artificial intelligence
Page: 18
LONDON – Key players from the United Kingdom’s music industry have pledged to improve the digital metadata for song recordings to help ensure that artists and rightsholders are paid fairly.
Under the voluntary two-year agreement published Wednesday (May 31) by the U.K. government’s Intellectual Property Office (IPO), the U.K. industry commits to progressively improve metadata for new music recordings and to accurately award credits on streaming services.
Signatories include representatives of major and independent record labels, publishers, creators and streaming services. Trade bodes BPI —which represents more than 500 labels, including the U.K. arms of Universal Music Group, Sony Music Entertainment and Warner Music Group — and the Association of Independent Music (AIM), which acts on behalf of U.K. independent record labels and music companies, are among the music groups backing the pledge
The U.K. government also says that it will be setting up a new working group with music business stakeholders “to explore and consider industry-led actions on remuneration for existing and future creators” over the coming weeks. The working group will review a range of issues facing creators and rightsholders, says the IPO, including the impact of artificial intelligence.
“The U.K. is a hotbed for world-beating musical talent but as technology advances we need our thriving music industry to continue to offer viable career opportunities,” John Whittingdale, a U.K. government minister for the creative industries, says in a statement.
Whittingdale calls the metadata pact a “landmark agreement” that will help ensure musicians “in the digital age are fairly credited and compensated for their contributions and creativity.”
As streaming has grown to become the dominant mode of music consumption globally, the need for accurate metadata — the descriptive data underlying music recordings detailing who contributed what to a song’s creation and who owns it — has grown in importance in determining who gets paid what share of a track’s streaming royalties. But digital platforms, artists and rightsholders have complained that streaming metadata is often incomplete or inaccurate, which delays payments to songwriters and labels.
“Currently a fraction of recordings being streamed properly credit the people that wrote and published the song,” says Graham Davies, CEO of U.K. songwriters trade group The Ivors Academy, which has also signed the metadata agreement.
In 2021 the Ivors Academy estimated that at least £500 million ($625 million) in global streaming royalties a year were being wrongly allocated or not being paid at all due to missing or incomplete metadata. In a statement, Davies says the pledge “gives us the first proper step on a journey of closing the metadata gap.”
Sophie Jones, BPI interim CEO, says she hopes the agreement will lead to “immediate improvements” in the “speed and accuracy of songwriter payments.”
Other music groups and companies backing what is officially titled the United Kingdom Industry Agreement on Music Streaming Metadata include The Digital Entertainment and Retail Association (ERA), whose members include streaming services; Hipgnosis Songs Fund; the Music Publishers Association (MPA); Musicians’ Union (MU); Featured Artists Coalition (The FAC) and U.K. collecting societies PRS for Music and Phonographic Performance Limited (PPL).
Wednesday’s metadata agreement is the latest in an ongoing series of government-led interventions into the U.K. music industry that spun out of a Parliamentary inquiry into the economics of the music streaming business.
That initial review concluded in July 2021 with the Department for Digital, Culture, Media & Sport (DCMS) Select Committee calling into question the major record labels’ dominance of the industry and branding the global streaming model as unsustainable in its current form, saying it “needs a complete reset.”
Since then, the government has initiated multiple investigations and inquiries into the U.K. music industry, including a market study review of the record business by the U.K. competition regulator. That ended in November when the Competition and Markets Authority (CMA) determined that low returns from streaming for most artists “are not the result of ineffective competition” between the three major labels.
Last year the British government set up two working groups – led by the IPO and made up of industry stakeholders — looking at issues raised by the Parliament probe. One focused on metadata, which has now concluded. A second working group is looking at improving transparency around licensing and royalties for artists. The IPO says it is due to publish the second group’s findings, including a code of practice, in the coming months.
Earlier this year, Oleg Stavitsky, co-founder/CEO of Endel, laid out a vision for how his company’s AI-driven functional soundscapes could help the major labels — even as anxiety around AI was reaching new heights. “We can process the stems [the audio building blocks of a track] from Miles Davis’ Kind of Blue and come back with a functional sleep version of that album,” Stavitsky told Billboard. At the time, he said his company was in talks with all the major labels about this possibility.
A few short months later, Stavitsky will have a chance to do exactly that: Endel announced a new partnership with Universal Music Group on Tuesday (May 23). In a statement, Endel’s CEO said his company will put “AI to work and help UMG build new and exciting offerings to promote wellness and banish the perceived threat around AI.”
“Our goal was always to help people focus, relax, and sleep with the power of sound,” Stavitsky added. “AI is the perfect tool for this. Today, seeing our technology being applied to turn your favorite music into functional soundscapes is a dream come true.” Artists from Republic and Interscope will be the first to participate — though the announcement omitted any names — with their soundscapes arriving “within the next few months.”
Endel focuses on creating “sound that is not designed for conscious listening,” Stavitsky told Billboard earlier this year. “Music is something you consciously listen to when you actually want to listen to a song or an album or a melody,” he explained. “What we produce is something that blends with the background and is scientifically engineered to put you in a certain cognitive state.”
Endel’s technology can spit out these soundscapes effectively at the click of a button. “The model is trained using the stems that are either produced in-house by our team, led by co-founder and chief composer Dmitry Evgrafov (who’s himself an established neo-classical artist), or licensed from artists that we’ve worked with,” Stavitsky said. “The trick is all of the stems” — Endel has used stems from James Blake, Miguel and Grimes —”are created following the scientific framework created by our product team in consultation with neuroscientists.”
Some people in the music industry have taken to calling sounds designed for sleep, study, or soothing frayed nerves “functional music.” And while it maintains a low profile, it’s an increasingly popular and lucrative space. “Science tells us that nature sounds and water sounds have a calming effect on your cognitive state,” Stavitsky noted this winter. “So naturally, people are turning to this type of content more and more.”
