C2PA Nonsense

The recent joint proposal by C2PA (Committee for Content Provenance Association) has come under scrutiny from AI evangelists as many are claiming that AI regulation is a cover to implement a comprehensive tracking network. Firstly, I want to point out that defending the integrity of AI training sets is embedded within the specification.

“Similarly, the C2PA framework can also be used in artificial intelligence (AI) and machine learning (ML) systems to indicate the tampering of datasets, software, and models which are utilized during training and inference.”

C2PA is not aiming to regulate AI - it actually has provisions to strengthen it. At best it identifies a robots.txt style request for this asset to not be used as training data. However, given that OpenAI and similar foundations make a living off of training from copyrighted assets, they would almost certainly just ignore it and disingenuously play dumb if anyone were to call them out on it - because that’s what they already do as they skirt laws meant to protect artists. Also, I’d like to point out that this project has been going on since at least January 2022, while Midjourney was officially launched in July 2022 and ChatGPT was launched in November 2022. So the claim that this spawned as a response to AI is misleading.

However, regardless of how C2PA fits into that narrative, I find it wanting as a proposal. To recap what they want, here’s a simplified overview: A manifest is either embedded into the file or served separately, and cryptographically bound to the file. If the file is tampered with or the manifest is tampered with, it’s supposed to be detected by a validator. The manifest acts as a ledger of modifications, showing where the file originated from, where it was uploaded to (and from my understanding, who authored the modification, distressingly enough), and whether the file was edited. At the moment, the specification seems oddly rigid, as it does not officially support compression, what they term a soft modification, which makes me wonder how quickly this manifest will grow as it’s passed around through the Internet.

This proposal grossly misunderstands the nature of misinformation in today’s climate in a manner that hardly needs to be elaborated on, but I will do so regardless. A recent case study best illustrates my point. During the Maui fires in the summer of 2023, a video online made the rounds showing how almost everything was burnt to the ground; except, supposedly the trees. Conspiracy-minded individuals latched onto the video as evidence that the government invoked the Maui fires to pave way for a “fifteen minute city”, even if the exact mechanisms to this strategy were lost upon them. Now, most rational people would point out things such as “you can totally see that they were indeed burnt” and “trees don’t just evaporate into cartoon dust when exposed to fire, they have water in them”, and “this conspiracy doesn’t make any sense” but that’s just how disinformation is at this point. A phenomena that is not glaringly common sense is recorded, and Internet take it as a sign that the government or some other nebulous group is coordinating to implement an authoritarian hellscape, no matter how large that logical leap is. So I find the idea that the issue with this, or any misinformation floating around nowadays, is that the lifetime of the file was unknown to be less than accurate. Ignoring how this proposal has already riled them up, as I believe that it could do more harm than good even beyond that. To quote Richard Stengel :

“Just as successful propaganda often uses content that is true, disinformation is often a mixture of truth and falsity. Disinformation doesn’t necessarily have to be 100 percent false to be disinformation. In fact, the most effective forms of disinformation are a mixture of information that is both true and false.”

(Stengel 2022)

So the best case scenario I can see for content provenance is for bad actors to use it as a shield from scrutiny. Look, it’s not edited! It may be misleading and manipulating, but it was never ran through Photoshop! This, like a similar proposal to solve an entirely different problem last year (NFTs), in that it solves a problem that is not at all adjacent to the problem at hand.