Welcome to the future: Spotify poaches AI music expert from Sony


On any given day, this is interesting news. In the context of what we’ve previously reported this week, this is fascinating news.

MBW has learned that Spotify has hired François Pachet – one of the world’s foremost experts on the application of Artificial Intelligence in the world of popular music.

Or in simpler – perhaps scarier? – terms: music written by computers.

Pachet is currently understood to be waiting out his contract as the Director of the Sony Computer Science Laboratory in Paris, where he leads the music research team. He has worked at Sony for 20 years.

In 2012, Pachet oversaw a project which created the first known pop songs composed with AI, as well as launching the first music label dedicated to the professional use of AI for music production.

Earlier this year, Pachet’s Sony team released two pop songs created using AI – Daddy’s Car, in the style of The Beatles, and The Ballad Of Mr Shadow, in the style of American songwriters such as Irving Berlin and Duke Ellington.

There’s been much debate in the past few days about MBW’s use of the term ‘fake artists’ to describe the secret pseudonyms of producers on Spotify, whose tracks have attracted hundreds of millions of plays on the platform.

We’d suggest that debate just kicked up a notch, wouldn’t you?

Pachet (pictured) has titanium credentials in the world of computer science and how it intersects with audio art.

In the four years leading up to 1997, he was Assistant Professor in Artificial Intelligence at UPMC.

His inventions have included ReflexiveLooper – a system that learns in real time the style of a musician and automatically generates accompaniments – as well as EDS, ‘the first audio feature generation system’, which is now fully owned by Sony Corp.

Pachet is also the co-creator of Flow Composer, the tech underpinning the two AI-penned pop music compositions released by Sony earlier this year.

Flow Composer has evolved to become Flow Machines at Sony, with a fully AI-written pop album expected for release before 2018.

As Sony wrote in a blog earlier this year “our Flow Machines software learns music styles from a huge database of songs… then, exploiting unique combinations of style transfer, optimization and interaction techniques, it can compose in any style”.

You’re waiting for the ‘2+2 = 4’ moment, aren’t you?

As MBW has revealed this week, Spotify has deliberately stuffed many of its playlists with music recorded by fake artists – aka. the anonymized aliases of select producers.

If, as many suspect, this move has a financial benefit to Spotify, what’s the next obvious step?

Could the company soon begin upstreaming AI-created music onto its hugely popular mood, genre and activity-based playlists?

Yes and no.

Pachet’s technology is certainly capable of spitting out reams of bespoke music for a ‘Peaceful Piano’ playlist, for example. Or ‘Ambient Chill’, or ‘Yoga & Meditation’ or ‘Sleep’, for that matter.

But even those bleeding-edge Flow Machine pop compositions – Daddy’s Car and The Ballad Of Mr Shadow – required a human being (the very talented Benoît Carré) to arrange and produce the material after Sony’s tech had written it.

When contacted by MBW today for confirmation of Pachet’s impending arrival at Spotify, the company declined to comment.

However, an MBW spy close to the firm did verify our story, before telling us that Pachet is understood to have been tasked with developing “creator tools” at Spotify.

These tools, said our source, could eventually “help composers to become more effective, and help more people become composers”.

That bit is quite exciting in a ‘here’s the future of the music business’ way.

But there’s another key point to be made, too.

Some MBW readers recently saw Pachet speak on a panel at the BUMA Music in Motion conference.

He was apparently very clear: AI-written music inherently shouldn’t require any royalties to be paid to copyright holders after it is published.

Brace yourselves.Music Business Worldwide

Related Posts

  • Jeoff Harris

    Seriously, Tim, your claims are downright fallacious. Just because you think your thoughts are compelling doesn’t mean that logic is on your side. I will quote my post from the Facebook page:

    If producers are ‘covertly’ releasing material without any other online presence, they’re still receiving the revenue from the streams. There’s incentive for producers to do that if they have no interest in maintaining a personal following and just want the music to exist for streaming and / or sync-liscencing because…money.

    Correct me where I’m wrong, but producers have a prerogative to release music without promoting it. Insinuating that Spotify commissioned them for pay when there isn’t any evidence to support that is dishonest, or, at the very least, extremely careless journalism.

    Spotify shouldn’t have to take a PR hit due to Tim Ingham’s lack of critical thinking skills. Use your platform responsibly or people will stop caring about what you have to say.

    • Dan

      There is no good explanation for Spotify to behave like this unless they are commissioning new tracks specifically to pay lower royalty rates. There was no shortage of ambient music before this. Why waste money on new music here when the listener is probably sleeping? If Spotify really does just need to fill an artistic void then why not put out a call for submissions instead of secretly tapping a handful of Swedish guys? Why not explain that in their original response instead of denying everything?

      Why do you think Spotify just spent a bunch of perfectly good money on this AI scientist? Are corporations generally in the business of spending money for no reason, or do you think they are expecting a return on their investment?

