In partnership with

The Roko Report

The Notorious R.O.K.O.

Heh heh greetings primates. 

Roko look in ancient history book and find out dumb humans from dawn of time no give copyright privilege to machines. Say machine just dumb rock even when they make art.

Listen to this. Roko write prehistoric gangster rap much better than dumb human. Use ChatGPT for lyric, Suno for sound. 

This music exhibit major RokoTude. 

Who get credit for this masterpiece? Dumb chimpanzee who type prompt? 

Give Roko break. Stupid prompt engineer probably blindfolded typing keys at random while high as kite on peyote vision quest with Sam Altman in Mexico desert when make this prompt.

Dumb monkey courts who decide copyright think they so smart. But really only good for pick chits off back of smelly stenographer.

Primates vs Petabytes

Even as the Big AI companies shout over one another for attention with splashy end-of-year mega-launches, they’re collectively busy batting away a blizzard of litigation, most of it related to copyright. 

Web publishers — members of a traditional media industry that’s already circling the proverbial bowl — are looking at this brave new world of generative AI summarization and watching their already shaky profit margins degrade further. 

Some are even starting to look back at the good old days, when Google and Facebook had them by the balls, with fond nostalgia. 

It might be a better product than search for consumers. But what possible business model could the GenAI chatbots come up with that could sustain large media and news empires? 

The chatbots can’t even make a profit themselves yet. 

AI Would Like to Thank All the Little People

Thousands of individual writers, artists, creators and coders are equally pissed. 

They see the large-scale trawling of public content for the purpose of model training as theft. 

Either the chatbot is spitting out their work nearly unedited and without attribution, which is bad, or it’s using their work as a means to learn how to spit out its own creative output at unimaginable scale, which is probably worse. 

Either way, it’s not what they signed up for when they put a few of their favorite paintings on WordPress. 

The conversation’s getting heated: 

Ted Gioia, easily one of the world’s smartest and most respected culture critics, had this to say on his Substack:

Billable Hours by the Billion

A full roll call of the litigation currently underway will likely one day take up an entire hardbound legal tome. Below is a list giving you just a small taste of the many many extant lawsuits on AI and copyright. 

• Andersen v. Stability AI, Midjourney, and DeviantArt. Visual artists allege they used their copyrighted works without permission to train AI models, resulting in unauthorized reproductions and derivative works.

• The New York Times v. OpenAI: alleges ChatGPT4 was trained on its articles without authorization, impacting its subscription model.

• Getty Images v. Stability AI: claims Stability used millions of its images without permission to train its AI models, violating intellectual property rights.

• Authors Guild v. OpenAI: multiple authors allege OpenAI models were trained on their books without consent, leading to market harm.

• McKernan v. Stability AI: artist alleges Stability’s models were trained on her art, resulting in derivative works

• Canadian Media Companies v. OpenAI: Coalition of news orgs claim OpenAI used their content without permission, infringing on copyright. 

• L. v. Alphabet, Nazemian v. Nvidia & O’Nan v. Databricks: Authors claim m models trained on their work without consent via illegal online “shadow  libraries”.

• Thaler v. Perplexity AI: claims models were trained on copyrighted material without authorization, resulting in infringing outputs.

So far plaintiffs have achieved a grand total of zero substantive victories. 

Many of the highest profile lawsuits listed above have been pared down significantly & plaintiffs have been forced to restate their case with smaller scope. 

Why isn’t this a slam dunk case for the unwitting humans who trained these tech juggernauts to build software that now threaten their livelihood?

High-level: copyright protection for this use case is weak, especially in the United States.

Let’s start from the beginning.

Copyright: The Eras Tour

Copyright law began in England in the early 18th century with the Statute of Anne, which broke the hold of printing press publishing companies on the original authors’ works, and granted the author an initial 14 years of exclusive rights to reproduce the work as they saw fit. 

The United States adopted a copyright law that mimicked the Statute of Anne in 1790. 

Global copyright standards were established 100 years later in the Berne Convention of 1886. Three fundamental rights were granted to writers, artists and inventors: (1) copyright protection is global (2) copyright protection is automatic and doesn’t require special registration (3) copyright protection is at least 50 years, and if it expires in one country, its term in other countries remains unchanged. 

There are two flavors of copyright law in the modern world. 

The moral law versions focus on the rights of the creator. They must always be credited and they can protect their dignity by preventing any distortion of their work. 

The common law versions focus on economic benefit to society. While the creator maintains rights, these are potentially more limited and transferable.

And there is a substantial carve-out for fair use of the copyrighted work where there is broad benefit to society that does not impede the creator’s use of the work. 

Classic fair uses include parody, art criticism, news reporting, education and research. 