Early in 2022, Endel estimated that the size of the functional music market is 10 billions streams a month across all platforms. (The company has since raised its estimate to 15 billion streams a month.) If true, that would mean functional music is several times more popular than the biggest superstars. “Every day, hundreds of millions of people are self-medicating with sound,” Stavitsky wrote in March. “If you look at the top 10 most popular playlists at any major streaming service, you’ll see at least 3-4 ‘functional’ playlists: meditation, studying, reading, relaxation, focus, sleep, and so on.”
But this has caused the music industry some concern. Major labels have not historically focused on making this kind of music. Most streaming services pay rights holders according to their share of total plays; when listeners turn to functional music to read a book or wind down after a long day, that means they’re not playing major label artists, and the companies make less money. In a memo to staff in January, UMG CEO Lucian Grainge complained that “consumers are increasingly being guided by algorithms to lower-quality functional content that in some cases can barely pass for ‘music.’”
But record companies can’t eliminate listener demand for functional music. It makes sense, then, that they would try to take over a chunk of the market. And Stavitsky has been savvy, actively pushing Endel’s technology as a way for the labels to “win back market share.”
Back in 2019, Endel entered into a distribution agreement for 20 albums with Warner Music Group. And the company announced its new partnership with UMG this week. In a statement, Michael Nash, UMG’s evp and chief digital officer, praised Endel’s “impressive ingenuity and scientific innovation.”
“We are excited to work together,” Nash continued, “and utilize their patented AI technology to create new music soundscapes — anchored in our artist-centric philosophy — that are designed to enhance audience wellness, powered by AI that respects artists’ rights in its development.”
How magical would it be if we listened to music and music listened back to us?” asks Philip Sheppard, the co-founder/CEO of Lifescore, a U.K. startup that creates soundtracks tailored to users’ functional needs, from sleep to focus to fitness.
Though the premise sounds like science fiction, a number of new companies are already working with technology that attunes music to listeners’ movements in video games, workouts, virtual reality — even the aesthetics of their Snapchat lenses. Much as a film composer accentuates pivotal moments in the story with perfectly timed swells and crescendos, these innovations are being used to create bespoke soundtracks in real time.
One of the most fertile testing grounds for “dynamic” or “personalized” music, as it is called, is the gaming sector. Gamers tend to be avid music fans who stream songs an average of 7.6 hours a week — more than double the rate of the average consumer, according to MIDiA Research — and for some time now, game developers have linked players’ in-game movements to changes in lighting, setting and other parameters to enhance their storytelling.
David Knox, president of Reactional Music, the maker of an interactive music engine for video games, says “the final frontier for innovation in gaming is music.” Until recently, video-game music has consisted of loop-based scores or licensed tracks. Because of its repetitiveness, Knox says many users mute game soundtracks in favor of Spotify or Apple Music.
To compete with this, Knox says Reactional’s “rules-based music engine” applies the same reactive technology used in gaming graphics to the soundtrack, enabling, for example, a player in a first-person-shooter game to fire a gun in time with the beat of a song. As the technology evolves, Knox says soundtracks could transform to reflect the state of a player’s health or the level of danger.
This same technology could work with augmented reality and the so-called metaverse. Minibeats, a company that creates social media lenses with interactive musical elements, is in the process of incorporating dynamic music technology, which it calls “musical cameras,” into its AR filters for Snapchat. For one of its first launches, Minibeats partnered with Rhino and Stax Records in February to promote the 30th anniversary of the release of Booker T. & The M.G.’s’ “Green Onions.” One Minibeats filter turns users’ heads into green onions and allows them to control when the song’s signature Hammond organ riff courses through body and facial movements. Another filter morphs users’ faces into spinning vinyl records, allowing them to control when the song’s guitar and keys start and stop by opening and closing their mouths.
When imagining the future of dynamic music, Mike Butera, the company’s founder and CEO, says Disney’s Fantasia comes to mind. The ambitious 1940 film, which mixes animation and live action and features Mickey Mouse in his iconic sorcerer’s hat, syncs vibrantly colored dream-like visuals with a score that enhances what’s transpiring onscreen. “Imagine if we transformed your day-to-day life into something like that?” Butera says. “The mundanity of drinking coffee, walking the dog and driving to work [turned] into something [that] can be soundtracked with your own personal score that you control, whether that’s through a phone camera or AR glasses.”
These startups all claim that they have received only glowing feedback from the music business so far, and many have formed key partnerships. Hipgnosis recently announced a deal with Reactional Music to help bring its catalog of hit songs to the startup. Bentley and Audi have made deals with Lifescore to get dynamic soundtracks into cars, and Warner Music Group counts itself as an investor as well. Minibeats says it’s “in discussion with all the major labels,” though beyond its Rhino-Stax partnership, the company declined to disclose more details.
These emerging capabilities are typically powered by artificial intelligence to adapt recorded music to malleable experiences, but unlike other AI companies trying to create machine-made music with the touch of a button, these dynamic music startups either license preexisting, human-made songs or commission composers to create new or more dynamic compositions.
Lifescore pays composers to record a number of separate elements of a song, known as “stems,” and then, Sheppard says, its technology works with the resulting audio files like “shuffling a deck of cards,” assembling newfound arrangements in configurations intended to support a user’s need for focus while studying or working, for example, or sleep.
In the case of preexisting tracks, companies like Minibeats partner with Audioshake, a firm that uses AI to break down songs into individual, standardized stems, so that they can easily manipulate a song’s instrumental mix — guitar, drums, vocals, etc. — in real time. Audioshake’s proprietary technology is especially helpful in the case of older recordings in which the copyright owner no longer has the stems.Audioshake founder/CEO Jessica Powell says one reason she thinks the music industry has embraced this innovation is its potential to spur music discovery. “I think the same way TikTok pushes new songs, gaming — as well as other use cases — have enormous potential to introduce people to music,” whether that be a catalog track or a new release.
