Blockchain's Role in Shaping an Ethical Future for AI Music
The Recording Industry Association of America's announcement that its filing lawsuits against the AI-powered music generation startups Suno and Udio were not a surprise, given the growing clamor and fuss around how the AI industry is trampling all over copyright laws.
If the two lawsuits actually make it to trial, and some kind of deal is not negotiated between the startups and the record labels that the RIAA represents -- Warner Studios, Universal Music Group and Sony Music -- they are likely to represent a watershed moment. One that could finally put the brakes on the breakneck pace of development in the AI industry and help to ease concerns that they're going to put millions of creators out of business.
Suno and Udio, like most other AI companies, have built their models by training them on thousands, if not millions of copyrighted works. They've basically admitted as much, as have some of their investors, and if they go ahead and fight the charges against them, it probably won't end in their favor.
The RIAA's lawsuits are almost identical, accusing Suno and Udio of pilfering basically every recorded piece of music in the world, in order to create extensive datasets that are used to hone the extraordinary capabilities of their AI music generation models.
The Evidence Is Overwhelming
The lawsuits argue that the term "generated" is a misnomer, for in reality what Suno and Udio do is pair user's prompts with various patterns in their datasets, then attempt to replicate that pattern. In other words, they basically generate copies of existing songs, with tweaks in the lyrics and instrumentation, that sound very much like the originals that inspired them.
Lawyers for the RIAA cited numerous examples of publicly available AI-generated tracks as evidence that Suno and Udio are using copyrighted songs as the basis of their models' training. For instance, they listed a song called “Deep down in Louisiana close to New Orle”, which bears a striking resemblance to the Chuck Berry hit "Johnny B. Goode". Meanwhile, the AI-authored "Prancing Queen" sounds so much like Abba's "Dancing Queen", that if you were drunk, you might well end up confusing the two and singing the original lyrics.
Those examples, and the many others listed in the lawsuits, provide overwhelming evidence that Suno and Udio almost certainly did train their models on copyrighted data. Such claims are backed by the public statements of both the companies' executives, and also some of their investors.
Take Suno investor Antonio Rodriguez's comments in an interview with the Rolling Stone earlier this year, where he openly admitted that he probably wouldn't have invested in the company had it agreed deals with record labels to use their content.
"I think that they needed to make this product without the constraints," he said.
Fair Use Claims Don't Hold Water
Suno and Udio, like many other AI companies, fall back on the argument that their use of copyrighted songs is legal because it amounts to "fair use", which is a doctrine that allows such content to be used in certain, limited circumstances. Admittedly, fair use is a somewhat vague concept. According to Wikipedia, justifiable examples of fair use include commentary, criticism, search engines, parody, research and news reporting. But by generating songs that are, to all intents and purposes, just remixes of existing music, Suno and Udio appear to have gone far beyond what fair use was originally intended to do.
The startups might have been able to get away with the "fair use" argument to counter individual criticism of what they're doing, but they're unlikely to be able to stand up to the expensive, high-powered legal teams employed by the RIIA to press home those copyright infringement charges.
In its lawsuit, the RIAA points to thousands of what can best be described as garbled versions of copyrighted hits from hundreds of the music industry's best known artists. Creating these mockups would be impossible if the originals were not embedded in their training datasets.
Suno and Udio are perhaps a little unlucky in that they choose to target the music making industry, for it makes them a much easier target than other AI companies, such as ChatGPT's OpenAI. While ChatGPT has been accused of copying text, something such as an AI-generated summary of a book can legitimately be scraped together from content that is free to use, such as search engine descriptions and online reviews. As such, it's likely going to be much harder to prove copyright infringement against such companies.
But in the AI music generation industry it's different. It's simply impossible for the companies to claim that their AI training datasets didn't include more than a few seconds of Abba's Dancing Queen, and suddenly work out how to generate the entire song, albeit in a garbled way.
