Hacker News Reader

Comment by Animats | Hacker News Reader

Musicians getting rich was a brief phenomenon. Historically, musicians ranked below bartenders. Making phonograph records was a process with expensive tooling, manufacturing, and distribution costs. Radio was an industry with limited airtime and a small number of stations. These factors created an industry with few songs produced in large numbers, resulting in the "rock star" industry.

That's over. Through heavy promotion, there are still stars, but they have a smaller fraction of overall plays and the margins are lower.

And we haven't even had much AI-generated music yet. That has to be coming soon.

When a Vocaloid learns your job, what are you going to do?

View on HN

Ha, you had me til the AI part. Personally I figure the musicians will have the last jobs, along with designers. Really artists generally speaking will be the most difficult AI, no?

Modern pop music follows strict patterns. I have little doubt that much of what's playing in the top 40 can be synthesized or already is :P

All that you need then is a human avatar that the kids can connect to and take as their role model.

The only problem though with using a strict pattern system is that it will, I believe, more than likely miss the nuances which a brilliant songwriter can weave into a structure. The song might be 90% as good, which makes a lot of difference I think. Max Martin is an absolute genius in this regard, and his years of success reflect that. He's also a musician, though he is truly making his living as a songwriter/producer.

Here's a great break-down of "Shake It Off" from a songwriting perspective, which kind of reveals the simplicity is deceptive in some ways:

https://www.youtube.com/watch?v=F5EWn6s65o8

The other day I was listening to some random pop song (maybe Demi Lovato, not sure) and it was sounding eerily Max Martinesque. I go to look it up, and there he was in the credits. There is just something there that stands out; I don't know if it can be trained into an AI, but my money is on no.

One thing you might be subconsciously catching on to is that Max Martin is committed to "balanced lines" when it comes to vocals. It's obious in "Hit Me Baby One More Time" by Britney Spears and his fingerprints on "Teenage Dream" by Katy Perry. Here's a great review and comment piece featuring 30 songs by Max Martin, in order to better study the guy's techniques which seem to work so well:

http://www.stereogum.com/1788251/30-essential-max-martin-son...

Also, I forgot to mention that the weird thing about music is that it follows trends just long enough to want a new one. Thus, whatever formula might win out for a period of time would likely be usurped by a variation or counter-intuitive trend. A funny thought in this line is that a hit-making, successful "Hair Metal Pop" AI would be going great until the "Grunge Rock Pop" AI started getting traction. Humans are fickle!

It does and it doesn't. The modern pop music that's most formulaic seems to be there mainly because then the managers can concentrate on promotions without worrying about what used to be known as A&R.

As an Olde Phart(tm), what DJs do looks very much to me like "a human avatar" and aggregator.

"All that you need then is a human avatar that the kids can connect to and take as their role model."

Already in production. See Hatsune Miku in live concerts.[1][2] Not only is the performer an avatar, the singing is synthesized by Vocaloid software, which you can buy.[3]

[1] https://www.youtube.com/watch?v=dhYaX01NOfA [2] https://www.youtube.com/watch?v=XyTmXLWCazw&nohtml5=False [3] http://www.vocaloid.com/en/

Yup, just need to "synthesize" words and melodies as the gp said and you're set.

"To create a song, the user must input the melody and lyrics".

I imagine the goal posts will for ever be moving when it comes to defining what art is, but ultimately optimizing for human emotional sentiment does not seem that difficult a problem. Art is mostly just a combination of past works optimized for the modern audience. Seems like the perfect problem for AI. We often judge a future AI in terms of how well it could replicate current things but the real question is what happens after it perfects that replication? AlphaGo alone shows that the AI is likely to discover new styles that we failed to examine before, why can't the same happen with Art?

Animats, kmnc...

Do either of you write or play any music? What supports your arguments other than a few marketing op-ed pieces put out by Google in the last couple of year?

You do realize that Google's self-driving cars and other AI research programs are mainly exercises in show business themselves, right? Well kudos to the people running public relations down in Mountain View, they've certainly convinced their target demographic that Google is the place to be for AI research!

> AlphaGo alone shows that the AI is likely to discover new styles that we failed to examine before, why can't the same happen with Art?

Because Go is a game with set of rules and clearly defined winner. Art has no rules and no winners and no losers. You can only program a computer, be it a neural network or some other machine learning algorithm, to deal with a finite set of inputs. Human oversight is still needed to set up and do the initial training of unsupervised learning algorithms and obviously needed during the entire process of supervised learning.

The number of variables that go in to a music production are pretty astonishing. I'm guessing you have pretty much no experience with writing lyrics, melodies, counter melodies, arranging a song, playing the drums, playing the bass, playing any accompaniment, singing, recording, mixing, and then mastering. There's probably around a billion subtle variables that go in to what ends up being 4 minutes of a stereo recording. I would imagine it would take hundreds of thousands of man-hours just to train the initial complex neural network that could get anything that would spit something out that remotely sounded like the Beatle's Yesterday. Perhaps it wouldn't take as much effort to reproduce some plastic pop star, but you know what? People don't really like that shit. Kids do. For like 3 weeks. Then the plastic melts.

You know what, this is basically pointless. You've made your mind up to remain ignorant about music. That's your prerogative. I'm safe in my knowledge that some genious in the Bay Area isn't going to be taking my job playing covers in a bar room any time soon. What, do you think a bunch of people sitting around in a dive bar want to watch a laptop sing them country music? But who knows, judging by how tasteless and uncultured most techies are, you guys would probably be fine with listening to the world's worst poetry, the world's worst instrumental lead, and the tackiest, lamest, emotionless expression of the inhuman condition a machine could possibly imagine.

