AI and Compositioning?

[Digidog]

Now that there are programs that are so advanced in calculating, deducing and compiling that they easily can write sensible texts, produce comprehensible reasoning on a topic and produce both realistic pictures and reasonably correct answers to questions; what do you TC:ers think that will mean for songwriting and composing?

There are already programs that can produce solos resembling various jazz artists, and ChatGPT can produce lyrics, that are still somewhat square. How long will it take until there is a composition generator, that from sheer statistics and algorithmic comparison can piece together music? At first small ensembles, four or six parts, but with increasing data access big band and symphonic scores.

I'm not too sure there is a future for song writers and composers, depending on how the use of reference data and statistical references is allowed. Based on how easily ChatGPT produce text, it won't be long until "ComposeGPT" can generate whole symphonies "Beethoven-style" with modulation to parallel modes in the second movement, and a fugue in the third.

Computerization makes every futuristic prospect look so bleak; boring, uninspiring and lifeless. Maybe I'm overly pessimistic, but in general new advances are immediately used and misused, regardless of what drawbacks there obviously are, or what setbacks they mean for human conditions.

What do you, fellow TC:ers, think?

[harrisonreed]

viewtopic.php?p=188884&hilit=AI+composition#p188884

[Bach5G]

Never. It’ll get close, maybe close enough for most people. But a lot of it will be generic muzak, cheap to produce (although lucrative for a few), forgettable and disposable.

I can’t see it consistently nailing that special something. Example: I was a big Steely Dan fan. I’ve never heard anything memorable from the post-Gaucho reunion phase. Whatever they had, they lost. I’ve never heard a successful cover of Fire and Rain by anyone. AI might give you Salieri but not Mozart.

Time for another look at:



IMVHPO of course. YMMV.

(To be fair, rewatching that clip, I wonder if F Murray Abraham delivers that special quality. That initial sideways glance … marvellous.)

[Digidog]

viewtopic.php?p=188884&hilit=AI+composition#p188884

That is an example of one level of creating music, but today's GPT:s have so much more statistical background data, and so much more complex algorithms that I would think that creating full scores are within their capability. If the programmers then have quantified and categorized the parameters of f.ex. a big band score, the program would definitely be able to piece together notes in totally meaningful melodical patterns and rhythms.

Then if the data were quantified and statistically specified for individual composers, I'd guess it could create scores similar to those composers that make its database.

The process would be that the program randomly creates a large number of scores based on the given statistics, and then succesively sort with a new level of data statistics and keep those with the best next-level sorting, in a repeating process until there is only one score left. Which then statistically should match the style and form of the given composer.

@Bach5G: In a way, I'm with you on Steely Dan's decline, but what I mean is that resemblance is statistically valid as an outcome; mathematically speaking, an approximate value.

[EDIT] I absolutely love Amadeus!! I should rewatch it!

[harrisonreed]

Yeah when I posted that, not actually that long ago, that was the only thing out there. I think the day where can co-create a symphony with AI by writing the melodies and basic structure and handing it off to an AI for finishing are right around the corner.