The charges don't look good, either for Suno or Udio, as the RIAA can make a compelling argument that not only are they stealing copyrighted content, but also making money from it, causing harm to the artists who own the originals of its mishmash AI hits. Should they be found guilty, they would not only have to pay back everything they earned, but could also be shut down entirely
A Deal Will Be Done
There are no real precedents to the charges against Suno and Udio, which makes it very hard to predict what the outcome of these lawsuits will be, and there is good reason to think that ultimately, some kind of deal might be done behind the scenes. Whoever presides over the case is likely to compel the two companies to reveal what data was used to train their AI models, and should those datasets expose their guilt, they'll have few options but to try and negotiate an agreement with the record labels.
The alternative would be losing the case and trying to restart their business with models trained on music that isn't copyrighted. The problem with that is that there's a lot less of it around, which means the quality of their AI-generated music would be much less convincing for it. And that could well be enough to see them put out of business.
Tech industry analyst Rob Enderle of the Enderle Group told SiliconANGLE that the most likely outcome is that Suno and Udio will ultimately start partnering with the music labels, under some sort of content licensing scheme. He proposed a system, ironically powered by AI, that would be able to listen to AI-generated songs and rate how similar they are to the songs on which they were trained on. Those that are too similar might be deleted, while those with enough variation might be deemed acceptable, though the original artist will be entitled to a share of the proceeds generated by the AI equivalent.
“Some of the proceeds could go to the artists who created the original source material as compensation for the use of their intellectual property,” Enderle told SiliconANGLE. He added that the mechanisms for musicians to license their music to someone who wants to adapt it already exist -- they're used all the time when artists perform "covers" of well-known songs, he said.
The challenge is that this process will need to be automated to keep up with the speed of AI music generation, and there are precedents for this as well.
Blockchain Can Make This Happen
The UAE-based "ethical AI" startup droppGroup has created a model that could easily be applied to the music industry. Its droppLink service is designed to reward content creators whenever their intellectual property is used by an AI model to create new/alternative content. The system is built on blockchain and uses cryptocurrency and smart contracts to automatically pay out compensation to artists.
droppGroup's blockchain system utilizes a "Data Genesis" protocol that makes it possible to trace the data that was used to generate an AI output, so it can identify the rights' holder and compensate them appropriately. Its Proof-of-Gen consensus mechanism validates the authenticity of the training data in question, with the participation of incentivized Computational Resource Patrons, who are individuals that work together to verify each request.
The final piece of the puzzle is the droppCoin cryptocurrency, which is automatically deposited into content creators' wallets whenever their data is used by an AI system.
A More Ethical AI Future Awaits
With the foundations of a more ethical AI economy in place, the RIAA's lawsuits could well be the catalyst that forces AI companies to start acting more ethically. If Suno and Udio are forced to do a deal, or worse, found guilty of copyright fraud, the consequences for the AI industry might be Earth-shattering.
Ever since ChatGPT burst onto the scene in late 2022, investors have been scrambling to throw as many millions of dollars at AI startups as they can muster. For sure they know these bets are a gamble, and not all of them will pay off, but if a substantial investment such as Suno, which pocketed $125 million in May, disappears in a puff of smoke, they're going to respond by being a lot more cautious in future.
The AI companies themselves will also take note. Indeed, industry leaders like OpenAI are already taking steps to avoid liability, signing content licensing deals with news organizations left, right and center. If the judge rules against Suno and Udio, it will serve as a stark warning to others that they'd better get their house in order, and make sure that whoever supplies the content that informs their generated outputs is compensated appropriately.
Ultimately, it seems that the RIAA's lawsuits are just what the AI industry needs. It needs to take a step back, think about what it's doing, and make sure that its models don't end up destroying the livelihoods of the people whose talents made it possible for them to exist.
***
DISCLAIMER
The views, the opinions and the positions expressed in this article are those of the author alone and do not necessarily represent those of https://www.cryptowisser.com/ or any company or individual affiliated with https://www.cryptowisser.com/. We do not guarantee the accuracy, completeness or validity of any statements made within this article. We accept no liability for any errors, omissions or representations. The copyright of this content belongs to the author. Any liability with regards to infringement of intellectual property rights also remains with them.