Taste

There’s a thing I find amazing about music that you can demonstrate with a pair of musicians who are about equally popular but have completely different sets of fans.

For example, let’s pick a solo violinist in a prestigious orchestra, and a singer-songwriter with an acoustic guitar. Poll the fans of each of them to find out why they think their idol is great. It is easy to imagine them arguing about whose music is better, who is more talented, why skill matters, and so on.

Such arguments are pointless if they do nothing to increase our understanding of why both artists are successful, so I’ll skip rehashing centuries of debate about art and what makes art great and get straight brass tacks: anybody who understands music knows that it is musicians taste which determines their success.

Anybody can play a violin, but if you want to knock the socks off fans of classical music, you have to learn to play it in the very particular way those fans expect. You might wind up doing your own thing after you have earned their respect, but you can’t get that without showing the bona fides.

This requires an enormous investment in time. If you’ve ever learned to play an instrument well, you know it takes years to develop the physical skills that allow you to play in time and in tune. To move audiences with your playing requires even more than that.

You must first understand what it is to be moved by someone else’s playing, which means spending years going through records, finding players whose work blows your mind, then spending even more time figuring out how to reproduce what they played with your own hands on your own instrument. Chances are that over time, you will find new players who blow your mind way more than the first player you fell for, and you’ll do that all over again.

Almost every musician I work with still cherishes their first musical love, but they’re always looking for the way up, someone they haven’t heard before who does things they’ve never heard and maybe can’t immediately understand. Sit in a group of killer musicians and you’re guaranteed to hear them trading notes: have you heard this record? Did you see this video? They don’t care if its popular, only that its mind-blowing. This exchange of ideas about greatness is the act of developing really good taste.

The relationship between our singer-songwriter and their audience is also based on taste, but it’s a different flavor. Some of the greatest songwriters play guitar or harmonica with the sort of earnest half-assedness of a novelist hacking away at a manual typewriter with one finger on each hand. The job gets done, as does the art, but the art is in the letters, not the fingers. Their rapport with the audience matters, the set list matters, the venue, the clothes, the instrument, the opener and the vibe matters. Maybe somebody plays a solo but its not going to be several minutes long with a bunch of stunt-playing. This is not to say there aren’t some singer songwriters with amazing chops, its just that there are enough of them who don’t and we know there are audiences who are nuts for them anyway.

Taste also explains why musicians start playing at all. It is easy to find people who just want to be famous and whose concept of fame is standing on a stage in front of tens of thousands of people being seen while music happens. There may be choreography and elaborate costumes. There may even be lip-syncing, I don’t judge. Most of the music I like would be boring as hell to fans of such artists, not because my music is better, but because we have different taste, and that’s OK.

There is a risk in pursuing the kind of musical fame I just described. The barrier to entry is low compared to becoming a solo violinist, so the pool of candidates competing for the finite attention of pop fans is huge by comparison. To become a star in this crowded hungry field means you will have to be shockingly noticeable, for one reason or another. Or better yet, several reasons.

It should be obvious then why so many of history’s great pop artists didn’t rely simply on the music to demonstrate their exceptional taste. Consciously or not, they expanded the domain of their taste to include fashion, sexuality, politics, or anything of public concern. There are just as many historically great musicians who do none of those things and, for all we know, pursuing any of them might have ruined their chances. Imagine John Prine deciding to experiment with dance choreography and a smoke machine.

The potential payoff if we can get this right is huge though. Rather than investing a huge amount of time to appeal to a small group of people with very exacting taste and deep pockets, we could write songs for the tens of millions of pop fans without so much preparation and production. So Let’s pretend somewhat cynically that you and I want to be famous musicians, but neither of us knows how to play an instrument or sing well, and neither of us have ever written any songs. Moreover, we are in a big-ass hurry about it. What are we to do?