      The endgame seems pretty clear to me, get AI to “compose” songs whenever possible and then pay some new Full Sail grad $32k a year to produce those computer compositions into something pleasant sounding. Then there’s no human actually writing the song who you have to pay royalties to. Sure, it won’t work for Rock or Hip Hop but they can save a lot of money here, and isn’t that the skill executives get paid so much for? We’re heading for a world in which there will be a financial incentive for the companies providing you music to remove human beings from the process of songwriting. A world in which every song is not just a work for hire, but a work for hire written by a machine!

      If it doesn’t bother you that automation by artificial intelligence is going to cause songwriters to lose jobs then why do you even work in this industry? Wouldn’t you be better off selling real estate or managing an Amazon warehouse or something?

      • Jeoff Harris

        Forgive me for the frankness, but I’m genuinely puzzled how you haven’t put this together.

        There is no reason to assume that Spotify is ‘involved’ with the creation of these tracks at all. Producers don’t have to promote music to post music on Spotify. They can invent an artist project out of thin air simply because they think they can make a couple bucks. That’s how it’s always been, sad for the fact that distribution is no longer an issue for these types.

        If their music gets some exposure, a music supervisor’s interest may be piqued and these producers could land a placement that could bring in serious money at very little cost to them. No promo necessary, especially if the music is good. You calling them ‘fake artists’ only serves to demonstrate how little you understand this revenue stream.

        And seriously, what evidence do you have to support the claim that Spotify is involved? Have you had correspondence with any of these ‘fake artists?’ Have you considered that they made a song, and just wanted to share it without spending time and energy building some arbitrary artist project from scratch? I would follow up with them before you make more accusations.

        Your intuitions about Spotify commissioning producers don’t add up to anything. There are approximately 50,000 new songs uploaded to Spotify daily, so, there is no need to inflate the catalog with extremely popular music. They could just upload thousands of less popular songs and spread the savings more covertly. Why are you not calling artists with 12,000 streams ‘fake?’ That argument would actually be stronger than yours, and it’s almost certainly not true.

        And where do you suggest the savings are even occurring? You’re implying that Spotify would pay producers thousands of dollars to produce music so they don’t have to pay them thousands of dollars in royalties later…think about that. They have exponentially more incentive to just not do anything and save all of the money. Sure, they have incentive to fill out their playlists, but there is just so much music out there that the savings would be negligible. Your model is not scalable, and of course Spotify has better ways to spend their resources.

        And I don’t know why they hired an AI expert. There are a plethora of reasons that could be a smart move. If AIs start writing better songs than humans, then great! Innovative tech doing what it’s supposed to do. But no matter what their reasons are, I’m sure they are much more sophisticated than your clumsy analysis.

        Spotify has too much on the line right now to risk their reputation so near-sightedly.

        • Dan

          You’re right, this is very embarrassing for Spotify and comes at an inopportune time for them. All the more reason for Spotify to have one of these “fake artists” state on record that they are paid the full royalty rate for these tracks. That would be very easy for Spotify to do and would make this problem go away.

          Can you think of a reason why they aren’t doing that and are instead denying everything with carefully worded statements that don’t ever mention mechanical royalties?

          • Jeoff Harris

            Yes, I can. They don’t want to reward someone seeking attention by giving extra credence to his ludicrous claims. It only exacerbates the problem. Their first statement wasn’t cryptic; it was perfectly thorough. You just weren’t satisfied with the way they responded, I assume, because you have no faster way to gain readership than to make more accusations.

            I doubt you have malevolent intentions with this, but this whole thing is seriously not thought through. Respect your craft and give this a rest, man. Your arguments are speculative and unsound. You don’t want to be known as the guy who pulls this kind of stunt.

          • Jeoff Harris

            And for some reason I thought I was talking to Tim Ingham. Forgive the aggression as it should be directed his way, not yours.

          • GB

            To a certain extent i agree, there’s a number of claims from MBW’s side but other than a denial from Spotify, very little else in the way of counter-argument. We now know the identities of the producers of a handful of those songs and while it’s their own prerogative as to whether or not they confirm or deny they were indeed hired by Spotify to write and produce those songs, to then be released by Spotify – there remains a lot of unanswered questions.

            Personally for me i find it hard to get passed the obvious success rate of these songs. It is no easy task to receive the playlist support and amass the streams those tracks in question have. Usually in order to keep the playlist additions going you have to show external factors such as live support, radio support and press support for the artist or song. Albeit this can usually be more so attributed to contemporary pop music as opposed to more ambient and bespoke works.

            However, the beauty of Spotify is that it can also act in a vacuum, regardless of those aforementioned external factors Spotify can see how a song is working on their own platform and continue to support it. These benchmarks are usually based around how many times the track is being repeat listened to per user, how many people save or share the track and how infrequent the track is skipped. If the track performs well against these benchmarks then it is re-added to the same playlist and also similar genre / mood themed playlists are sought for it’s inclusion.

