Hacker News Reader

Comment by Areading314 | Hacker News Reader

This may be academically interesting, but the music still sounds fake enough to be unpleasant (i.e. there's no way I'd spend any time listening to this voluntarily).

View on HN

I disagree with this: some of the stuff being played live on Twitch right now I would definitely voluntarily listen to, especially if it was actually performed by people with real instruments and not just midi.

There was a baroque-pop song just now that had a ritardando that almost gave me chills. Probably copied from Chopin, but still.

Yes, the lack of human factor is very noticeable if you are a musician. I believe it's pretty similar to when grandmasters can tell if they are playing a human or a bot. Something that's hard to explain. Now saying if it's better or worse is just subjective.

Or where an AI can tell if they're playing a human or a bot.

Grandmasters also use different tactics against bots: https://en.wikipedia.org/wiki/Anti-computer_tactics

The concept of how different human intelligence is from "AI" fascinates me, as it would seem to say a lot about the nature of intelligence and how far we are from GAI (pretty darn far).

One thing with it is the distance from getting a computer to compose anything to getting it to compose close to human level could be 90% of the distance to getting it to compose at a super human level.

Even several months before Alpha Go beat Lee Sedol there were people saying the AI could be good but never great. Now everyone admits it's super human.

I agree that getting really texture and subtlety in the work is really hard. However a really good system has the possibility to be better than any human has ever been, and IMO that's super exciting.

What does "super human" mean for a cultural artifact like music? With a game, "super human" can mean "beats humans". For supervised learning, where there's some label other than human opinion, "super human" can mean "more accurate than a human labeler". But for music composition, you need a difference source of information (listener ratings? sales/streaming plays?) to determine what "better" means, which would be a meaningful structural change rather than just an extension of this technique.

So long as the training set only tells it what human-composed compositions look like (MIDI files of existing music), a system has no signal to find "super human" territory.

You're right measuring the skill of a composer is hard, however as you say listener ratings and sales/streams would be a pretty good metric. Even for songs in the top 100 charts how high can an AI get? If all the top 10 songs were written by an AI that would be pretty convincing it was super human I think.

On the issue of generalising alpha go learned go by watching human played games and by playing against itself and is now better than any human. So it's not true that a system is limited only to the skill of the examples it's shown, it's possible to design something which can surpass the examples it learned from.

Yup, this is why I said that the output is somewhat uninteresting musically - there are actually machine-learning music models that do a lot better from that POV, despite being seemingly unambitious in many other ways. (I've mentioned one elsewhere in this thread, but BachBot and DeepBach also deserve mention!)

agreed, it captures mostly rhythmical elements which make it sound "like some music" but it tends to sound like a broken record soon. Maybe its decoder was not designed to capture very long-term dependencies and that makes it sound so "myopic". It s very promising though and it seems in the future we 'll have half-decent pieces suitable e.g. for background music.

How does it sound fake? It sounds lovely to me, especially much of the Chopin+pop piece sampled in the article.

Lack of narrative. It's less disturbing if you're in the mode of listening to background music, but for active listening, classical music is so much richer.

I believe they selected some of the best-sounding pieces. Playing with their tool (which is a lot of fun) is more hit-and-miss: some interesting results, but also some slightly unpleasant music.

Reply on news.ycombinator.com