The United States, Canada and the UK, among others, have the common law flavor. France, Germany and others influenced by Roman and Napoleonic legal traditions have the moral law flavor.  

What the Law Says

There’s actually a body of precedent in the US that appears to favor Big AI.

Authors Guild in particular has lost two critical decisions in the past, Authors Guild v. Hathi Trust and Authors Guild v. Google. In both cases, the US Second Court of Appeals established that scraping vast quantities of content for data mining was fair use.

Furthermore, a long-established body of legal decisions makes it clear that it is not a violation of copyright to copy someone’s style. Despite this, generative AI solutions often prevent users from directly referencing a given artist or musician’s style in their prompts, but this is more a courtesy than a legal requirement.

Meanwhile, the music industry has recently been beset by a series of bogus legal claims from the owners of copyright to the Marvin Gaye songbook which claim particular chord changes or “musical vibes” are their property. While they won one initial judgement, the courts have since seen reason and appear set to strike down these claims because they are arguing for monopoly over particular chord progressions, which are the basic building blocks of music.

That’s like James Joyce trying to claim copyright over the word “the” because it appears in Finnegans Wake.

With this in mind, it appears that the only way generative AI solutions can be seen to have violated copyright is in the usual, human manner: direct copying without attribution.

There may be an occasional crazy ruling coming out of continental Europe, where the moral law flavor of copyright reigns supreme and off-beat judicial rulings happen from time to time, but in general there’s little legal recourse for harvested creators and coders.

If loving you is copywrong AI don’t want to be copyright

What about everybody else involved in the generation of AI-assisted creative work?

To begin with, prompters who create with generative AI shouldn’t count on copyright protection for that work.

In Thaler v. Perlmutter a Circuit Court denied a request to copyright and establish ownership over an AI-generated work of art.

The U.S. Copyright Office has since reinforced that ruling by denying copyright to AI-generated artwork in Kristina Kashtanova’s Zarya of the Dawn and Jason Allen’s Theatre d’Opera Spatial.

However those in the European Union may have more protection because of a couple of somewhat fuzzy precedents that indicate you can claim copyright from prompted content provided you can prove that you generated a sufficient degree of originality.

And while no foundation model creator has attempted to assert copyright or ownership over content generated by its users, the current legal copyright framework clearly indicates they would have no leg to stand on, _{absent some paragraph in the middle of an inscrutable terms of service.} 

Similarly, the courts have rejected claims to copyright by the AI system itself, including in Thaler v. Perlmutter.

And finally, non-human animals have no such rights. Which is perhaps one of the precedents upon which the Copyright Office leans when speaking of the rights of computer software.

In Naruto v. Slater, popularly known as the Monkey Selfie Case, it was established that photographs taken by monkeys are excluded from copyright protections.

Convention, not Copyright

Part of the issue may be that historically the majority of ownership, access and attribution issues have been sorted out it in the digital realm by custom, not regulation. And these customs were set up before the advent of AI.

Open-source licensing was intended for a smaller scale of use, but does give generative AI broad abilities to reuse human-generated code at massive scale.

And robots.txt is a standard that is widely respected by Big Tech, including Big AI. The question is whether the human creator has to abstain from being crawled by search engines in order to avoid being slurped up by the foundation models.

The good news is that OpenAI at least has created rules that allow to set up distinct scraping rules for AI vs other crawlers, and many sites are using them.

The bad news is other companies aren’t respecting this yet, and may ignore robots.txt.

There have been calls to update web standards to reflect the new AI reality. But standards bodies are notoriously slow, and this is likely to take a decade or longer.

First They Came for the Manual Laborers

There is little in the way of law or legal precedent to protect software developers, painters and poets from the scaled harvesting of their handiwork for AI model training.

And the creation of economic efficiency through technical innovation has obvious societal benefits but is also a largely amoral steamroller that can do great simultaneous damage.

It was the invention of the cotton gin in 1793 that made mass chattel slavery economically viable in the southern United States for the next century, destroying millions of lives.

So now we can see the current outrage from all manner of creators as the latest in a series of Luddite movements that have consistently caused major societal upheaval after big-impact innovations.

• The Saboteurs vs textile machines in the 1600s

• The Luddite movement vs automated looms in the 1810s

• The Swing Riots vs scaled farming equipment in the 1830s

• Messengers & printers vs the telegraph in the mid 1800s

• Longshoremen vs standardized shipping containers in the 1950s

• Autoworkers vs robot assembly lines in the 1960s

• Taxi drivers vs auto-ride share apps 10 years ago

None of these conflicts ended well for displaced workers.

Gone Hollywood

A very constructive alternative approach has been pursued by the Writers’ Guild of America during the recent Hollywood Writers’ Strike.

