Can we protect art online? Moving from the physical realm to the digital world.

This piece serves as an introduction to explorations into applying techniques to prevent digital art theft on online art-sharing platforms.

The initial problem is quite simple to phrase and explain: “Artists create and post art online, and given the ‘digital’ nature of digital art’s shareability, it is difficult (if not possible) to control the distribution of the work.”

The problem takes many different forms for artists, ranging from simple offenses such as reposts, to more extreme offenses as tracing, unauthorized use in advertisement or sale of products, etc. All of which deserve to be addressed. Perhaps it is the most heinous offenses such as unauthorized use of artworks in sale of products or advertisement, and tracing that concerns the reader, as these can be most related to physical cases of theft. However, it is important to note that artists are materially affected by traffic being diverted away from their works by reposts. As such, each of these offenses are still offenses, and are worthy of address.

Currently artists, in terms of preventative action, can only apply a physical watermark to their work in the form of a signature or an overlay (or a mixture of both). This is a preventative action that originates from our physical realm of art. Lots of artists sign their works, and as a testament to how important this practice is, even forgers have been known to hide their signatures in their forgeries. However effective (or ineffective) this method is in the physical realm, once applied to the digital world, seems to lose its potency. In the least offensive cases of reposts, signatures and watermarks are simply ignored, and traffic is still diverted (no control over distribution). In the most heinous cases of theft, signatures and watermarks are simply removed. Readers must understand that it is not reasonable to put the burden on the artist to reduce the visual aspect of their work in order to ensure their signature is tied to their creation.

It is here that I must mention the empty and irresponsible (and arguably fraudulent) claims made by pushers of Non-Fungible Tokens (known as NFTs) in their ability to protect artists, which in reality behave like nothing more than digital receipts, and which fail to solve any of our mentioned problems as well as contribute negatively to the environment. Allow me to stop the discussion here, and say that we will not be talking more about this.

So what is the solution to our problem? And perhaps, it might be of interest to the reader to know as well what I consider the core of the problem to begin with, as knowing the perspective in which I think about the problem may in fact give more comfort to my claims. When we talk about control over the distribution of digital artworks, can this really be done? After all, isn’t one of the fundamental properties of digital works the ability for it to be widely distributed, and having indistinguishability between the “original” and “viewed” copies? Is it possible to have a best of both worlds where our digital works can be shared widely, but only in a legitimate manner, or better yet only in the manner in which the artist wishes (no matter how strict)?

The core of the problem is one of authentication. Since our problem is really that we wish to allow the artist to be able to have the rights to control their digital works, we have to be able to authenticate that a specific piece of work is created by a specific artist (or more broadly put, a specific piece of work is owned by a specific individual). This is in fact how it works with art in the physical realm as well. For instance, there are systems in place to authenticate and document the ownership of the Mona Lisa, which belongs to the French Republic (and is on display in the Louvre). If this documentation and authentication system did not exist, or more interestingly was not recognized, then control over this piece of artwork would collapse (even in the physical realm). For instance, if war breaks out, and the ownership/authentication of this piece of art was not respected then it would not matter. Understanding that at the very base level, there are necessary conditions that must be met for our system to serve its purpose. This shows that there are required conditions, but also presents the warrants for why authentication cannot be decentralized. It would not make sense in the physical realm to allow just any person (or entity) to authenticate ownership documents, because this would break the condition of recognition we previously mentioned. Readers have to understand that these problems also present themselves in the physical realm, and that in the digital realm there are even more challenges as copies of digital artworks should be made indistinguishable.

All of this to say, my readers, is that if we understand the basic conditions outlined for authentication to function, then the platform in which digital artwork is shared must hold the responsibility to authenticate as well as adjudicate (or assist in allowing adjudication) and simultaneously be held responsible for upholding these tasks. However, it must be said in order to be fair, that the question of ownership over a piece of artwork in the physical can be just as up in the air and just as difficult to verify as the authenticity of the work itself. And in the digital world, it is not easier. Readers can imagine in the digital world where anyone can obtain an indistinguishable version of the artwork in question, that verification would be difficult.

