The program changes minor chords in the song to major, so suddenly the chorus seems peppy and bouncy. And its version of Crazy Train fits far too well and will have you laughing out loud if you're a fan of the original:
I like how some people are decrying this online as the death of music, like Songsmith is going to change how music is made. (Oasis writing music with Garage Band to the contrary, most people still write their music by getting some musicians together and figuring out how stuff sounds.) I think it's really neat that it can do this! The program's working very well - I'm installing the beta today and we'll see how well it handles real-people voices - and people expecting it to magically create the memorable hooks and snares from the original are people who still miss that a machine can't generate truly great music, because the best music comes not from formula but from the human mind.
Holy crap, that is absolutely hilarious. The bluegrass polka version of Crazy Train! I gotta get out the banjo!
And of course people hail this as the death of music. It's a strange tic in our culture: Every time we do anything except play the same old music in the same old way, it's a sign of something dying. If you double the number of music listeners overnight, that's a harbinger of death. If you get fifty million Rock Band players analyzing the drum tracks of Rush tunes that are thirty years old, that's a harbinger of death. If you get people competing to find off-the-wall but briefly listenable orchestrations of classic tunes, that's a harbinger of death. One wonders what life is supposed to look like.
Many people also have no idea how actual composition works.
Not to mention that doing different forms of electronic accompaniment has been around for 40 years, and algorithmic composition for quite a while as well.
Only listened to the first part, but I think he might have a hit on his hands. Brilliant lyrics!
The last hit I witnessed in Germany (I rarely check) was something along the line "I count to three and take you home, come along we'll make a threesome" - people love that kind of lyrics...
'If you tried, if any Mind tried, could you impersonate my style?' the Chelgrian asked. 'Could you write a piece - a symphony, say - that would appear, to the critical appraiser, to be by me, and which, when I heard it, I'd imagine being proud to have written?'
The avatar frowned as it walked. It clasped its hands behind its back. It took a few more steps. 'Yes, I imagine that would be possible.' [...]
'So what,' the Chelgrian asked, 'is the point of me or anybody else writing a symphony, or anything else?' [...]
'Ziller, are you concerned that Minds - AIs, if you like - can create, or even just appear to create, original works of art?'
You can analyze an existing progression. You cannot predict the possible future progression.
You could analyze Please Please Me and determine how to generate similar Beatlesy sounds. You couldn't predict Help. If you had Help, you could make Help sounds but you couldn't figure out that Rubber Soul was on its way. The sonic style changed constantly and evolved incredibly.
Or imagine trying to copy Frank Zappa. You might replicate a single song, but it wouldn't be possible to predict the next songs in a given album. He evolved and shifted too rapidly to follow.
Why shouldn't an AI be able to listen to ALL songs that are uploaded to YouTube at any given time and from that synthesize something new? Real musician presumably don't do anything different - at least they don't start from scratch, they build on existing music and styles.
A lot of cutting-edge music doesn't come from the field of music. John Cage's work was influenced by literature. Brian Eno created ambient music by listening to music in places he went. An AI couldn't discover breaking-ground things like that. Not through synthesis.
The idea is that truly great music isn't music that takes an existing style and moves it. Great music is the music that creates something that isn't already there.
Why would a human, but not an AI, be able to do that?
Because AI operates according to structures that have been designed by human hands. The human mind is incomprehensibly more complex than any AI. So you could make AI that does innovate, but its innovations would still be far more predictable and routine than any that a human mind could come up with.
AI's cannot read literature?
John Cage would read Finnegans Wake, draw lines up and down pages, and derive poems from words formed by what his line would hit. A computer would never think to do that on its own. There's a fine line between genius and insanity, it is said - and computers can't be insane.
Because AI operates according to structures that have been designed by human hands. The human mind is incomprehensibly more complex than any AI. So you could make AI that does innovate, but its innovations would still be far more predictable and routine than any that a human mind could come up with.
Not necessarily. Using evolutionary approach, AI can come up with innovations on par (and possibly even beyond) to what humans could do:
The biggest issue with this contention is that no one knows how far AI can go. Furthermore, I posit that there are people who will never accept what AI does, viewing it as always lesser to human development. Someone (I don't remember who) quipped, "What is AI? Whatever a computer can't do".
It's not, in the sense that you're talking about. We don't operate through a vast system of yesses and nos. We're not just a huge binary cloud, in other words. And nobody's entirely certain just how we work.
Comparing us to computers is really ignoring most of the limitations of computers, and invites far to simplistic - and inaccurate - a comparison.
I think that "human as machine" has some interesting implications, but I think that by the time we make machines as complex and sophisticated as ourselves, they won't be machines any longer. If I see a machine that's organic and that has a superpowerful mind, I won't think it's a machine anymore.
It isn't possible for anyone who hasn't heard a particular Beatles or Zappa (rock, not classical) song before, to recognize it as a Beatles or Zappa song?
When I heard latter-day Beatles I didn't recognize it. Two years ago, all I knew by them were the oldies. I'd hear Let It Be or Across The Universe or Eleanor Rigby, and I'd ask what band it was playing. I think I had that revelation something like 10 times, and finally, after Why Don't We Do It In The Road?, I took the time to listen to their entire discography.
