How can you tell if music is AI-generated?

A new song is making the rounds and, in the immortal words of Kylie Minogue, you can’t get it out of your head.

But what if it was created by a robot or the artist himself is a product of artificial intelligence (AI)? Are streaming sites required to label music as produced by AI? So is it important that you like what you hear?

A survey published last week 97% of respondents recommended It failed to detect a song created by the AI. But there are some signs if you know where to look.

Here’s a quick guide.

AI music became one of the hottest topics last summer after accusations that the band The Velvet Sundown were created by AI sent them viral.

The band, which had no record label and little social media presence, quickly reached hundreds of thousands of monthly listeners on Spotify after releasing two albums just a few weeks apart, and the music world became suspicious.

The band initially denied the allegations, later describing themselves as a synthetic project “driven by human creative direction and composed, performed and visualized with the support of artificial intelligence.”

They claimed the project was an “artistic provocation” and not a gimmick, but many fans felt betrayed.

Internet sleuths were suspicious of the band’s airbrushed photos, which featured non-descript backgrounds and a warm orange filter.

There was also no record of them performing live; There were no positive comments from fans or concert photos or videos online. band members did not give interviews and it was observed that they did not have individual social media accounts.

Looking at an artist’s real-life and social media presence can be a useful indicator of whether they’re real or not. But experts tell the BBC that rapidly evolving, complex technology means it’s becoming increasingly difficult to know when a song has been created using AI.

Still, they say, although misleading, there are signs that listeners can be wary of.

LJ Rich recalls that when he started creating AI music about five years ago, it could only produce three seconds at a time, and it took about 10 hours to create one minute of audio.

An entire song can now be quickly called up with a single prompt, fueling what industry experts describe as an “explosion” of AI music, sometimes called “slop,” on streaming platforms.

The musician and tech speaker says a song with a formulaic feel (sweet but without much substance or emotional weight), along with vocals that feel breathless, could be a sign of AI.

AI songs tend to stick to general verse-chorus structures and often lack a satisfying ending. Rich says AI is also more likely to create lyrics that follow correct grammatical structure; But some of the most beautiful or memorable words written by humans are not always meaningful.

Just ask Alicia Keys and her “concrete jungle where dreams are made” or The Rolling Stones and their flirtation with double negatives on (I Can’t Get No) Satisfaction.

The former BBC Click presenter continues: “If it doesn’t sound sentimental, it’s a really big role.” “Does it create the tension and resolution that are a fundamental part of the music we love? Is there a story in it?”

Another indicator is unrealistic productivity levels. Prof Gina Neff, from the Minderoo Center for Technology and Democracy at the University of Cambridge, explains how an artist who recently released multiple sound-alike albums simultaneously is believed to be artificial intelligence.

Their songs resembled a mix of 80s rock bands; Like “true classic rock hits put in a blender.”

“This will be fine as background music for most people,” he continues, “but of course it won’t work for creating future superstars who draw from the past and then create something completely new from it.”

Sometimes what stands out is a song that sounds almost too perfect, devoid of minor flaws and differences.

According to Tony Rigg, a music industry consultant and lecturer in music industry management at the University of Lancashire, this can mean there is no tension in the vocals and the production is overly polished.

He adds that odd phrasing, unnatural emotional delivery, and generic or repetitive lyrics can also be clues.

“The AI ​​hasn’t felt heartbreak yet… It knows the patterns,” he explains. “What makes music human is not just the sound, but the stories behind it.”

It is also worth paying close attention to the vocals. AI “singers” often sound a bit slurred. Consonants and plosives (hard sounds like “p” and “t”) are not quite right. You may hear “ghost” harmonies where background vocals appear and disappear randomly.

However, Rigg calls these signs “hints, not evidence” and admits that it is not very easy for a casual listener to identify songs created by artificial intelligence.

In addition to being used to create entire songs, AI is becoming a tool some established artists use to fuel their creativity.

Currently, artists have no obligation or consistent way to let fans know if and how they use AI.

Some are very obvious: For example, the Beatles used machine learning Removing John Lennon’s voice from a cassette recording in the 1970s and releasing the song Now and Then, which they call their “last song”, in 2023.

Artists including Imogen Heap and Timbaland have created AI personalities and released singles under their names.

Last month, Heap released the song Aftercare with AI model ai.Mogen training on her voice.

She created the voice model as a chatbot – a “desperate attempt” to keep up with the overwhelming number of messages and requests, including from fans – but recently the model has appeared on several songs, allowing Heap to take part in more collaborations than she otherwise would have due to time constraints.

While “if you really know my voice it sounds different,” he says he put a lot of effort into making the AI ​​version of his voice sound human and didn’t think listeners would be able to tell.

Heap is not trying to mislead listeners; ai.Mogen is listed as a co-contributor on the track.

But he hopes that if people feel a human connection to the song without knowing that some of the vocals were sung by an AI model, they might re-evaluate their preconceived negative ideas or fears about AI.

“I hope people listen, don’t notice, and find peace in that,” he tells the BBC.

He actually says he’s not against using AI to create music, but it’s not something he can do yet.

Heap believes there needs to be more transparency about the content of a song and how AI is used.

“We need this for music, we need this for artificial intelligence,” he says, citing the example of someone who reads the label of a microwaved meal and learns its ingredients.

There is currently no legal requirement for streaming platforms to tag AI-generated songs, despite increasing calls for them to flag such tracks.

In January, streaming platform Deezer launched an AI detection tool, followed this summer by a follow-up system: Tags AI-generated music.

Deezer says its detection system can flag tracks made by the most efficient AI music creation tools and is working to expand its ability to detect music made by others. The risk of false positive results (for example, a trace created by a human being incorrectly marked) is stated to be very low.

This week, the company said that one-third (34%) of the content uploaded to its platform is generated entirely by AI, or about 50,000 pieces per day.

Manuel Moussallam, Deezer’s director of research, said his team was surprised at how many tracks were flagged when the detector first started up and were “pretty convinced there was a problem.”

The tool quickly flagged the music of the band The Velvet Sundown, which went viral over the summer, as “100% AI-generated.”

Other platforms have recently announced steps towards greater transparency.

In September, Spotify said it would We’re introducing a new spam filter towards the end of this year to identify “bad actors” and prevent listeners from being advised to “slop.” It removed more than 75 million spam tracks last year.

It also supports a method that allows artists to tell where and how AI is used in a piece, through a system created by a consortium of industry members called DDEX. This information will be added to a track’s metadata and displayed in its app.

Spotify says this is about “strengthening trust” as well as recognizing listeners’ desire to learn more.

“This isn’t about punishing artists who use AI responsibly or downgrading them to revealing information about how pieces are made.”

If you’ve fallen in love with a new artist, does it matter if they or their songs are created by AI?

Some believe that the existence of artificial intelligence is irrelevant; Participation is enjoyable and the music people love already serves its primary purpose.

Others say music fans should be able to make informed choices about what they listen to.

Artists share deep concerns about the impact of artificial intelligence, including hundreds of musicians. Dua Lipa and Sir Elton John protest the use of their songs in training artificial intelligence tools.

According to LJ Rich, the use of artificial intelligence in music raises many “weird and beautiful ethical questions” that have not yet been answered.

“For example, if music makes the hairs on the back of your neck stand up, does it matter whether an artificial intelligence wrote it or not?”