-
On an academic level...cool.
But to get computer programs which can replace improvisers and other Human functions- I am of course biased against that and openly admit that.
The IrealPro App uses programming but to help the Improviser .. not replace him.
I don't think you will be able to do it for another 20 years sonot too worried...but I don't welcome it to be honest.
-
02-20-2018 09:53 AM
-
I think our art form is well shielded from such threads due to there being absolutely zero financial incentive to undertake such a project commercially
Originally Posted by Robertkoa
If it costs you more than 700$ to develop it, you'll probably lose money regardless of how good the system is in jazz improvisation.
-
AI can never actually replace human expression, so as long as humans value human expression in the Arts - provided humans can continue to express something meaningful or compelling- then live performance will always be around, and hopefully valued according to it's quality and rarity.
The problem is, as I see it, that recorded performances (Films, recorded music etc) can be and are being replaced by computerised impostors. 30 years ago I would not have believed that thousands of young hipsters would one day pay half their paycheque to witness some bozo manning a coupla turntables whilst waving his arms around behind a wall of pre recorded sounds with very limited live sonic interaction ( "oh wow, that high pass filter sweep he just did with his left hand was just so amazeballs!"...).
So who knows what it's gonna take to entertain the masses in future? Surely a trapeze artist will always be good for a thrill, visually obvious risk taking should always excite people, but will anyone care how many risks a Jazz guitarist takes on the band stand? How will they know spontaneous risk taking from pre composed playing? It seems amazing that non players had a sense of what was going on in that regard 50 years ago, and we can only hope that future generations will use all that extra leisure time they'll have in pursuit of the appreciation of the high arts.
Because let's face it, Jazz, I mean the good shit, is indeed a High Art, who's realm is inhabited not merely by humans (as opposed to machines), but by super humans, titans, masters of the universe, demigods even! A computer may have beaten Gary Kasparov, but no machine will ever match Wes Mont-fuckin'-gomery!
...
-
Talking about how the BIAB improviser can be surprisingly convincing I think emphasises a point that I kind of focus on as a musician, whether as an improviser and a composer.
You need to get improvising down to the point where you can do it mechanically without much need to think - chord tones, extensions, neighbouring tones, etc. You can do it, but it might not sound too good - at first.
Once you can do that mechanically, that's where the element of personal taste and the human touch really starts to become more important and the real art starts to come out in my opinion. I guess I'd put it like this - a computer can do the mechanical aspect of improvising surprisingly easily, as the BIAB improviser kind of shows. But so far, and I'm assuming for a decent period of time, it requires a human mind with experience and discernment to make the choices of what to play when that makes a piece really musical - even if every solo is made by combining "licks" or "fragments" or "figures" that have been drilled hundreds and thousands of times before.
Kind of like how writing a speech isn't so much about inventing new words and figures of speech as it is taking pre-existing ones and putting them into a new context.
-
I propose a Butlerian Jihad
-
Yeppity yep! Us humans will play a "preprogrammed" bit of material mid solo which will lead to certain spontaneous decisions about what to play next, do we mimic the arc of the line? Invert it? Paraphrase the rhythmic aspect somehow, follow with something more emphatic? Bring down the exuberance a little? etc etc. In other words we interact tastefully and/or emotionally with what we just played, if we're lucky. Machines probably don't do that, and if they ever did, then they're only ever going to closely approximate a human in this regard, just to show it can be done (sort of...). Originally Posted by Shadow of the Sun
But let me put it this way, if a robot can be built to stand on a stage and deliver improvised stand up comedy, would anyone laugh? ...
-
1000010001010101011111001111
Punchline
00010110101110011111
-
Originally Posted by christianm77
lol
-
I don't wanna burst no bubbles but if you captured Charlie Parker's brain's electric signals that controlled his motor functions during his solo on Anthropology, and ran a PCM algorithm to digitize them, it would look not too dissimilar to that. Originally Posted by christianm77
If you had high sampling rate, you could even reconstruct the original analog signal with a half decent processor.
Joking a side, you can train a robot to play sax in the style of Charlie Parker if you used digital format of all his improvisations as training sets. All the jazz listening a human jazz musician can do to train their ear for the style can be achieved in a few nano seconds with a computer. In fact more jazz than a human can listen to in their life time. There is no magic to learning phrasing and dynamics of a musical style by listening, it's the same old statistics. Thinking that improvisation is a "magical" human merit is the same primal tendency that lead humans to think moon was god before they had telescopes.
I am not saying machine learning is there yet, but theoretical principles are well understood.