            This doesn’t happen as easy as you might think so in order for those tracks to be reaching the millions of streams they have, it certainly does raise an eyebrow. To be fair, the producers in question, Andreas Romdhane and Josef Svedlund, do have a strong track record of working on hits, the fact these songs are doing so well could be attributed to the fact they’re very good at what they do. But it’s all very coincidental that songs under different artist names all attributed to one source have done so well on Spotify and begs the question that if they did legitimately release one or two songs under different names, why after they had proven success did they keep releasing music under different names and not simply continue with one sole artist name and have all streams come back to a central point and actually build a credible if not anonymous artist?

          • Jeoff Harris

            Wow. That was an extremely thorough response. I wish you’d written this article. lol

            I’ll have to think a little more about your comments. I’ll admit my knee-jerk bias is to act protectively toward a company who is doing so much amazing work for the music industry. That said, I think we should be doing more thorough research before we write about it. Sites like MBW need some accountability and a little more journalistic rigor, in my opinion. Your framing made way more sense to me than the original article.

          • iLexx

            Just as somebody who has been reading MBW articles on this issue aswell, I have to say that I got everything Bastow was saying from Tim’s article. I found it very interesting that these points were lost on you until Bastow literally spelled it out for you.

            I think MBW has done and continues to do a phenomenal job as to pertains to giving us Music Business news and one of the only major sites in the music industry that actually partakes in journalism on this level. The very reason Tim’s articles have gotten the traction they have gotten is because of the type of journalism in which they are conducting here. It feels refreshing because there isn’t enough of it happening else where.

            I know I’m just jumping in here and sharing my perspective but by all accounts, Tim isn’t the problem here, infact he and MBW deserve an applaud for what they have been doing.

          • Jeoff Harris

            Well, his most recent article was certainly a step up in the spelling-things-out department. From my vantage point, we were living in the land of pure conjecture until this most recent post, and Tim’s initial inferences seemed to neglect the incentive of producers to do exactly what the Epidemic people are doing…but I must admit that he’s got more to say than I was giving him credit for in my previous comments and I’m leaning away from my post a bit now.

            Probably should practice being a little less aggressive before I’ve put in as much effort as he has into investigating.

            Sorry, Tim! 😭

      • vintermann

        Spotify is an AI company, just like Google and Facebook and Microsoft are AI companies these days. In my opinion, both as a listener and a machine learning hobbyist, they have the best music recommendation system in production today. Especially that release radar is so good, means they’ve gone a long way to solve the “cold start problem” of figuring out who would be interested in a new track there’s not much listening data on yet.

        But if you can do that, you can probably also get a good idea of what music people would like, which they’re not getting much of. Like for instance simple, calm piano pieces.

        So they go out and commission more of that, possibly sharing examples of the sort of music the algorithm thinks is undersupplied. It’s no surprise they do it quietly at first. They’re probably experimenting, giving it to different producers and composers to see who does best with it. A pair of pop producers, a classically educated composer … and maybe a machine learning expert or two.

        If AIs can learn to compose indistinguishable from a human, it won’t happen overnight. There will certainly be trial balloons – and it’s also certainly best not to advertise them.

        • Dan

          An AI DJ is one thing, an AI songwriter is another thing entirely.

          I’d prefer a human in both roles.

  • Josh Rosenthal

    What about the fact that these songs are the worst dogshit ever ?

    • Pablo Ferrigno

      what do you think they will sound like in 20 years?

  • Johnny J Jems

    Songs Suck horrible

    • Pablo Ferrigno

      missing the point. 20 years ago the internet sucked horrible.

      • Johnny J Jems

        I understand Pablo but im still angry for what happened to me sorry

  • Johnny J Jems

    You can’t replace a real talented Songwriters and Artists like myself and Sony and BMG stole from me in the past Bigtime, they owe me, Staff writers are a bunch of thieves

  • Every Musicoin

    thought it’s about Musicoin, which $MUSIC was defined.

  • iLeonD

    The craft of songwriting across most genres these days are at an all time low. If AI can create a new creative fire in people, I’m all for it. The issue is that from what we can see, technology makes people lazier. When Pachet and his crew improve the technology in which these songs are created over the next decade, that’s where the fun begins.

    • Edwin Joassart

      Tech doesn’t necessary make ppl lazier. What it does is level up the play field. With better tech you’ll have less effort to do to get to the same result, so is everyone else. So if you want to crush it, you’ll need to go the extra miles…

  • Okay, they both suck!

  • Just listening to this….you will never be able to take the human element and emotional insight out of songs. Computers just cannot duplicate that.

  • M.N Calristein

    Bleck. Not interested.