Though this technology is new, interactivity has been long seen as a way to create powerful bonds between fans and songs. Powell points to popular video games like Guitar Hero and Rock Band as successful examples. Karaoke is another. One could even point to the more recent novelty of stem players, like those Ye peddled during the release of his album Donda 2, as a way of engaging listeners. At a time when much of music discovery is passive — scrolling TikTok, streaming an editorial playlist or turning on the radio — musical interactivity and now personalization promises a stronger bond.
Knox at Reactional says interactive music also has economic potential. In-game purchases — which allow players to buy customizable elements like cars, weapons and outfits — dwarfed global recorded-music revenue in 2020, with players spending $97 billion in-game compared with music’s $35.9 billion (retail values), according to MIDiA Research. “In the same way you put hooks into a game, allowing someone to pay to change their appearance at a certain point, a game developer working with us could create a hook that unlocks access to the Reactional platform, letting players buy their favorite songs,” he says.
Since at least the advent of the transistor radio, consumers have used music to soundtrack their lives, but until recently, personalization of those soundtracks was limited to song selection and playlist curation. The songs themselves were unchangeable. Those on the forefront of dynamic music contend that it marries recorded music with the goose bumps-inducing, real-time experience of listening to something live.
“You know how you listen to a live performance, and the musicians are influenced by your energy in the room?” asks Erin Corsi, director of communications for Lifescore. “That’s what this is. Though this also feels like something new, it feels like we are finally able to get back to how music started.”
Beatdapp co-founders and co-CEOs Morgan Hayduk and Andrew Batey were not initially focused on fighting streaming manipulation. Batey spent years in digital marketing, while Hayduk formerly worked as a lobbyist for the Canadian music industry in the area of copyright protection. At first, they teamed to build an auditing tool that would enable labels to evaluate inconsistencies between their sales reports and streaming services’ server logs. Conversations with label executives indicated that “there were pretty often material discrepancies,” Hayduk says.
As he and Batey tried to understand those inconsistencies, it became clear that streaming manipulation was causing some of them, and Beatdapp embarked on developing a tool to detect fraudulent streams — which Hayduk defines as the leveraging of “bots, stolen accounts or manipulated platform features” to steal streaming income — and prevent them from impacting payouts.
According to a recent report from the Centre National de la Musique (CNM), a government-backed organization that supports France’s music industry, in 2021, over 1 billion music streams — between 1% and 3% of all streams generated in the country that year — were fraudulent. “The methods used by fraudsters are constantly evolving and improving,” the report noted, “and fraud seems to be getting easier and easier to commit.” If that percentage was applied to IFPI’s estimate that global streaming revenue totaled $17.5 billion in 2022, fraudulent streams would amount to $350 million in potential lost income for legitimate rights holders.
Beatdapp’s software sifts through massive amounts of data from partners — including labels, distributors and streaming services — to identify and investigate suspicious patterns. In one case, it identified 10,000 accounts all playing the same 63 tracks. The pair say the company now analyzes hundreds of billions of streams, and while they declined to identify partners, they recently started working with SoundCloud and Napster, according to two sources.
“If we can make this industry less attractive for financial fraudsters, that will make a positive difference for everybody who’s working on music,” Hayduk says. “That’s what animates us.”
Why is streaming fraud an important issue?
MORGAN HAYDUK: It hurts everyone who makes a living in the music industry and, left unchecked, creates this promotional race to the bottom where everyone believes they have to cheat to succeed. In cybersecurity terms, it’s important to shrink the attack surface of the industry.
ANDREW BATEY: In an industry where it’s already hard to make something and then promote something and then get paid, you should at least get paid correctly.
How much data is Beatdapp analyzing at this point?
HAYDUK: We’re looking at about 320 billion streams now. That’s about 13 trillion individualized streaming data points when you account for all of the metadata associated with each of those streams. We expect to add data in the neighborhood of another 50 billion streams in [the second quarter] and about another 2 trillion data points on that.
BATEY: It’s not just the individual stream. You might make 12 decisions in an app, such as how you search — if you clicked on the artist first and then you looked at their song list. We’re capturing all of that, anonymized across users. All of that context helps us because if somebody consistently hits, let’s say, the exact 11 things for every song they play, that’s a pretty obvious case of fraud if they’ve done that 3,000 times in a week.
How has the industry’s perception of streaming fraud changed since you started Beatdapp?
HAYDUK: Just hearing people acknowledge the issue is probably the biggest shift. It used to be verboten to speak publicly about streaming fraud. It was all behind closed doors. But I don’t think you can fix a problem until you accept its existence. We’re starting to get there now and [are] seeing a more widespread willingness to put in place solutions.
How has your perception of the problem changed as your data set has expanded?
HAYDUK: The biggest revelation to us has to be that this is way closer to death by a thousand paper cuts than it is a top-of-the-market problem. If you asked us where most of the fraud came from 18 months ago, we probably would have pointed the finger at bigger artists because we would have thought they had the most to gain. But we were missing the point of most of this activity. It’s not about changing perception; it’s about making money. This isn’t a phenomenon that’s driven by major labels and major independent label artists or their top artists. The overwhelming majority, like upwards of 80% of what we see is fraud, is coming from — call it non-music content. It’s not being released for popular consumption or because these are artists who are trying to get noticed. These are releases that have no commercial purpose except as [instruments of] fraud.
BATEY: When we first started, we genuinely thought fraud would be 1% to 3%. Now we think it’s closer to 10% [though some of this is caught]. Also we would have guessed that most of the fraud would occur on the platforms where people were — Spotify, Apple, YouTube. But because it’s a lot of financially motivated fraud, what we actually see is that it’s easier for the fraudsters to attack all the mid- and long-tail [digital service providers] as well, where they’re less likely to get caught and they’ll get a similar or better per-stream payout. Why not target all of these smaller DSPs with zero protections in place and get paid across all of them?