-
Originally Posted by christianm77
There are 10 types of people in the world, those who read binary, and those that don't.
-
Nice comment Tal.
Not to mention that a field of AI research in music is focusing on the transcription/digitization process. I think within 10 years we will be able to transcribe any audio into western musical notation (insofar as it is expressive by western musical notation).
-
10 years is a very generous timeline it is likely to come much sooner
-
I agree, I think... But the other point I was trying to make was that the nuanced spontaneity of the human can't really be mimicked. Let's say you could recreate Lenny Bruce as a robot/hologram, let's say you feed it every joke Lenny ever told. From a distance, this may be every bit as funny as the real thing for a few minutes, until a heckler yells out and the robot has no response, or an unfunny one... Originally Posted by Tal_175
-
I don’t think that’s correct. One advantage of recent machine learning techniques is the ability to model not just common behavior but also rare events. It’s a matter of design, but a programmer could incorporate heckler comebacks to the training data. Originally Posted by princeplanet
-
For the record, I don’t have a PhD in AI like tal but I have a masters in AI/statistics and I work in the industry.
-
Maybe, if Lenny Bruce himself was the programmer! Originally Posted by omphalopsychos
-
I don’t see why he would have to be.
-
Don't forget that the goal is not to create a statistical model of Charlie Parker's brain in order to "predict" what would his next solo would be if he was alive. Originally Posted by princeplanet
The goal is for computer to play an improvised solo in a way that a fan of Charlie Parker's can believe that could reasonably be a Charlie Parker solo, ie the Turing test.
-
Imagine you told a nuanced joke to a comic robot, what would make it laugh? Presumably, and optimally, everything that has ever been known to make humans laugh in all of history (big programming job right there). But a lot of humour comes from spontaneous nuanced reactions that have never been uttered before in ways never before witnessed. It's beyond Chaos, or Randomness. We're not talking about chess here...
-
Hahaha Originally Posted by mr. beaumont
My joke was funnier in hexadecimal
-
If I'm belabouring the Comedy analogy, its because the same parameters apply in Jazz. Nuance, spontaneity, interaction, to name just 2 (joke, see?) ....
Look at the jokes on this page: Christian's "1100011" joke. Even my "lol" reply... it's nuanced such that you could not have programmed a computer to have made a subtle response like that in an attempt at humour. Mr B's "There are 10 types of people in the world, those who read binary, and those that don't." ... Would a computer be able to respond with Christian's "Mine's funnier in Hexadecimal"?
Or are our jokes so lame that a robot would wipe the floor with us all?
-
Computers have been wiping the floor with the best chess players in the world and laughing at their moves for 20 years now. Originally Posted by princeplanet
Or are our jokes so lame that a robot would wipe the floor with us all?
Tell that to Norwegians who consider Mangus Carlsen to be their national hero.
We seem to be able to distinguish between human achievements and computer achievements and appreciate them differently.
-
Princeplanet, I think you're missing a big point. We're not so good at AI today because programmers are writing very complex software to incorporate domain knowledge. We're good at AI because programmers no longer need to do that. The programming is no longer explicit. The last 30 + years of machine learning research have focused on developing modeling frameworks that are adaptable to new problems, such that (this is a gross simplification) a programmer needs only to process observations and pass them through that model architecture. Because of this framework, we don't program rules anymore. Tal was mentioning this, i.e. the dichotomy between rule-based AI and machine learning. In machine learning the programming isn't explicit. You feed a computer observations and it learns the correct response through trial and error, much like pavlovian conditioning. It's this way that we're able to get so nuanced, not because we are able manually to encode nuance.
-
Are you also aware of Martin Norgaard's MATLAB analysis of Parker? Conclusion was that 99.3% of his playing was pattern based. Even though you can't trust such dissertations due to programmer bias or ineptitude (unkind word, but you know what I mean...), it seems true that Bird did have his stock patterns and devices. But so what? It was still "spur of the moment" that made him reconnect all the data in mysterious and wonderful ways. Witness the famous "outtakes" sessions- all different- all perfect - all genius... Originally Posted by Tal_175
Bird even told spontaneous "jokes" in his playing, littles cues, snippets or quotes in response to the room (a cute lady walks past) that his band mates were always hip to. Good luck programming that shit into Robot Bird!
-
My point about Chess was that it is infintely easier to predict than a good Jazz musician's solo. Possible chess moves are finite. Possible Jazz solos, along with all the nuances in their performance, are infinite. Originally Posted by Tal_175
I hope...
Reply With Quote
Video: The Harmonic Minor Scale in Jazz
Today, 10:47 AM in Improvisation