The obvious choice today would be to subscribe to one of these generative AI services that will translate our wishes into marketable product. Our desires are clear so its easy to tell the models what we want: we need a song that sounds similar to some other song we know, with catchy lyrics that describe something we are feeling with a tempo suitable for dancing. We’d likely want to use our own voice on any recordings we make since we want to be seen performing the music and people would likely catch on if our speaking voice wasn’t similar to our vocals.

If we can dance a little, maybe we’d film ourselves lip syncing the song we asked the AI to make for us, or maybe we’d have it instead generate some theatrical music-video style production with our music playing in the background.

Our nascent ersonal tastes having guided us through these choices, we now upload the video to TikTok and Youtube and wait for our fame. A few days pass, without fame. We decide to investigate.

We notice that our video appears in some lists of other videos of other songs in a similar vein, with a similarly small number of views. Our next step, therefore, is to distinguish our content from the others which are being recommended to some rando perusing recent TikTok uploads. How do we make our current song better, or make a second song better than the first? Clearly what we need to give the AI more specific instructions, but how?

First we try typing more emphatically at the AI, clarifying that we don’t want just any old song, we want a really good one, and so do what you must to add some pizazz to our single. We burn up some tokens and a take a new mp3 out of the slot.

This new version of the single sounds amazing to us after being modded by the AI, so we update the video and upload it again, and our views double from 2 to 4. We are clearly on the right path.

The problem we soon encounter is that the songs we improved through tweaking the prompt, while different, are all about the same in terms of catchiness. If we’re going to improve the single any further, we’re going to have to be even more specific to the AI about how it should improve the song.

In order to do that, however, we’re going to have to work out what it is not right in the last version. Since we don’t know, maybe we send the URL to one of our favorite artists via Instagram, but we never hear back. We ask the AI what it thinks we should do, but it responds with generalities describing characteristics of catchy songs. We could ask it to do some of those things, but every time we do so we have to upload the video again to see if the clicks go up, and now the video is starting to show its age and our views are going down.

We have all the technology that we need to produce and distribute music virtually free of cost. We have inexpensive cameras, lighting, and video editing software, and we can film almost anywhere we want. Instruments, either physical or virtual, have never been cheaper than they are now. All that is missing is taste which, in the past, could only be acquired through years of listening, studying, and practice. Or, as many bands still do, hiring a producer with all that taste.

To fix what is wrong with a song (or better yet, to write a song that doesn’t need fixing) requires taste adequate to identify what if anything is wrong with the song, and to write a good song requires the taste to comb through your memory of sounds, words, and physical motions. Things that you spent your entire life listening to and internalizing.

Even doing all that does not guarantee taste; for every famous musician there must be at least a thousand people who listened, studied, and practiced their asses off yet produced music that did not resonate with listeners and who never became famous. It’s because skill is more common than excellent taste. Good taste is pretty rare, otherwise it would not be so remarkable when we discover it.

Large language models and the various other contemporary generative “AI” systems are probabilistic by design. Their response is always an inference about the most probable response to a prompt based on training data which includes virtually every bit of data that humans have produced. Not the most tasteful thing but rather the most likely thing. The most average. If we’re trying to appeal to the broadest audience this is great.

If we trained these models on only the works of artists we thought had exceedingly good taste, we would get outputs that were akin to the average of all the work those people produced. The catalog of prolific artists with excellent taste is probably half B-sides. So we’d have to limit the training program to just their best work, but at that point the songs the model produces sound too much like those hits for us to be able to market them.

We’re not too concerned about being considered great yet, we just want to be famous, so we’re cool with using the model that was trained on everything. We’re aiming squarely at the middle.

Although I’ve spent nearly my entire life writing and performing music, and for some time fancied making a career of it, I have other loves as well. I learned to program computers as a child, and have been making a living of it for thirty years. Lately it is impossible to follow news related to my field without being inundated with discussions about whether programming is a dead or dying field and how we will need to change our approach from writing software by hand to one of prompting a Large Language Model to do it for us.

Let’s pretend then, somewhat cynically, that you and I want to be wealthy software developers, but neither of us knows a programming language or how computers work.

Is this how you want your work to be remembered?