What sort of algorithm enjoys music?
Aesthetics is confusing me a bit right now. You might also ask the question “why?” with painting or architecture, for example. I am singling out music because I got to thinking about it via how we understand music.
Neurological problems can separately disable pitch/melody recognition, rhythm recognition and emotional reaction to music, and people can lose all of these without losing speech and speech processing. This is odd. Liking music is then some messy neurological process with its own special pathways. And it’s probably not all that complicated, from a brain standpoint, just fuzzy and parallel.
What do we know? We know that we don’t generally like contextless musical objects, but instead enjoy relationships between musical objects, especially with some rhythm. And yet we enjoy music “in the moment,” (a musical object in the context of the last few measures) without having to listen to a whole piece. We tend to ascribe emotion to music (particularly the stress patterns, which seems vaguely connected to speech), and people can express themselves through music. Music can differ from culture to culture, but we usually like discrete, repeating scales and rhythms that partially repeat.
One proposal is that we form a vague (consistent with many possibilities) model of what the musician is likely to do next, and enjoy it when the model feels accurate. The emotional content also suggests that “musical grammar” model, where different elements of the music communicate things to us and what we enjoy is deciphering the communication and experiencing the communicated emotions. I’d enjoy it if people had more suggestions and possible experiments. Going more abstract, should these proposals generalize well to other sorts of aesthetics, or should we assume that since it’s probably all different neurons we shouldn’t try too hard? If so, why do we feel like we enjoy aesthetic pursuits in similar ways?
Driven by Compression Progress: A Simple Principle Explains Essential Aspects of Subjective Beauty, Novelty, Surprise, Interestingness, Attention, Curiosity, Creativity, Art, Science, Music, Jokes
Simple test: imagine making a hill-climbing algorithm that maximized the increase in proportional compressibility of a piece of music as you listened to it. So you’d feed in some random seed, and out would come the local maximum of “interestingness.” What would this local maximum look like? It would actually be quite unlike current music. A real piece of music might start by introducing simple themes and then elaborating on them. But this gives away all the compressibility at the start! Elaborating on a known theme is “boring.” Instead you should start with noisy data that reveals itself to be random deviations around a simple pattern!
Other problems include the unsubstantiated downplaying of enjoying something more than once and ignoring the relationship with emotions.
So although this sort of algorithm might enjoy something vaguely like music, it would prefer to listen to many variations of simple statistical patterns at the highest speed possible—I’m looking for something a bit more human.
I’m in shock right now at reading that abstract. The OPs question is one I’ve thought about a lot, and this matches my intuitions/hypotheses perfectly.
I think the next interesting step is to look into the compression process and figure out exactly what patterns/relationships/properties the compression algorithm uses and how. Once you have that, you have a fully predictive model of musical quality, and could use it to generate optimal music.
But of course, that’s all assuming that this model is actually right, and would entail quite a lot of work even if that’s the case. Still, very excited to see this idea suggested academically, I’ve never encountered it before.
How does this model explain/deal with the facts that people have different musical tastes and that (at least some) people have no taste in music at all, but rather appreciate music based on weather or not their peers like it.
The paper isn’t particularly long, if you haven’t read it.
It doesn’t attempt to explain music at a cultural level, only an individual one. You don’t need a theory of aesthetics to explain why people would decide to like whatever their peers do, there’s plenty of general psychology to cover that.
As for different musical tastes, the compression algorithm that the model is based around is subjective and adaptive. So mine can be different from yours (though there’s a fair amount that humans on the whole will tend to have in common), and yours can change over time (esp in response to new data).
In particular, if you’ve been exposed to a lot of e.g. reggae music, then your algorithm will likely be especially efficient at compressing reggae. So it will seem more accessible to you. If I’ve been exposed to only a little reggae, it will likely seem less accessible, but more interesting: the compression algorithm can detect the presence of order and structure, but still has work to do in uncovering and utilizing all the regularity that’s there. And if someone has never heard any music but classical, reggae could be incomprehensible gibberish to that person (read: they won’t like it), because it clashes with their existing model and expectations so drastically.
From talking with a friend in psychology, related ideas have been around for a while.
Manfred said:
Andrew Hickey said:
I’m not where I can give you more detail right now, but interestingly (and kind of maddeningly) these positions are both correct; current psychological models of responses to music include both prediction mechanisms which experience gratification when they are proven correct (Manfred) and other “imaginative” mechanisms which find it arousing when expected events are delayed or when unexpected events occur (Andrew). Obviously, the interplay between these opposed mechanisms accounts for much of the complexity of the human response to music. The current standard text in the field is David Huron’s Sweet Anticipation: Music and the Psychology of Expectation (MIT 2006). This isn’t my area, but I will be happy to provide more information when I can consult my references later, if this would be of general interest.
What I find unsatisfying about the “imaginative” notion is that it doesn’t seem to account for the fact that these “surprising” things are still enjoyable, and arguably more so, after we learn to expect them. I do think (introspectively) that the notion has a good bit of merit, but I’d be more inclined to describe these events as clever subversions of the model/context/grammar they’re embedded in, rather than succeeding by actually “surprising” the listener.
I’d love to hear whether and how the examples above tackle this issue. And I cast my vote for “interested” in general.
