During Elon Musk’s recent interview with Andrew Ross Sorkin at the New York Times DealBook Summit (around 56:00), Mr. Sorkin asked him a question about copyright and AI training which drew an interesting reply. And probably made Mr. Musk an expert witness.

SORKIN

Can I ask you an interesting IP issue…one of the things about training on data is the idea that these things are not being trained on peoples copyrighted information historically, that’s been the concept.

MUSK

That’s a huge lie.

SORKIN

Say that again?

MUSK

These AIs are all trained on copyrighted data, obviously.

SORKIN

So you think it’s a lie when OpenAI says…none of these guys say they are training on copyrighted data?

MUSK

That’s a lie.

SORKIN

A straight up lie.

MUSK

A straight up lie. 100%. Obviously it’s been trained on copyrighted data.

This is why I keep saying that the Big Tech narrative around artificial intelligence is straight out of the Internet in 1999. In 2024 it’s largely the same people making the same arguments about why they should be getting away with mass infringement with even less of a leg to stand on than they had in 1999. The difference is that instead of the so-called “DMCA license” based on whack a mole and being the bigger bully as it was in 1999, Big Tech is now trying to get away with the so-called “fair use license” based on whack a mole and being an even bigger bully. The difference is that this time you have publicly traded companies with obligations to shareholders as well as multi-million dollar venture backed startups where the investor bet is based on getting away with it.

This all starts with the training of databases upon which AI is based. Mr. Musk gives us a succinct assessment: If it smells like bullshit and looks like bullshit it’s because it actually is bullshit. Or as he says, a straight up lie. And he’s in a position to know.

But understand–it’s not that these AI purveyors refuse to get a license; it’s that they don’t want to if they can get away without it. And be clear–this scraping of all data on the Internet has already happened. I would argue that it has been happening since the Google Books case if not before. Big Tech will be happy to get a license when copyright owners get a final, non-appealable judgement on a copyright-by-copyright basis in every country of the world. Unless Big Tech’s lobbying muscle can deliver another safe harbor for them first, preferably a retroactive safe harbor just like the idiotic retroactive safe harbor in Title I of the Music Modernization Act–and don’t think they didn’t notice that stupidity that handed leverage over tied up with a bow on it.

This is all really unfortunate, though. It didn’t have to be this way, but just like Google et al got really bad advice from Lessig et al that ended up costing them hundreds of millions, the AI companies are getting really bad advice about fair use, etc. And so it goes.

UMG Recordings, Inc., Capitol Records, LLC, Sony Music Entertainment, Atlantic Recording Corporation, Atlantic Records Group LLC, Rhino Entertainment LLC, The All Blacks U.S.A., Inc., Warner Music International Services Limited, and Warner Records Inc., filed two separate mass infringement cases based on copyrighted sound recordings that were copied and exploited without permission by two multi-million-dollar music generation services, Suno and Udio (links to complaints) (and no relation to Uday and Qusay). (If you follow Ed Newton-Rex, who now runs the nonprofit Fairly Trained, the names Suno and Udio will be familiar to you.)

The two cases have some commonalities and really come down to this:

“Given that the foundation of [these businesses] has been to exploit copyrighted sound recordings without permission, [they have] been deliberately evasive about what exactly [they have] copied. This is unsurprising. After all, to answer thatquestion honestly would be to admit willful copyright infringement on an almost unimaginable scale.”

Sound familiar? Can the EFF be far behind?

But here’s the thing: It didn’t have to be this way. The labels, and in particular Universal, have been way out ahead of the AI training issue to create a legitimate market. All these AI companies had legitimate alternatives; they chose to be “deliberately evasive.” The AI companies should welcome the labels’ efforts, especially those that evolve specific principles and goal posts that make success easy to identify. It is axiomatic that successful commerce requires a functioning legal system that respects property rights, not chaos. As the revered Peruvian economist Hernando de Soto taught us in his seminal work, The Mystery of Capital, “[T]he great practitioners of capitalism…were able to reveal and extract capital where others saw only junk by devising new ways to represent the invisible potential that is locked up in the assets we accumulate.”

But yet these two AI companies (and many others potentially) knowingly found themselves a new name: Defendants. RIAA CEO Mitch Glazier said of the lawsuits, “The music community has embraced AI and we are already partnering and collaborating with responsible developers to build sustainable AI tools centered on human creativity that put artists and songwriters in charge. But we can only succeed if developers are willing to work together with us. Unlicensed services like Suno and Udio that claim it’s ‘fair’ to copy an artist’s life’s work and exploit it for their own profit without consent or pay set back the promise of genuinely innovative AI for us all.”

The reality is that both these services made a choice: Let’s try to get away with it. If you listen to Mr. Musk, it’s not just these two–it’s all of them. And their day will come. In the words of James Baldwin, no more water, the fire next time.