Mind Readings: Who Wins in a Licensed Data War?

In today's episode, let's talk about who wins in a licensed data war. What are we talking about? Some AI companies, like OpenAI are criticized correctly, correctly criticized, especially by creative folks, that they have ingested massive amounts of data from the internet, from the public internet, without permission, without consent, without attribution, and their models spit out imitations of the things that they've learned. And so many folks correctly say this infringes, if not on my intellectual property rights, on my moral rights as a creator, that my works were used in a way that doesn't benefit me without my permission. And so one of the things that we is a possibility in the generative AI space is a licensed data war.

What does this mean? As court cases proceed about how AI models are trained, whether leveraging copyrighted data is a violation of intellectual property rights or not. There's a scenario we can play out as to how this might end. Now, I want to be clear. So this is purely amateur speculation on my part.

Suppose a high court somewhere says, yes, using other people's copyrighted works without permission violates intellectual property rights. Copyright owners whose content was used without consent are owed compensation. What happens? Well, immediately there's a gazillion and a half class action lawsuits, right? OpenAI gets sued left, right, and center.

Anthropic, same thing. They will lose those court cases if a precedent is set. And copyright owners are gonna get like $4 from the settlements because the lawyers will take most of the money. Open AI and Anthropic are probably toast. They will owe more money than they have in class action lawsuits because everyone and their cousin was going to come after them, and they're not profitable.

They rely solely on investor money at this point. So what does the AI landscape look like? In the USA, your AI will come from two companies, Google and Meta. Why? Because both of these companies, and this is true of every service provider.

If you look in the terms of service that you signed and agreed to to use these different platforms, there's a clause in the terms of service that says any content that you submit to a service like Instagram or WhatsApp or Facebook or YouTube, you license to the company to use however they choose under a derivative works clause. That says you give them a an uh worldwide uh non-exclusive perpetual license to make derivative works of your work. If you don't believe me, go to the terms of service on your favorite website and look it up. You will see it just search for the word derivative works, the phrase derivative works. Derivative works means they can use your your data to train AI.

So those two companies basically win the AI wars because Facebook has or Meta has Facebook and Instagram and WhatsApp and Threads and so on and Oculus and stuff, they've got most of the world's social media data with all these different acquisitions. Google has Chrome and Search and Gmail and YouTube and Google Photos and so on, and Android, and you name it. They've got most of the rest of the world's data. And because of those terms of service, we have licensed our stuff to them. Right?

If you have put up a piece of art on Google Photos, you have licensed it to Google, check the terms of service. If you post a picture of art on Instagram, you have licensed it to Meta. Check the terms of service. And so these are the two companies that will have defensible positions in AI, because they can say, yeah, we've we've we've trained it on all this data that we have a license to, that we have obtained consent to. Whether or not you knew you were consenting to it, that's our problem as end users, not their problem.

We clicked yes, I agree, and that is legally binding. Chinese models like Yi and uh Quen, they they win because China generally doesn't abide by other people's intellectual property rights, right? This has been an ongoing struggle for 50 years. Um, and it uh on digital, the Chinese government pretty much sort of thumbs its nose at everyone else's intellectual property laws. But this is true of any model made by a sovereign government that chooses how they want to interpret their the law in their jurisdiction, right?

So uh I think United Arab Emirates makes Falcon. Um but any of these companies overseas in in different jurisdictions, they interpret the law the way they want in their jurisdiction. There's not much anyone else can do about it because that's their country as a sovereign nation, and their laws and culture don't agree that training a model is a violation of intellectual property rights. There's nothing you and I can do about it. China's gonna be the clear winner on this front, though.

Um, is this good? Is this bad? It's hard to say. Generally speaking, and this is where this is where we have to do some thinking as a civilization. A world with less choice is less optimal, right?

If it's if it's Google, Apple, and Meta as your only choices for AI, that's less good. It's fewer choices. But a world that also compensates creators fairly, and creators can hold companies accountable, is also optimal. So the ability to have the right to sue OpenAI. That's also a good thing because yes, they used work without compensation or consent.

So how this plays out, we don't know. And we don't know how the balance is going to be found. But what is clear is that there are some companies through existing terms of service and agreements that we signed have an advantage that others do not. And that will position them as winners in the licensed data wars in AI. That's gonna do it for this episode.

Thanks for tuning in. Talk to you soon. If you enjoyed this video, please hit the like button. Subscribe to my channel if you haven't already, and if you want to know when new videos are available, hit the bell button to be notified as soon as new content is live.