France’s CNM recently came to the conclusion that fraud is getting easier to commit.
BATEY: I 100% agree with that. There are so many ways to exploit platforms. If your job is to deliver the best user experience possible, it often means making it easy for them to access that content and creating really cool ways for them to experience or engage with that content. [When that happens,] there are more ways to manipulate that content for the purpose of exploiting it for a payout.
HAYDUK: And the tools that you need to commit fraud effectively and at scale are easier to access now than ever before. The tools that facilitate fraud in e-commerce or ticketing or financial services are also repackaged and repurposed to commit streaming fraud. You can generate fully automated online bot farms using cloud computing in a way you couldn’t 10 years ago.
How do you avoid generating false positives when you’re hunting for fraud?
HAYDUK: We know that a false positive is worth considerably more in the loss column than a false negative, so we adjust our models to account for the fact that they need to be conservative in the right ways.
BATEY: You can’t get it wrong. If you miss a fraudster, it’s OK. We hope we catch them later. If we call something fraud that’s not, that’s way worse.
Some have suggested that a user-centric payout system might mitigate fraud.
HAYDUK: Our view is that it’s not going to make that big a difference. It’ll change the tactics, but it won’t change the motivation. It’s a big pot of money on the internet, and generally speaking, the DSPs are still fairly soft targets. A different payout structure will just change the tactics that fraudsters use to aggregate money and divert it their way. Obviously, there’s a whole different case for the merits of payout systems if you’re an artist or you’re a label.
There’s a lot of industry concern about artificial intelligence right now. To what extent does AI make it even easier to commit these types of fraudulent activities?
HAYDUK: It’s a tool. We work for some good AI companies that care about not being a tool for fraudsters. That said, the new models are incredibly powerful, and you can create content at scale. There’s no putting the genie back in the bottle when some of these tools emerge. The tougher we make it to get away with fraud, the less valuable the tool becomes in the hands of someone who’s wielding it for a bad purpose.
How incentivized are DSPs to care about fraud?
HAYDUK: Their biggest partners care, especially in light of what we said earlier: Market share shifts matter to the partners and, therefore, it matters to the DSPs. I think consumers also care because bad recommendations on the DSP side make for bad user experience. And given that every platform is offering roughly the same catalog to the consumer, if your recommendations are substandard, that makes consumers more inclined to choose your competitor.
Some music industry executives worry that public discussions of fraud undermine user confidence.
HAYDUK: How many times a week does your bank email you about the extra efforts they’re taking to protect you from fraud in the financial sector? It doesn’t make me want to boycott my bank when they tell me that. Fans probably want to hear that, as an industry, we’re taking steps so that the artists they care about are paid correctly.
BATEY: If you’re the consumer, your account was hijacked, and now you’re getting a bunch of recommended songs that don’t make any sense, you’re not blaming the fraudster — you’re blaming the platform.
What is your dream scenario for fraud mitigation in the industry?
HAYDUK: Our view is there are some things you can’t do in a vacuum. DSP A can’t look at the data from DSP B to help inform its own detection models. It’s way too competitive between the platforms to give up the level of data required to do fraud detection at the highest levels. Having a platform in the middle acting as Switzerland, working for the collective benefit of everyone without minimizing the level of competitiveness between the platforms, is the right approach. And it’s also an approach that we’ve seen play out in other verticals with similar dynamics.
The three major label groups have been in talks with the big music streaming services to find a way to get them to remove recordings with AI-generated vocals created to sound like popular artists, Billboard has learned. The idea under discussion with Spotify, Apple Music and Amazon Music would operate much like the one laid out by the Digital Millennium Copyright Act but would cite violations of rights of publicity, rather than copyright, according to sources at all three majors. Unlike the DMCA, however, this arrangement appears to be voluntary.
The 1998 DMCA gives online services that use, store or transmit copyrighted works a “safe harbor” from secondary liability for copyright infringement as long as they abide by a notice-and-takedown system that allows rightsholders to ask them to remove copyrighted content. That law would not apply to most AI-generated soundalike tracks because they do not infringe protected elements of copyrighted recordings or compositions but rather a trademark or a right of publicity, the protection celebrities may be able to receive to protect their names and likenesses from unauthorized commercial exploitation.
Songs that imitate the voices of big-name talent have become a trend over the past month, reaching widespread attention in mid-April when the track “Heart on My Sleeve,” which apparently used AI to mimic the style and tone of vocals by Drake and The Weeknd, was uploaded to streaming services and then swiftly removed. (The song did not credit those artists, although they were referred to in social media posts about it.)
Citing rights of publicity can be more complicated than copyright, because they are matters of state law in the U.S., backed by limited legal precedent. Rights vary by state, protections for deceased artists vary even more widely, and the use of soundalike vocals for creative purposes may in some cases be protected as free speech. Further complicating matters, these rights almost always belong to artists, not labels, which would presumably file notices on their behalf with authorization. Right now, however, this is the most obvious legal argument with which to keep AI-generated soundalikes off major streaming platforms.
In an April 26 earnings call, UMG CEO and chairman Lucian Grainge seemed to signal this approach to investors. “The recent explosive development in generative AI will… create rights issues with respect to existing copyright law, in the US and other countries, as well as laws governing trademark, name and likeness, voice impersonation, and right of publicity,” he said. “Further, we have provisions in our commercial contracts that provide additional protections.” It is not clear if takedowns issued by the majors would rely on these provisions, state law, goodwill, or some combination.