The digital artwork sharing platforms suffer from this unique brand of authentication issues: enrollment becomes a race-to-the-bottom. Here, enrollment means the act of authenticating a piece of digital artwork for the first time in the system (platform). A race-to-the-bottom in our context means that enrollment in the digital world of artwork becomes a burden of artists who wish to have their work protect must become the first person to claim ownership in that platform. Artists cannot be expected to retroactively protect their works by being the first person to enroll, and even in the extreme cases where artists choose to forgo all previous works (unrealistic), a first-come-first-claim system will still fail to guarantee protection should the platform fall out of favor. Is this then, an impossible task?

Let’s for the moment, ignore the race-to-the-bottom. Readers, allow me to discuss briefly a system that can be created to allow artists to enroll their works and receive some guarantees. Imagine that when an artist enrolls their work by sharing on the platform, that their work is automatically watermarked (digitally) and digitized in a way that any appearance of this work on the platform (whether it be a saved copy directly from the timeline, a screenshot of the page, a picture of the display from a smartphone camera, etc.) can be detected by the platform and properly credited automatically (or removed). In this environment, artists would no longer have to worry about losing control over the distribution of their works. Allow as much or as little sharing as desired. If readers wish to read more about the methodologies deployed to make this possible, please search for quantization and digital fingerprinting, or wait for my additional writeups regarding this topic.

Next, let us discuss the race-to-the-bottom problem. How would one solve a race-to-the-bottom in the physical realm? We imagine a theoretical physical realm of art where the first person that comes in with a copy of the Mona Lisa is the determined creator (or generally put, owner) of this artwork. What is the problem? Well, readers can imagine naturally that three more individuals can follow with their copies of the Mona Lisa, and claim that they are in fact the creator. It can be suggested that in the physical realm, we can apply whatever analytical methodologies to determine which copy of the Mona Lisa is “genuine”, and determine that the holder of the genuine copy is the creator of the artwork. However, as readers can imagine, even in our real world possession might not be the best method to decide ownership. What if the person holding the genuine Mona Lisa did not create it, what if they simply bought it, or worse, what if they stole it? For the sake of simplicity, let’s discuss the case where the holder of the genuine artwork is claiming to be the creator. Readers, let us explore a funny scenario I imagined to reason through our situation.

A child walks into the classroom, and shows the teacher an impeccable work of art. The teacher asks where the child got such a remarkable painting. The child tells the teacher that they painted it. What can the teacher do to confirm that this is the case and that the child is not lying? Well, for starters, if the paint is fresh and the child has some traces of paint on them, we could make a conclusion using those physical clues. But what if the child was indeed painting, but did not create this piece of work, which happened to be made at the same time? Maybe we can ask the child to paint again. If the child produces something trivial or uncorrelated, we can perhaps argue that the child is lying, that they did not create the artwork in question. But what if we ask the child to submit another painting, and they simply obtain these paintings from their friend, unbeknownst to the friend? Is it possible/reasonable to ask the child to paint in real time? Maybe in the worst case we can ask the child to produce once more, twice more, fifty times more, paintings that we determine show that they are the genuine producer of the artwork in question. Eventually we should have a reasonable idea whether or not the child is the genuine creator.

This funny scenario oddly enough presents a hint of a solution to our race-to-the-bottom problem. No matter how quickly a child can produce a litany of works to claim they are the creator, they cannot possibly continuously produce enough works that pass muster, and especially not if the genuine creator of the works is aware of these claims. In the best case scenario for the abuser of this system, the abuser will claim first with a selection of stolen works, and hope that the genuine creator is unaware of the theft. Once the creator is aware of the claims, it is only a matter of time before the abuser is caught, because in the worst case scenario the genuine creator under this system only has to continue producing works as normal. Continuing our allegory, if the teacher decides not only to simply wait for the child to submit their artwork, but rather starts actively looking in the classroom for artworks similar to the work in question, or better yet, goes beyond the classroom to look for artworks similar to the work in question, then cases of abuse can be found faster. In this scenario, the original genuine creator of any artwork does not even need to be aware that their work is being stolen. The burden is fully (and rightly) on the platform to protect the artist from theft.

So how can we possibly know that a work is from a specific artist just by looking at a selection of works? How can we efficiently go out of our “classroom” and into the world to look for artworks?

For next time, readers, we will look into techniques applied in handwriting identification, medical image segmentation and image recognition. It is shown that machine learning can determine by learning statistical distributions whether a image belongs to the same group or not. Readers can imagine that it should not be out of the realm of possibility that the same models can learn statistical distributions regarding whether a piece of digital artwork belongs to a specific artist.

Thank you for reading, I hope it was interesting.