over if I tried. But we work with the World Central Banks to protect the authenticity. This is a 24-year relationship. We just actually signed an extension of that through the end of this decade. We're extremely proud of this work. You know, we talk about our mission as a company is to promote a prosperous, safer, and more sustainable planet. When you talk about safe, trust in global currencies is right up there, right?
57:11I mean, we're talking about spending a lot of time living in a digital world. That is a massive area that we need to help add authenticity. But currency is foundational. If we lost faith in our currency as a society, all bets are off, right? So we're really, really proud of that work. We have another product, Digimark Recycle, where we can transform the sortation of plastics. And really excited to make great progress there in multiple geographies as well. Fulfilling our more sustainable part of our mission. So this is, as I mentioned, you know, there was a business. We've since sold this business.
57:51But Digimark at one point secured more than half of the driver's licenses in the U.S. and multiple secure credentials overseas. So this is what we've done. For almost 30 years, we have built massive, multi-stakeholder, multinational, mission-critical systems of trust and authenticity. So this is just applying it in the digital domain, and we're really excited to be on this journey. Yeah, but in the digital domain, is there a big use case that you can point to? I mean, currency, presumably you're looking at watermarking physical currency. And yeah, are there some use cases in the digital domain?
58:32Yeah, so we've been part, our innovations and inventions have been part of the, as I mentioned earlier, the Blu-ray anti-piracy and the digital cinema initiative by Arbitron and Nielsen for their people measurement systems. And we do have customers that are using our technology in the image space today to track some of their assets. And as I mentioned, this is all things old or new again. This is how the company was started. A brilliant and wonderful human being named Jeff Rhodes started Digimark almost 30 years ago to protect his images. because he was a photographer as well as an astrophysicist and looking out and saying, I really want to share my images on this brand new thing called the WWW and how do I do it and keep attribution, right?
59:28So this is back to our roots. This is back to our origin story. Yeah. Pretty cool. Yeah, well, that's fascinating. And just going back on the standard, The standard, you're sort of standard agnostic. Is that right? I mean, you have the process, and I don't know what the standard would be, but if there was a certain standard for watermarking and reading or detection of watermark, your tech would adapt to that? or are you a proprietary standard that the world has to adopt to in order for your products to be useful? We're not going to change the world by ourselves. We're not intending to change the world by ourselves, even if we wanted to.
1:00:19We don't think the world has enough time. I don't know if... It's funny. I was thinking about the DMCA and some of the regulation outside of DMCA. DMCA is great, but if you think about when DMCA was written, And I don't know this to be true, and I should probably look into it before I say it publicly on a podcast. But there is an argument to be made that some of these Gen AI engines have as much processing power as the whole Internet did back then, right? I mean, it's possible. It's at least an order of magnitude probably right, right? And so I don't think people understand how quickly this is changing.
1:00:52I think this is your tagline, right? Your world's changing. AI's going to change the world. So you get it. So you get it. But so we don't want to do this alone. And we don't think the world has time to wait because it's unfathomably large how quickly things are changing and how permanently things are changing every day. So, like I said, we're in discussions with government officials. We are proud partners of standard-setting organizations such as C2PA because this is going to be an ecosystem. And what we want to be is we want to be a tool provider in that. We want to educate people on the power of digital watermarks because, again, what's been wonderful is seeing people really coalesce around the view that, yeah, digital watermarks have a real fundamental foundational role to play in addressing the risks of Gen AI.
1:01:36That part of the debate's over, which sometimes takes a long time in these ecosystem and these standard setting things. People get it, so we're past that. Now let's figure out how we can all work together. And again, we're a mission-driven company, and our mission is to make the world a prosperous, safer, and more sustainable planet. And this is a wonderful way to help. Yeah. Okay. I keep on thinking of new questions. So, so I'll try and make this the last one, but this applies to PDF documents as well. Yeah. Yes. So we will, again, not, not yet the technology there, we haven't commercialized it yet.
1:02:11And we know, but yes, it'll be documents, PDF documents as well. That's one of the subset of, of areas of, I don't want to get too far into that because we haven't released it yet, but that will be part of it. Yeah. Okay, Riley. Well, that's all fascinating. And just finally, how do you price this and is it affordable? Yeah, well, I love that question because it lets me end with what I wanted to end with, visit us at digimark.com, www.digimark.com, and you can see our pricing on our website. There's many packages available just for web sales to get going. And then we do have enterprise offerings that are beyond that.
1:02:54But there's great starter packs for any of your listeners who are interested in protecting their digital assets today. And before you go, give Shopify a try. It's only a dollar a month for their trial period. And you can sign up at shopify.com slash I on AI, E-Y-E-O-N-A-I, all run together. all lowercase. Go to shopify.com slash IonAI to take your business to the next level today. That's it for this episode. I want to thank Riley for his time. If you want to read a transcript of the conversation today, you can find one as always on our website, IonAI, that's E-Y-E hyphen on.ai. And remember, the singularity may not be near, but AI is already changing your world.
1:03:49So pay attention.
From the publisher
This episode is sponsored by Shopify. Shopify is a commerce platform that allows anyone to set up an online store and sell their products. Whether you're selling online, on social media, or in person, Shopify has you covered on every base. With Shopify you can sell physical and digital products. You can sell services, memberships, ticketed events, rentals and even classes and lessons.
Sign up for a $1 per month trial period at http://shopify.com/eyeonai
On episode #145 of Eye on AI, Craig Smith sits down with Riley McCormack, President and CEO of Digimarc, pioneers in digital watermarking and cloud-based product data.
In this episode, we explore the critical role of digital watermarking in securing our digital assets, especially amidst the surging influence of AI. Riley guides us through the potential risks and benefits this technology brings to the forefront as AI continues to transform our digital world.
We then navigate the intricate territories of NFTs and distributed ledger technology, understanding how digital watermarking is reshaping these fields by ensuring trust and authenticity. Our discussion also delves into the Digital Millennium Copyright Act of 1997, highlighting its relevance in upholding copyrights and fostering trust within the digital stratosphere.
We conclude with a look at how digital watermarking impacts content creation and its role in shaping a secure and sustainable digital future, from collaborations with central banks to innovative products like Digimarc Recycle.
Craig Smith Twitter: https://twitter.com/craigss
Eye on A.I. Twitter: https://twitter.com/EyeOn_AI
00:00 Preview, Introduction and Shopify
03:05 Introduction to Digital Watermarking
07:01 Evolution of Digital Watermarking
14:21 Digimarc's Role in Digital Watermarking
21:13 Exploring the Protection Of Digital Content
28:56 Key Characteristics of Digital Watermarking
35:12 Application and Implementation of Digital Watermarking
42:46 Watermarking in Blockchain
49:11 What is Digimarc Validate?
01:03:01 Outro and Shopify