Some executives have raised concerns that AI soundalikes that imitate the voices of popular artists could result in consumer confusion. Still, a few artists like Grimes and Holly Herndon have embraced the technology, training their own AI voice models and making them available to the public.
Meanwhile, companies like Uberduck, Supertone, Lingyin Engine, and Covers.ai are marketing models with which to replicate voices. Covers.ai, which launched last week, has said that it received over 100,000 sign-ups in anticipation. Tencent Music Entertainment executives announced in November that with the company’s Lingyin Engine they had created and released over 1,000 songs containing synthetic AI voices already, one of which amassed 100 million streams.
This stance taken by the leading streaming services counters a recent announcement from the blockchain-based music platform Audius, which announced that artists can now “opt-in” to allow AI-generated works on their artist page. To organize this new music and avoid confusion, Audius would create a separate tab on the artists’ page especially for user-generated content.
Representatives for Universal, Sony, Warner, Spotify, Apple Music and Amazon Music did not respond to requests for comment.
A U.S. senator representing Music City had tough questions about artificial intelligence’s impact on the music industry during a Congressional hearing on Tuesday, at one point asking the CEO of the company behind ChatGPT to commit to not using copyrighted songs to train future machines.
At a hearing before the Senate Judiciary Committee about potential regulation for AI, Sen. Marsha Blackburn (R-Tenn.) repeatedly grilled Sam Altman, CEO of OpenAI, over how songwriters and musical artists should be compensated when their works are used by AI companies.
Opening her questioning, Blackburn said she had used OpenAI’s Jukebox to create a song that mimicked Garth Brooks – and that she was clearly concerned about how the singer’s music and voice had been used to create such a tool.
“You’re training it on these copyrighted songs,” Blackburn told Altman. “How do you compensate the artist?”
“If I can go in and say ‘write me a song that sounds like Garth Brooks,’ and it takes part of an existing song, there has to be compensation to that artist for that utilization and that use,” Blackburn said. “If it was radio play, it would be there. If it was streaming, it would be there.”
At one point, Blackburn demanded a firm answer: “Can you commit, as you’ve done with consumer data, not to train [AI models] on artists’ and songwriters’ copyrighted works, or use their voices and their likenesses without first receiving their consent?”
Though Altman did not directly answer that question, he repeatedly told the senator that artists “deserve control” over how their copyrighted music and their voices were used by AI companies.
“We think that content creators need to benefit from this technology,” Altman told the committee. “Exactly what the economic model is, we’re still talking to artists and content owners about what they want. I think there’s a lot of ways this can happen. But very clearly, no matter what the law is, the right thing to do is to make sure people get significant upside benefit from this new technology.”
Blackburn’s questioning came amid a far broader discussion of the potential risks posed by AI, including existential threats to democracy, major harm to the labor market, and the widespread proliferation of misinformation. One witness, a New York University professor and expert in artificial intelligence, told the lawmakers that it poses problems “on a scale that humanity has not seen before.”
The music industry, too, is worried about AI-driven disruption. Last month, a new song featuring AI-generated fake vocals from Drake and The Weeknd went viral, underscoring growing concerns about AI’s impact on music and highlighting the legal uncertainties that surround it.
One of the biggest open questions is over whether copyrighted music can be used to train AI platforms – the process whereby machines “learn” to spit out new creations by ingesting millions of existing works. Major labels and other industry players have already said that such training is illegal, and cutting-edge litigation against the creators of such platforms could be coming soon.
At Tuesday’s hearing, in repeatedly asking Altman to weigh in on that question, Blackburn drew historical parallels to the last major technological disruption to wreak havoc on the music industry — a scenario that also posed novel legal and policy questions.
“We lived through Napster,” Blackburn said. “That was something that really cost a lot of artists a lot of money.”
Though he voiced support for compensation for artists, Altman did not get into specifics, saying that many industry stakeholders had “different opinions” on how creators should be paid. When Blackburn asked him if he thought the government should create an organization similar to SoundExchange – the group that collects certain blanket royalties for streaming – Altman said he wasn’t familiar with it.
“You’ve got your team behind you,” Blackburn said. “Get back to me on that.”
In 1994, at the dawn of the internet era, Rolling Stone asked Steve Jobs if he still had faith in technology. “It’s not a faith in technology,” he responded. “It’s faith in people.”
Today, at the dawn of the artificial intelligence era, we put our faith in people too.
It’s hard to think of an issue that has exploded onto the public scene with the furor of the debate over AI, which went from obscure technology journals to national morning shows practically overnight. This week, Congress is convening the first two of what will surely be many hearings on the issue, including one with OpenAI CEO Sam Altman and another with musician, voice actor and SAG-AFTRA National Board member Dan Navarro.
As members of the global Human Artistry Campaign, made up of more than 100 organizations that represent a united, worldwide coalition of the creative arts, we welcome this open and active debate. It’s gratifying to see policymakers, industry, and our own creative community asking tough questions up front. It’s a lot easier to chart a course in advance than to play catch up from afterward.
We don’t have long to get this right, either. The internet is already awash in unlicensed and unethical “style” and “soundalike” tools that rip off the writing, voice, likeness and style of professional artists and songwriters without authorization or permission. Powerful new engines like OpenAI’s ChatGPT and Jukebox, Google’s MusicLM and Microsoft’s AI-powered Bing have been trained on vast troves of musical compositions, lyrics, and sound recordings — as well as every other type of data and information available on the internet — without even the most basic transparency or disclosure, let alone consent from the creators whose work is being used. Songwriters, recording artists, and musicians today are literally being forced to compete against AI programs trained on copies of their own compositions and recordings.
RIAA Chairman/CEO Mitch Glazier
Othello Banaci
We strongly support AI that can be used to enhance art and stretch the potential of human creativity even further. Technology has always pushed art forward, and AI will be no different.
