This is the first article to talk about AI (music) fake news, isn’t it?
Journalists and bloggers like to write attention-grabbing stories. They love to be the first to tell their audience something unique; something that’s completely unheard of and sure to get tongues wagging. There’s nothing wrong with this. It’s the competition for finding and breaking big news stories that improves our society’s awareness of important topics. Take Watergate for example. The relentless investigative work of Woodward and Bernstein was so influential, that it eventually compelled President Nixon to resign.
Unfortunately, there’s a dark side to the search for massively impactful, never-before-heard news. Journalists and bloggers often magnify a story to the point where it actually ceases to be true, and becomes fake news. This happens quite often in scientific journalism — ever heard of the God particle? AI in general, and AI music in particular, have been plagued by this problem for quite some time. I get that it’s juicy to write about esoteric AIs that rely on not-so-specified wizardy to create music by themselves. The thrill of knowing that an AI can come up with creative stuff — something that’s always been distinguishable about human beings — is quite intense. However, what concerns me is that in these reports, oftentimes journalists write things that are just plain false. These false claims could easily be corrected, if they were strict when doing the research, instead of believing the marketing propaganda machine of the companies which built these AI music systems. Is this too much to ask?
What compounds the issue is that, once a story is out ‘in the wild’, it often takes on a life of its own. Secondary and tertiary stories (often more) will simply republish vast amounts of copy without a second thought.
In particular, there’s one bad habit very common in AI music reporting that I find unacceptable and unbearable. I call it the ‘first syndrome’. It seems that being the first at doing something in AI music is the most important factor. Beethoven wasn’t the first composer writing symphonies, but this didn’t impede him from becoming one of the greatest symphonic composers ever. Have you ever heard about Giovanni Battista Sammartini? Probably not. He was one of the first composers to write symphonies. Apparently then, being first at doing something doesn’t necessarily give you an edge — but this is a personal opinion. The real problem arises when someone claims that someone else did something for the first time, although they didn’t. Unfortunately, the ‘first syndrome’ is exactly what happens frequently in AI music coverage today, as we’ll see in the examples below.
This is the first time an AI music system has written a pop song
In 2016, Sony CSL released a pop song in a musical style similar to the Beatles, called Daddy’s Car. The catch is that the song was partially composed by an AI system called Flow Machines. I say ‘partially’, because the AI created a lead sheet (melody and harmony) and relied on a human musician to arrange, produce and mix the song.
The news got a lot of attention on journals and blogs. Some of these wrote that Daddy’s Car was the first example of a pop song fully created by an AI. Simply put, this is fake news. As Sony CSL explained on its channels, Daddy’s Car was co-created by AI and a living, breathing, flesh-and-blood musician. Also, there are tons of AI music systems predating Flow Machines that can create songs in popular musical styles. To name one, the Cybernetic Composer is an AI that generates full rock, ragtime and jazz songs. It was developed in 1992, long before the claims made in 2016.
This is the first AI music composer
AIVA is a French music tech startup company that uses neural networks for generating piano sheet music. Once the AI comes up with a piano score, human musicians select and refine the music. They then orchestrate it and record the score with a live orchestra. A number of media outlets reporting on AIVA claimed that this was the first AI music composer. This is an outrageous claim, because it flatly ignores 60+ years of research and development in AI music. At the time of writing, there are probably thousands of AI systems that can compose music. If you don’t believe me, just take a look at this survey of music generation systems, which only scratches the surface. The first AI music composer that I’m aware of was built in 1957. It composed a string quartet in a contemporary style called the Illiac Suite. This could be literally AIVA’s grandad.
This is the first time an AI music system composed a soundtrack for a video game
In 2017, AIVA produced a soundtrack for a game called Pixelfield “Battle Royale”. They claimed that this was the first time an AI composed a soundtrack for a video game. A number of media outlets broke the story, without even questioning the claim.