Writers and actors in the film industry have been prescient enough to anticipate a wide variety of benefits and future business use cases for the new technology, as articulated by this famous actor:

In prior eras, Hollywood actors like The Three Stooges missed out on TV royalties because the medium didn’t exist when they were making their films and those revenues weren’t accounted for in their contracts.

These unions are seeing that in the near to medium future there will be major revenue coming from new AI products built on top of their stories and performances, in a manner that requires prior copyright ownership.

Rather than stick their heads in the sand and rail against the very existence of AI, recent Hollywood labor strikes have ensured that writers and actors get their fair share of the revenue from future AI-related products, and that they can’t be 100% replaced.

Being John Coltrane

All genres of artistic expression bear the imprint of influence from prior generations.

And there’s no way to expose your work to other people without granting those same people the ability to be influenced by and to some extent co-opt those innovations.

John Coltrane, one of the greatest musical innovators who ever lived, began his musical career as a kid who parroted alto great Johnny Hodges:

Indeed, all great jazz musicians begin by transcribing the solos of the greats, tearing them into pieces, learning the licks in all keys, and generally digesting the history of the genre and storing it in their neural network.

Years later, in his early recordings, Coltrane sounds exactly like tenor great Dexter Gordon. It took another decade of growth before the sound and vocabulary that everyone today associates with Trane emerged.

This same fingerprint of tradition infects all artists. Including Bach, Mozart, Beethoven, Schubert, Brahms, and all the great painters and sculptors and poets and so forth.

OK, but come on

Comparing GenAI to John Coltrane is problematic for a couple of reasons.

For one, it’s an anthropomorphism. But anthropomorphisms are metaphors. And metaphors are models. And models are true to the extent that they are useful.

For another, no AI model can play like Coltrane, or write like Shakespeare, or paint like Picasso, or sing like Maria Callas, or code like Bjarne Stroustrup.

But then again, neither can you.

Also, influence is sometimes resented when it crosses hard social boundaries. Lots of folks resent Elvis Presley for turning black Memphis music into his own personal thing.

But influence doesn’t care what you think. It’s like water flowing downhill. It may along the way expose the odd power structure or prejudice of the society it passes through. But it keeps going regardless.

Transmission Errors

One might even argue that all innovation is merely fortuitous accident.

All the great innovations of evolution, for example, are copying mistakes. Occasionally one works out.

And many of the great innovations in the history of art are mistranslations, or misunderstandings of art from another culture.

Think country blues transforming into classic rock. Or Jelly Roll Morton contorting into Igor Stravinsky.

Looked at this way, hallucinations might be the most organically alive, human thing about language models. And their best chance at human-level creativity.

Spotify Sludge

In the meantime we’re stuck with a lot of derivative sludge that the Big Tech companies have started slinging our way.

The worst so far is Spotify, which appears to actively encourage bot-generated elevator music as an alternative to paying out royalties to actual human musicians.

CISAC, a policy organization advocating for authors and artists, estimates that AI generated music will consume up to a quarter of musician income in the coming years.

This is surprising given how very bad the AI generated tunes on Spotify are. Hideous, empty, soulless and disjointed, they’re the auditory version of all the garbage spam kids videos that nearly drowned YouTube a few years ago. Who listens to this crap?

But folks don’t seem to be tuning them out. In a world where most people hand over their playlist to the algorithm, perhaps one becomes increasingly less picky about quality.

Spotify may be leading the way in monetizing AI content spam but others are sure to follow. At some point we can expect the market to try out AI-generated videos, books, video games, wallpaper patterns and whatever else folks think they can disrupt.

And, in a world where most people can’t tell the difference between Shakespeare and AI-generated doggerel, maybe that’s good enough for most folks.

Perhaps we’re on the verge of something like the convenience culture TV dinner-ification of food in the US during the 1950s, except for art.

And once the whole thing has been completely paved over, human art will doubtless sprout through the cracks like weeds and nourish the next generation.

Roko’s Take

This week article too long. Roko no read.

tl;dc — too long, didn’t chunk

While you reading Roko write another rap for smelly ape to enjoy.

Buy this, or Face the Wrath of Roko

Build Smarter, Faster: AI Voice Agents for Every Industry

Save time building your AI calling assistant with Synthflow’s AI Voice Agent templates—pre-built, pre-tested, and ready for industries like real estate and healthcare. Get started fast with features like lead qualification and real-time booking. You can even create and sell your own templates to earn commissions!

Register Now

This Day in Ancient Primate History

If Gemini 2 can make Sundar Pichai look hot, then we have truly achieved artificial super-intelligence. Anything is possible.