At the same time, however, human artistry must and will always remain at the core of genuine creation. The basis of creative expression is the sharing of lived experiences — an artist-to-audience/audience-to-artist connection that forms our culture and identity.
Without a rich supply of human-created works, there would be nothing on which to train AI in the first place. And if we don’t lay down a policy foundation now that respects, values and compensates the unique genius of human creators, we will end up in a cultural cul-de-sac, feeding AI-generated works back into the engines that produced them in a costly and ultimately empty race to the artistic bottom.
That policy foundation must start with the core value of consent. Use of copyrighted works to train or develop AI must be subject to free-market licensing and authorization from all rights holders. Creators and copyright owners must retain exclusive control over the ways their work is used. The moral invasion of AI engines that steal the core of a professional performer’s identity — the product of a lifetime’s hard work and dedication — without permission or pay cannot be tolerated.
David Israelite
Courtesy of NMPA
This will require AI developers to ensure copyrighted training inputs are approved and licensed, including those used by pre-trained AIs they employ. It means they need to keep thorough and transparent records of the creative works and likenesses used to train AI systems and how they were exploited. These obligations are nothing new, though — anyone who uses another creator’s work or a professional’s voice, image or likeness must already ensure they have the necessary rights and maintain the records to prove it.
Congress is right to bring in AI developers like Sam Altman to hear the technology community’s vision for the future of AI and explore the safeguards and guardrails the industry is relying on today. The issues around the rapid deployment of novel AI capabilities are numerous and profound: data privacy, deepfakes, bias and misinformation in training sets, job displacement and national security.
Creators will be watching and listening closely for concrete, meaningful commitments to the core principles of permission and fair market licensing that are necessary to sustain songwriters and recording artists and drive innovation.
We have already seen some of what AI can do. Now it falls to us to insist that it be done in ethical and lawful ways. Nothing short of our culture — and, over time, our very humanity — is at stake.
David Israelite is the President & CEO of the National Music Publishers’ Association. NMPA is the trade association representing American music publishers and their songwriting partners.
Mitch Glazier is chairman/CEO of the RIAA, the trade organization that supports and promotes the creative and financial vitality of the major recorded-music companies.
As a musician, educator, and author, I’ve spent the last few years examining AI’s challenges to the music ecosystem. But recently, after a comical misunderstanding on a U.S. podcast, I ended up playing devil’s advocate for the AI side of the AI and music equation. The experience was thought-provoking as I took on the role of an accidental AI evangelist, and I started to refocus on the question of, “Why are we fighting for ethical use of AI in music in the first place? What are the benefits, and are they worth the time and effort?”
As we hurtle from the now-quaint AI chatbot ChatGPT, to the expected text-to-video and text-to-music capabilities of GPT 5 (rumoured to drop in December), to Microsoft’s acknowledgment that AGI is feasible (artificial general intelligence, or a sentient AI being, or Skynet to be alarmist), to viral AI-generated hits with vocals in the style of Drake & The Weeknd and Bad Bunny & Rihanna, it can be easy to focus on the doom and gloom of AI. However, doing so does us a disservice, as it shifts the conversation from “how do we harness AI’s benefits ethically” to “how do we stop AI from destroying the world?” There are many, with a cheeky chappie nod to Ian Dury, “Reasons to Be Cheerful” (or at least not fearful) about music and AI. Here are nine reasons why we need to embrace it with guardrails, rather than throwing the baby out with the bathwater.
Fun: Yes, damn the ethics – temporarily. Generative AI technologies, which ingest content and then create new content based on those inputs, are incredibly fun. They tease and capture our curiosity, drawing us in. We might tell our employers that we use text-to-image services like DALL-E and Stable Diffusion or chat and search bots like ChatGPT and Jasper to optimise workflow to stay ahead of the technological curve, but they are also so seductively entertaining. Elementary AI prohibition won’t work; our solutions must be at least as stimulating as our challenges.
Time-saving: According to a survey by Water & Music, this is what music makers desire most. Many musicians spend countless hours managing social media accounts, wishing they could focus on music-making instead. AI solutions promise to grant that wish, allowing them to auto-generate text for social posts and announcements, and providing inspiration and potential starting points for new tracks, giving them the gift of time and helping them write, record, and release music more quickly. Let’s use automation wisely to free up musicians for their art and their economics.
Education: Despite dwindling funds for music education, technology offers new ways to make music accessible to everyone. Affordable AI tools can help students break through privileged barriers, providing access to personalised learning. I work with Ableton, which makes a variety of music production hardware and software. Successful initiatives such as the Ableton Push 1 campaign, which provided discounts to those who traded in their Push 1 midi controller and then refurbished and provided them for free to schools that needed them, demonstrate how digital tools can empower the economically marginalised, enable them to explore new musical styles and techniques, and nurture their passion for music.
Imperfect charm: AI’s imperfections and quirks make it endearing and relatable. AI’s unpredictable nature can lead to happy accidents and repurposing technologies for new musical uses. The fact that LMMs (large language models), which analyze huge swaths of text and then generate new text based on the patterns it learns, can be flawed retains a touch of human magic and humour in our interactions with them. Let’s enjoy this fleeting VHS fuzziness before it’s gone.
Affordable: Setting aside the environmental costs for a moment, AI has become exponentially accessible. AI allows creators to produce incredible results with basic tools. Last July, I purchased an expensive GPU-filled MacBook with dreams of making mind-blowing AI musical creations, but by September, I was doing just that using only my old phone’s browser. This so-called “democratisation” of music production can level the playing field for musicians worldwide, allowing more people to pursue their passion. Can we get it to increase their income too?