Zappa you can't tell. In the last 3 months I've listened to all 60 of his albums because he never repeats himself, not even song to song. Even on Freak Out!, his debut, you can't guess from the first song just how incredible the third song will be, or that from there it will move to doo wop or kid's songs. He does it all. If you've heard that album, you can't tell what Absolutely Free or We're Only In It For The Money will do. After that he does a postmodern doo wop album and moves on to jazz fusion and movie tracks, returning to pop rock just in time to release an album consisting of only guitar solos.
With Zappa, if he's not singing you can't tell it's him writing the songs. He's the single most diverse composer I've ever heard.
I've just made an impromptu collection of Zappa songs - http://omegaseye.com/Zappa.zip. His rock songs never sound the same, his pop doesn't sound like anything else. The closest he ever gets to generic is Fountain of Love, which has a surprising guitar line, a very complex choral arrangement, and quotes from Stravinsky's Rite of Spring at the end. I could give you more songs from any of these albums, or show you other albums of his, and unless I told you you would be unable to guess it was Zappa.
Interesting... so part of Obama's rhetorical skill is to start his sentences on a beat.
Songsmith looks like a mostly useless piece of software from a musical point of view... I can't imagine any real musicians using it. But from a content-creation-for-the-masses point of view, it's clearly a hit. I look forward to seeing more crazy shit put together with this software.
meter is an important aspect of public speaking. if you vary your meter at inappropriate times your speech doesn't come off as natural and flowing. OTOH if you maintain a rigid meter the whole time you sound robotic.
In it's current state songsmith could not kill modern music. It does not sound organic enough. It has no soul. If music created by humans using music sequencers has its quirks. Songsmith music sounds very formulaic and unimaginative and becomes boring after the first verse or so even if the tune is catchy.
To heck with soul -- now that I've listened to several of the tracks in the original article, the problem with Songsmith is that it's hard to reverse-engineer a rhythm that you're singing against. [1] It mangles the rhythms in Roxanne, in which Sting's vocals are designed to play around and against the timing of the rhythm section, and its treatment of the first few seconds of Intergalactic are comedy gold, as it desperately tries to figure out the beat. It makes the Beasties' syncopation feel wickedly subversive. ("Wait", says the computer, "you're allowed to sing off the beat? You utter bastards.")
It's a hilarious simulation of the more painful aspects of the amateur music experience, [2] but it doesn't have the rhythmic chops to out-rap a five year old.
Maybe it would do better if we gave it a metronome.
---
[1] "hard" in the sense of "actually, I think it's logically impossible".
[2] If only metronomes were issued at birth. Though, actually, that's happening now: I have hope that Rock Band and DDR will help to raise an entire generation that has a sense of rhythm.
We have already established that Songsmith does not break any new musical ground...but the more I think about it this is hardly an impressive technological feat either. I have a Casio keyboard from the early nineties that has about 100 tunes from different genres. You can change the key of these songs as they are being played and it makes them sound like an entirely new creation. We have also been able to analyze vocals and determine the tone for a while. Songsmith basically slaps the two ideas together. This could have been in beta by 1995.
Sure, but the key innovation here is YouTube. Cats have played piano for as long as there have been cats and pianos [1], but only now do the folks sitting around at home have the opportunity to turn their piano-playing cat into an international video superstar with ten minutes of work.
As a subject for academic papers in music or electronics, Songsmith doesn't rate. But as an amusing social media hack, it's just great. I can't wait for the advent of the next new genre: "pieces that were supposedly written by Songsmith, but have secretly been painstakingly assembled by actual musicians". Will that cause a Milli-Vanilli-style scandal, only in reverse? [2]
[2] I published this, then suddenly realized that Milli Vanilli is nearly twenty years old, and some of you might not have been born yet. Thank god for Wikipedia: http://en.wikipedia.org/wiki/Milli_Vanilli
Isn't it kind of sad though that MSFT invested x dollars in this and it only has value as (potentially) the next i-meme?
On an unrelated note, I am going to to start using that footnote technique of yours because I often have several tangential points, which makes my comments messy.
I am going to to start using that footnote technique of yours...
Hooray, I'm an innovator! ;)
[1]
----
[1] Though technically I think I stole the amusing footnote idea from Terry Pratchett, and from the Steve Meretzky/Douglas Adams team that wrote the HHGTG game.
Let's not forget the second half of most modern songwriting - lyrics. Music certainly has MANY mathematical elements to it. There is no doubt that computers can do a lot of lifting in orchestrating and making arrangements just using basic music theory. Some of it may even be creative and useful. Most of it will be elevator arrangement trash.
Still, you're not going to be able to have computers write meaningful lyrics. A lot of "human" factors go into writting lyrics and the lyrical melody.
http://www.youtube.com/watch?v=e1e_h1OJfS4
The program changes minor chords in the song to major, so suddenly the chorus seems peppy and bouncy. And its version of Crazy Train fits far too well and will have you laughing out loud if you're a fan of the original:
http://www.youtube.com/watch?v=VFZvLprEIpw
I like how some people are decrying this online as the death of music, like Songsmith is going to change how music is made. (Oasis writing music with Garage Band to the contrary, most people still write their music by getting some musicians together and figuring out how stuff sounds.) I think it's really neat that it can do this! The program's working very well - I'm installing the beta today and we'll see how well it handles real-people voices - and people expecting it to magically create the memorable hooks and snares from the original are people who still miss that a machine can't generate truly great music, because the best music comes not from formula but from the human mind.