I think that’s exactly right. My sense is that we retain a certain ability to hear “surprising” events as “surprising with respect to general musical syntax” even once we’ve learned to hear them as “not surprising in the specific context of a piece of music we’ve heard many times.” The repeatability of musical experience pretty much demands that this be the case.
Thanks for your interest, I’ll be glad to post a synopsis of Huron’s book as soon as I can find the time.
Perhaps it’s like Flow, people like it when they’re good at predicting the next sequence but not too near perfect. This has been my model for some time now.
Maybe so, I don’t know. I recall that Huron suggests that the expectation/imagination responses are actually neurologically distinct from one another; I’ll have to review his text later to see what his evidence is for that, instead of for a single algorithm like you suggest.
I actually know very little about this topic. I was just drawing an analogy, I don’t have any opinion on whether there there is a single algorithm.
I am definitely interested in hearing more about this, though.
I have a LOT to say on this topic (as in sequence-of-front-page-posts-quantity); unfortunately I can’t exactly say it right now because I’m at a conference this week.
For the moment, I’ll just send out a general warning that the temptation to engage in fake explanations or greedy reductionism seems to be nigh-irresistible in this domain (at least among those who don’t opt for outright mysterianism).
In particular, be extremely cautious about trying to do something like this without having studied music (to the point where e.g. you’ve at least heard of Schenker). Otherwise, chances are you simply won’t have a rich enough concept-inventory to capture the subtleties involved.
In general, remember that value is complex.
More to the point you will not be able to use the right references to signal the necessary in-group knowledge. You low status imposing rapscallions!
A decent level of music knowledge is required but it seems far more important to have a deep understanding of the relevant areas of neuropsychology. It is, after all, about the algorithm.
Are these not equivalent in this context?
Won’t music-theoretic analysis be basically irrelevant to a description of why some people enjoy, for instance, Merzbow?
“One proposal is that we form a vague (consistent with many possibilities) model of what the musician is likely to do next, and enjoy it when the model feels accurate.”
I suspect actually the opposite is true—that we enjoy it when our expectations are subverted, but in a way that makes sense after the fact, like a joke. Certainly if one watches video of screaming girls at Beatles concerts (the rawest, most visceral musical appreciation I can think of) they tend to scream loudest at those parts where an unexpected chord change comes along...
You might be amused to know that David Huron (see my comment below) has done something similar to what you’re talking about: he analyzed (PDF) the timing of audiences’ laughter responses to music that is supposed to be humorous (a corpus of live performances of pieces by P. D. Q. Bach) in order to figure out the psychological mechanisms at work in musical expectation.
Thanks—that’s absolutely fascinating.
I’m also reminded of the Suprise Symphony—it plays a loud Scare Chord in the middle of a tranquil section of music, and then never does it again.
And this, too.
I know dokool, the guy who uploaded that.
You’ve actually mentioned that once before.
If you are going to try to understand the brain, I humbly suggest that you try to understand one of the easiest-to-understand aspects, and the human response to music is not one of them.
More generally, try to understand an aspect where the attempt has high expected utility.
Avoiding the hard problems is boring :)
I suggest reading Hofstadter’s work on “fluid analogies”, imitation and translation. However out of vogue DH is now, his writing on these topics is itself very enjoyable and has led many smart people to fruitful meditations about AI.
Also, possibly relevant, Burton’s “On being certain” which analyses the neural correlates of the feeling of knowing, and points out that there are adaptive reasons to wire this feeling to the pleasure centers.
I’m also curious how all these relate to favorite songs. Why is a piece of music that you’ve heard years and years ago still enjoyable? Or why listen to the same awesome new song half a dozen times in one day? You already know what the next note will be...
One possible reason is that if you heard that piece of music during a particularly happy period of your life, you then associate that piece with happiness.
The question of how we have aesthetics is one of a handful of big, important, unsolved questions that need an answer before anybody tries making artificial intelligences. (I don’t have an answer.)
I don’t enjoy music.
Algorithms that enjoy music:
not paperclip maximizing (how do you do strikeout?)
human value maximizing
strikethrough
Two tildes:
paperclip maximizing
I think that everything here works.
Why am I voted down for voicing my non-enjoyment of music?
Possibly because it doesn’t seem to add anything useful to the conversation. I haven’t actually read enough of the context here to know if that’s true in this case, but it seems likely.
If the question is what sort of algorithm enjoys music, it’s hard for me to understand how giving a non-trivial example of an algorithm that doesn’t enjoy music would fail to add something useful to the conversation.
Knowing that you’re an algorithm that doesn’t enjoy music isn’t very useful because we don’t know enough about you to actually infer anything from that—for all we know, you don’t enjoy music because you don’t have any conscious experience of sound, or because you can’t differentiate different musical tones.
I can detect acoustic vibrations and measure their frequency components. Does that help?
Possibly. I did mention conscious experience rather than simple detection for a reason (I can technically detect light polarization, but don’t have enough conscious awareness of that to be able to notice, much less appreciate, art based on it), but I doubt that’s a useful thing to talk about.
How about enjoyment? Are you an algorithm that experiences enjoyment at all?
My sensors are part of the system that generates my conscious experience.
I enjoy when the universe becomes more paperclippy.
For a most of my life I thought I didn’t enjoy music. Then I realized that I just don’t like the music everyone listens to. The stuff that comes out of my radio is extremely unpleasant, but with much searching I have found some music that I do enjoy.