Tech Stacking: Experimenting with new combinations of generative AI APIs (application programming interfaces) opens up a world of DIY creativity. APIs are essentially pre-made functionality that developers can slot into their code easily, allowing them to focus on innovation rather than spending their time creating generative AI applications from scratch. This collision of technologies can encourage collaboration between musicians and developers, fostering a dynamic and innovative environment that crucially must be aligned with new licensing and rights payment models.
Elevated Chatter: As AI becomes more prevalent, the quality of conversations surrounding it has improved. People are starting to debate the legality of using copyrighted material to train AI, particularly in the music world, with a variety of strong arguments being made on behalf of human creators. In my research, I tried to address the complexities of AI in the music ecosystem, and now, I find these discussions happening everywhere, from John Oliver to barber shops. This elevated discourse can help us as a global industry make informed and necessarily swift decisions about AI’s role in our lives, better integrating its reasonably cheerful benefits and not being overwhelmed by its many chilling implications.
Inspiring the next generation: Introducing AI to young minds can be inspiring and terrifying. In my undergraduate module at Windmill Studios Dublin, I tasked students with inventing new music IP using existing cutting-edge technologies, with one rule of thumb: they could use a hologram but not bring someone back from the dead. Initially, I felt terrible about presenting such a potentially dystopian vision to young minds. But what happened next amazed me: all 40-odd students (from two classes) came up with outstanding commercial ideas. Their creativity and enthusiasm reminded me that the adage, “no one knows anything,” holds as true for music as it ever did.
Time to adapt: Perhaps the biggest reason to be cheerful at the moment, we still have enough time to address AI’s challenges. As Microsoft announces the “first sparks of AGI” and we face a saturated streaming market, we must work quickly together to ensure an equitable future for music makers. In my book, “Artificial Intelligence and Music Ecosystem,” my fellow contributors and I delve into the pressing issues surrounding AI, and now, more than ever, we need to take action to steer the course of music’s evolution. As AI continues to develop, it’s crucial for musicians, industry professionals, and policymakers to engage in open dialogue, collaborating to create a sustainable and equitable music ecosystem. Otherwise, what looks like Drake and sounds like Drake may not actually be Drake in the future.
The Human Artistry Campaign is the first step in this direction and interlinks with the concerns of MIT’s Max Tegmark’s “Pause Giant AI Experiments Open Letter” (which currently has 27,000+ signatures and called for a pause in the development of AI to give governments the chance to regulate it) for a big-vision picture. As we work together, we can ensure that AI serves as a tool for artistic growth and sustainability. But we must move fast and nurture things as AI’s growth is exponential. Do you remember back when ChatGPT was on the cutting edge of AI? Well, that was only five months ago, and we’re now talking about the possibility of sentience. That’s why I am working with some of the biggest companies in music and big tech to create AI:OK, a system to identify ethical stakeholders and help to create an equitable AI music ecosystem. If you are human and would like to help shape this future, it might be the most significant reason to be cheerful of all.
Dr. Martin Clancy, PhD is a musician, author, artist manager and AI expert. He is a founding member of Irish rock band In Tua Nua, chairs the IEEE Global AI Ethics Arts Committee, serves as an IRC Research Fellow at Trinity College Dublin, and authored the book Artificial Intelligence and Music Ecosystem. He also manages Irish singer Jack Lukeman and is a leading voice in advocating ethical use of AI in music.
What if we had the power to bring back the dead? As far as recordings are concerned, we might be getting pretty close.
The viral success of the so-called “Fake Drake” track “Heart on My Sleeve,” which apparently employed AI technology to create realistic renderings of vocals from Drake and The Weeknd without their knowledge, has raised the possibility that perhaps any voice can now be imitated by AI, even artists who died decades ago.
Last week, producer Timbaland did just that. “I always wanted to work with BIG, and I never got a chance to. Until today…” he said in an Instagram Reel, pressing play on an unreleased song clip that sounds like Notorious BIG, rapping on top of a Timbaland beat, despite the fact that the rapper was murdered in a drive-by shooting 25 years prior. (A representative for Timbaland did not respond to Billboard’s request for comment. A representative for Notorious BIG’s estate declined to comment).
But this is not the first time a deceased stars’ voice has been resurrected with AI. The HYBE-owned AI voice synthesis company Supertone recreated the voice of late-South Korean folk artist Kim Kwang-seok last year, and in November, Tencent’s Lingyin Engine made headlines for developing “synthetic voices in memory of legendary artists,” like Teresa Teng and Anita Mui. To see more even examples of this technology applied to late American singers, take a few minutes on TikTok, searching phrases like “Kurt Cobain AI cover” or “XXXTentacion AI voice.”
Some artists – like Grimes and Holly Herndon – have embraced the idea of this vocal recreation technology, finding innovative ways to grant fans access to their voices while maintaining some control through their own AI models, but other artists are showing signs that they will resist this, fearing that the technology could lead to confusion over which songs they actually recorded. There is also fear that fans will put words into artists’ mouths, making them voice phrases and opinions that they would never say IRL. Even Grimes admitted on Twitter there is the possibility that people will use her voice to say “rly rly toxic lyrics” or “nazi stuff” – and said she’d take those songs down.
In the case of artists like Notorious BIG or Kurt Cobain, who both passed away when the internet was still something you had to dial-up, it’s impossible to know where they might stand on this next-gen technology. Still, their voices are being resurrected through AI, and it seems these vocals are getting more realistic by the day.
It calls to mind the uncanny valley nature of the Tupac hologram which debuted at Coachella in 2012, or even the proliferation of posthumous albums in more recent years, which are especially common to see from artists who passed away suddenly at a young age, like Juice WRLD, Lil Peep, and Mac Miller.
Tyler, the Creator has voiced what many others have felt about the posthumous album trend. At an April 26 concert in Los Angeles, he noted that he’s written it into his will that he does not want any unreleased music put out after his death. “That’s f-cking gross,” he said. “Like, half-ass ideas and some random feature on it…like no.” It remains unclear if Tyler’s dying wishes would be honored when that time comes, however. Labels often own every song recorded during the term of their contract with an artist, so there is financial incentive for labels to release those unheard records.