Once again, this is demonstrably false. Firstly, it’s misleading to say that the soundtrack was composed by an AI. From my understanding of how AIVA works, there was a lot of human intervention in orchestrating a piano music sheet generated by the AI. Most importantly, there are a ton of instances of AI-generated game soundtracks that predate that of Pixelfield. The most famous one is probably Spore’s soundtrack, composed (should I say programmed?) by Brain Eno. Spore hit the market in 2008, almost 10 years before Pixelfield. Differently from the soundtrack produced by AIVA, which is static, Spore has an AI music system that runs in realtime while you’re playing the game, creating music on the fly. Quite impressive, isn’it?
This is the first album with a collaboration between human and AI
Taryn Southern is a forward thinking songwriter who likes to experiment with technology. She embraced AI music, because she saw the incredible creative potential of human-AI collaboration. In 2018, she released the album I am AI, where she co-created the music with a number of AI music systems like Amper, AIVA and Magenta.
Several journals and blogs reported the story claiming that I am AI is the first album co-created by a human and AI. Guess what? This is false. Brian Eno spent his entire career playing around with AI systems to co-create his music. He even coined the term generative music. Eno released the groundbreaking album Generative Music 1 in 1995, more than 20 years before I am AI.
This is the first record deal an AI composer signed with a music label
Just last month, news came out that Endel, an AI that generates ambient music, signed a record deal with Warner Music. 20 albums generated by Endel will be released through the label. The story spread fast. Even reputable journals like The Guardian reported it. This would indeed be groundbreaking news. The only issue is that there is no (traditional) record deal between Endel and Warner music. Journalist Cherle Hu found out that the relationship is basically a distribution partnership. There’s nothing newsworthy here, because AIVA, for example, already partnered with Believe Distribution, a Sony-owned label, to release its latest album. Kudos must go to Music Ally, who correctly pointed this out.
The misrepresentation of reality
As we’ve seen, the way media outlets report AI music stories is often sensationalist, headline-grabbing fodder. They twist the truth and repurpose it to appease and impress their readers, completely ignoring or disavowing facts. I understand the urge blogs have to get more eyes on their content through click-bait titles, but that can’t justify spreading claims that are blatantly false.
How do bloggers spread fake news? Sometimes they overblow a story by, for example, momentarily succumbing to the ‘first syndrome’. Other times, they aren’t thorough enough with their research, and simply report a story they’ve read elsewhere without double-checking the facts. A few minutes of quick Wikipedia research, for example, could have easily prevented the embarrassment of claiming that AIVA is the first AI composer ever built.
Of course, the misrepresentation of reality isn’t only an issue in the field of AI music, though. All AI verticals suffer from the same problem. To call AI hot right now is an understatement — AI-interest is at an all-time-high, and people are fascinated by the fact that machines are becoming fluent in so many traditionally human-like tasks. This shift itself, could be enough of a reason to perpetuate this sloppy journalistic habit. Whatever the reason, the misrepresentation of reality in AI is a severe issue that needs to be seriously addressed.
Spreading fake news creates false expectations among people. AI has improved by leaps and bounds over the last five years or so. AI agents can outperform humans at very complicated tasks, like medical diagnosis and driving. But there’s still a long way to go before they’ll be able to showcase a level of general intelligence comparable to that of humans. Bombarding people with stories that increase the merits of the AI agents beyond measure, contributes towards overhyping the market. Once people realise that the systems aren’t as intelligent as they’ve been told, the AI market will be at risk of collapsing.
The risk of this happening can be reduced by thorough research and sound journalistic practices. As readers, we can’t double check all the information we get exposed to every day. However, we can be aware of a few red flags. Whenever you read a title with a grand claim like ‘this is the first time an AI has…’, an alarm should go off in your head. Most of the time the claim will be false. This isn’t only true for AI music. It also applies to AI in general, and beyond that, to technology and science. I hope that journalists take these ideas on board when covering innovative industries. That way, when they knock on doors saying, “Show me the AI”, developers, companies and investors can know they’re in the right hands.