Why do AI zealots hate humanity so much? Are you ashamed of our past? Do you treat your parents and friends and family as ultimately replaceable by a machine? Art is much closer to the home and heart than you're realizing. I mean, it's you whose gonna end up driving real love away from your own life. I'm surrounded by a bunch of lovely morons who are fine with their shameful existence and are fine from spending the rest of their years being far from perfect and living side by side with misery and sin. You'd better get acquainted with reality at some point. You're still just a bag of flesh and bones with a set expiration date!

Here's how I'm wasting away my remaining years: https://www.facebook.com/photo.php?fbid=10153729538523439

Rest in peace, Merle. It's a shame he never got around to learning about and writing about techies... although if you use your imagination his "Here In Frisco" is pretty damned close:

It's four a.m. in New York City, three a.m. in Dallas

The night is still early here in Frisco

Market street's still going, the same old shows are showing

And I'm still all alone here in Frisco

They say it's raining in Chicago, but it's cold and clear in Denver

Been windy all night long here in Frisco

Trolley cars are clinging, the big Bay Town's swinging

And I'm still all alone here in Frisco

The way I feel tonight, I won't be staying long

But when I leave, I'll leave my heart just like in the famous song

Trolley cars are clinging, the big Bay Town's swinging

And I'm still all alone here in Frisco

---

Yup, and I'm still all alone here in Frisco, sitting around with my Artificial Love Machine and my Oculus VR wired directly in to the Google-Apple-Amazon mothership circling the earth... full-stack, full vertical integration, corporate assisted lifestyle... pumps me full of vitamins and movies made by computers.

That this doesn't sound like the living embodiment of hell on earth, I don't know what the fuck does...

It's nine o'clock on a Saturday

The regular crowd shuffles in

There's an old man sitting next to me

Making love to his tonic and gin

And the waitress is practicing politics

As the businessmen slowly get stoned

Yes they're sharing a drink they call loneliness

But it's better than drinking alone

Sing us a song you're the piano man

Sing us a song tonight

Well we're all in the mood for a melody

And you've got us feeling alright

Generative music already is a tool / method in some artists' portfolio (particularly in ambient or experimental circles). However, it doesn't tend to be the only approach.

AI could perhaps work soon for certain formulaic music (like say bubblegum pop) as well as some forms of synth music (I easily can see generative minimal techno for instance). I will say that people don't tend to like overly formulaic music though; they tend to like the twists, turns, and "personality" that people can bring to art. Even in pop, formulaic pieces are quickly forgotten. But it's probably possible to write an algorithm now that does this sort of stuff really well.

Beyond pop formula, I think AI advocates sometimes underestimate the difficulty. There are no real "rules" in music. The music of the world is very diverse, the "rules" even in pop / folk vary from culture to culture.

Many styles of music are way beyond AI at this point IMHO. They are either much more technically complex (Western classical) or rely on improvisational skills that I would think would be difficult to parametize (jazz, Indian classical) Heck, even with something like say blues music, which pretty much uses the same template for most songs (well, a similar rhythm and similar chord changes at least), you've still got the soloists and the singing. Vocaloid ain't got nothing on Muddy Waters or BB King etc. :)

Also I don't think people in general cares that much about the music as much as the idea of the identity behind the music. For AI music to be popular in a context where the an actual artist has to be emulated you either have to fool people into believing it's a real human behind it, or antopromoprohize the ai sufficiently to be promoted like a human.

Music has meaning, it's a tool for communicating very complex ideas.

I don't think anthropomorphising something that generates pleasant sounds is going to get you anywhere, by the time you get anywhere interesting, you've become the artist.

Here[1] is an interview with Alex Ebert where he discusses a non lyrical song teaching him about death.

[1] http://youtu.be/gVK51JvGzzk

These guys: https://www.jukedeck.com/

There will come a point when one of the easiest ways to make money for a musician - sync licensing - is overtaken by computer generated tunes.

That's really interesting. They talk[1] about being able to automate it in the future but it seems like there may be a lot of human involvement still.

That article goes further, that their goal is to be Instagram like, helping the unlearned make more professional sounding things.

That doesn't seem more revolutionary than the Auto-Tune.

[1] http://www.globaluniversityventuring.com/article.php/5136/ju...

Exactly. Music isn't algorithms, it's communication, it's emotional connection. You might be able to fake even that, but it's not going to be very good. The best and most potent music is stuff that connects to human experience. You're not going to get something like Adele's "Hello" or Pink Floyd's "The Wall" out of a computer until you get computers that are sentient and share the same experiences as humans. At which point they'll be people too anyway.

I don't think the future is "totally" AI generated music. Rather, I think that AI will provide tools to enable humans to make music more easily. For instance, taking a basic riff and "jazzing" it up, maybe suggesting a drumline, providing lyrical "autocomplete" suggestions based on the song's subject and rhyme structure, etc.

If you're interested, such a tool was recently debuted aimed at guitarists:

http://digitech.com/en-US/products/trio

"Making music" is a very small part of the "musician" package; a lot of it is pure ethos. Difficult to identify with algorithmic output, even if you put a fancy brand on it.

In George Orwell's "nineteen eigjty-four", the Versificator machines cranked out music and cheap lit for the masses (proles). I wouldn't bet against Orwell.

Totally agree. Big music and film stars are a pretty new phenomenon and in the future they will probably be viewed as a short term fad. The big companies probably will keep making lots of money though.

You clearly have no clue about music or the music business.

Reply on news.ycombinator.com