Some who look at this optimistically liken the ability to render an artists’ voice onto a cover or original track as an emerging, novel form of fan engagement, similar to remixing, sampling, or even writing fan fiction. Similar to where this new technology seems to be headed, remixes and samples also both started as unsanctioned creations. Those reworkings were often less about making songs that would go toe-to-toe with the original artists’ catalog on the Billboard charts than it was about creativity and playfulness. Of course, there were plenty of legal issues that came along with the emergence of both remixing and sampling.
The legality of bringing artists’ voices back from the grave specifically is also still somewhat unclear. A celebrity’s voice may be covered by “right of publicity” laws which can protect them from having their voices commercially exploited without authorization. However, publicity rights post-mortem can be limited. “There’s no federal rights of publicity statute, just a hodgepodge of different state laws,” says Josh Love, partner at Reed Smith. He explains that depending on where the artist was domiciled at the time of their death, their estate may not possess any rights of publicity, but in states like California, there can be strong protection after death.
Another potential safeguard is the Lanham Act – which prohibits the use of any symbol or device that is likely to deceive consumers about the association, sponsorship, or approval of goods and services — though it may be less of a potent argument post-mortem. But most cases in which rights of publicity or the Lanham Act were used to protect a musician’s voice – like Tom Waits v. Frito Lay and Bette Midler v. Ford – were clear examples of voices being appropriated for commercial use. Creative works, like songs, are much more likely to be deemed a protected form of free speech.
Some believe this could be a particularly interesting new path for reviving older catalogs, especially when the artist is not alive to take part in any more promotion, for the estates and rights holders who control the artists’ likeness. As Zach Katz, president and COO of FaZe Clan and former president of BMG US, put it in a recent press release for voice mapping service Covers.ai: “AI will open a new, great opportunity for more legacy artists and catalogs to have their ‘Kate Bush’ or “Fleetwood Mac’ moment,” he said. “We are living in a remix culture and the whole fan-music movement is overdue to arrive in the industry.”
Though Covers.ai, created by start-up MAYK, was only just released to the public today, May 10, the company announced that it already amassed over 100,000 sign ups for the service leading up to its launch, proving that there is a strong appetite for this technology. With Covers.ai, users can upload original songs and map someone else’s voice on top of it, and the company says it is working to partner with the music business to license and pay for these voices. Its co-founder and CEO, Stefan Heinrich, says this idea is especially popular so far with Gen Z and Gen Alpha, “the product we’re building here is really made for the next generation, the one coming up.”
Between Supertone, Lingyin Engine, Covers.ai, and others competitors like Uberduck coming into the marketplace, it seems the popularization of these AI voice synthesizers is inevitable (albeit legally uncertain) but playing with dead artists’ voices adds another layer of moral complexity to the discussion: is this more akin to paying respects or grave robbing?
MAYK’s artificial intelligence-powered voice recreation tool officially launched to all users today (May 10).
Covers.ai lets users upload their own original songs and then try on other AI-voices on top of it, including the voices of Billboard Hot 100-charting talent. According to a company press release, Covers.ai’s tool topped over 100,000 sign-ups prior to its launch.
Its founder and CEO, Stefan Heinrich — an entrepreneur who previously worked in high-ranking positions for Cameo, TikTok, Musical.ly and YouTube — explains that, for now, most of the models available for users to work with are “community models.”
“This is open source,” he explains. “There are users that make these models with various celebrity voices out in the wild, and those can be uploaded and marked at ‘community models’ on our site. At the same time, we are working with artist teams to license the voices of specific talent so we can find a way to compensate them for their official use.”
Eventually, Heinrich says he also hopes to find a way to license song catalogs from rights holders so that users can mix and match tracks with various artists’ voices they find on the site. Through these licensing agreements, he hopes to find a way to create a new revenue stream for talent, but to date, these licenses have not yet been finalized.
MAYK is backed by notable music investors including Zach Katz (president/COO of FaZe Clan, former president of BMG US), Matt Pincus (co-founder and CEO of MUSIC), Jon Vlassopulos (CEO of Napster, former global head of music at Roblox), Mohnish Sani (principle, content acquisition, Amazon Music) and more.
The launch arrives as conversations around AI and vocal deepfakes are at a fever pitch. Just last month, an unknown songwriter called Ghostwriter went viral for creating a song called “Heart on My Sleeve” using supposed AI-renderings of Drake and The Weeknd’s voices without their knowledge. Soon after, Grimes responded to the news by launching her own AI voice model to let users freely use her voice to create music.
In just a few minutes of searching, it’s apparent that TikTok is already flooded with songs with AI-vocals, whether they are original songs employing the voices of famous talent, like “Heart on My Sleeve,” or mashing up one well-known song with the voice of a different artist.
This AI vocal technology raises legal questions, however.
Mimicking vocals may be a violation an artist’s so-called right of publicity – the legal right to control how your individual identity is commercially exploited by others. Past landmark cases — like Tom Waits v. Frito Lay and Bette Midler v. Ford Motor Company — have established that soundalikes of famous voices cannot be employed without their consent to sell products, but the precedent is less clear when it comes to creative expressions like songs, which are much more likely to be deemed a protected form of free speech.
Heinrich hopes that Covers.ai can help “democratize creativity” and make it far more “playful” in an effort to get music fans from the lean-back forms of music discovery, like listening to radio or a pre-programmed editorial playlist, to a more engaged, interactive experience. “I think what music is really changing right now,” he says, noting that Covers.ai’s earliest adopters are mostly Gen Z and Gen Alpha. “The product we’re building here is really made for the next generation, the one coming up.”