Can a robot write a symphony? (AI reliably writes good classical symphonies before 2025)
23
Ṁ2541
Dec 31
14%
chance


Can an AI reliably write classical symphonies that I like as much as existing ones?

Symphonies must be purely AI written without human intervention and without excessive cherry-picking. The AI system must be able to cheaply produce additional symphonies that are diverse enough. Prompts will not count as human intervention (as long as the same prompt works multiple times, or there is a way to automatically generate prompts). Humans performing based on AI-produced sheet music counts, if the process is well-documented. Same goes for any kind of hybrid performance.

Examples of things that don't count: one-off PR events, music clearly different in style from 19th-century symphonies, humans cherrypicking parts of the symphony and then putting things together.

comparing symphonies:

I will sample 10 symphonies from the most promising system (and take suggestions from the comments into account). If no such promising AI system exists, this market will resolve NO.

For the human list, I will pick some symphonies written in the 19th century (also taking suggestions from comments). I will exclude some existing well-known symphonies from the human list: https://www.classical-music.com/features/works/20-greatest-symphonies-all-time

Then I will listen to symphonies from both lists (which might take a long time) and resolve based on average subjective score.


Note that this is not a test whether I can distinguish AI-generated symphonies from real ones.


If there is promising material but I fail to run the test for some reason (like I become inactive on manifold), it resolves N/A.

Note: I might change or add more detailed resolution criteria in the future.

Feel free to ask in the comments.

Get Ṁ1,000 play money
Sort by:
bought Ṁ10 YES

is suno.ai an update for anyone?

by 2100, maybe.

If I understand correctly, this market assumes that either: a) orchestras are going to play symphonic music generated by AI (unlikely), or b) synthetic orchestras will be indistinguishable from real ones (also unlikely).

In the case where the capability is clearly there, but there isn’t a reasonable sample of material for the test in the description, how would this resolve?

If there is not enough material to run the test, it will resolve NO.

Synthetic orchestras don't need to be literally indistinguishable from real ones in order for this to resolve YES. For example, the second half of https://www.youtube.com/watch?v=aC8I2YvL6Uo sounds fine to me (although it is a bit short).

I expect that The Human Spirit will prevail here by virtue of not caring about classical symphonies enough to spend a billion dollars on a machine that makes some passable ones.