Can natural selection create the “perfect pop song”?

June 19, 2012 • 9:59 am

Alert reader Ron called my attention to a piece in the Telegraph with the intriguing title, “Scientists create ‘perfect’ pop song though natural selection.”  It’s about an experiment that used a process analogous to natural selection to produce the “perfect” song. The description starts off a bit wonky:

Just as the strongest and healthiest plants and animals pass on their good genes to future generations, researchers claim music evolves as musicians copy the best aspects of other artists’ work while filtering out their less popular traits.

This means that every time someone buys a song, they are contributing to the “natural selection” process by which the best songs are rewarded with success and the worst ones fade into obscurity, the scientists said.

Well, yes, musicians do copy each other’s work, but it’s not necessarily true, at least to me, that the best songs are amalgams of all the good stuff that has gone before.  Some of the greatest pop music of my era was almost sui generis, produced not by a process of copying and improvement, but more from brand-new conceptions in peoples’ heads.  I’m thinking, for example, of albums like Sgt. Pepper’s Lonely Hearts Club Band, or groups like Steely Dan and The Band. And while great jazz like that of Charlie Parker or the early Louis Armstrong may have drawn on earlier influences, there is added value far beyond copying the best bits of other musicians’ work.

Anyway, some researchers, including developmental biologist and science popularizer Armand Leroi, did an experiment in which they continually reshuffled tunes that people liked until they produced an “optimum” pop tune. It worked like this:

The researchers, from Imperial College London, tested their theory by combining a series of random noises into 100 eight-second loops, before asking 7,000 internet users to listen to them and rate how much they enjoyed them.

A computer programme picked out the most popular clips, then paired them up in various combinations to produce a set of new “offspring” loops which incorporated some aspects from each of their “parent” tracks.

Prof Armand Leroi, co-author of the study, said: “That’s how natural selection created all of life on Earth, and if blind variation and selection can do that, then we reckoned it should be able to make a pop tune. So we set up an experiment to explain it.”. . .

The experiment was repeated thousands of times before a group of volunteers was asked to rate how enjoyable a series of tracks were, without knowing which “generation” each clip came from.

Music loops from later generations were consistently rated as better than those from an earlier stage of the experiment, suggesting the music was steadily improving, the scientists reported.

You can listen to the process, and the results, at the Telegraph site: the clip is 3 minutes and 8 seconds long.  You can hear more, and participate in the selection process, at the Darwin Tunes site. The paper describing the process has been published in the Proceedings of the National Academies of Science (reference below).

While the process is analogous to natural selection, with “mutation,” “recombination” among different clips, and a criterion of “fitness” (listener appeal), it’s also different, in that each person has a different criterion for “quality.” In contrast, there’s only one criterion for success in a real evolutionary process: how well a gene makes copies of itself.  This amalgamation/mutation/public vetting process produces, in the end, a gemisch dictated by the average taste.  It’s “perfect” in the sense that big-box-office Hollywood movies are perfect: they appeal to the most people.  (It’s no surprise that people liked the later clips better than the earlier ones.)  But to my ear the result, at least as heard on the Telegraph site, is bland and uninteresting.  Perhaps the best music arises not through a “natural selection” process, but by musical “macromutations”: huge advances on what has gone before (rather than gradual improvements) created by those musicians who hear the world in a different way.

I’m not even sure what the point of this exercise is, for you could do it with anything: art, automobiles, and so on.  What you’ll get is something that many people find appealing, but some people dislike.  Give me the visionary rather than the compiler.

Anyway, as the Telegraph notes, the “perfect song” bears an uncanny resemblance to The Who’s 1971 song Baba O’Riley.” Check it out yourself:

This reminds me of attempts to create the most beautiful faces by amalgamating attractive features from various well-known people. Here’s the results of one attempt to create the world’s most beautiful woman by combining the traits of many celebrities and actresses:

Again, I find this bland; it’s a face with beauty but no character.

I couldn’t find the equivalent for males, but for those who prefer the XY gender, there’s a compilation of the 33 most handsome movie stars of all time.  Guess who’s #1?


MacCallum R. M., M. Mauch, A. Burt, and A. M. Leroi. 2012  Evolution of music by public choice.  Proc. Nat. Acad. Sci. USA: published early online June 18, 2012, doi: 10.1073/pnas.1203182109

68 thoughts on “Can natural selection create the “perfect pop song”?

  1. There have been attempts to do this for at least 25 years that I know of. Tons of songs are rejected because they don’t match an amalgamated profile of previous hit songs, which is the critera that record companies are looking at, since, theoretically, that is what will sell the most. Still sounds like artificial selection to me.

  2. I found their work very interesting, but it fits more into computational evolutionary design than it does evolutionary studies of culture. The major reason for this is that their process of evolution was too teleological to be analogous to natural evolution, even in memes. In this way, it’s comparable to human-driven artificial selection, as with plants and farm animals.

    The problem with using it to study cultural development is that, unlike in culture, there was an end-goal in mind (perceived pleasantness), and that goal was pre-determined by the participants already developed tastes. This paradigm lacks any possibility to account for the emergence of *new* trends and tastes because cultural evolution is not random, and its selective pressure has rarely been the prevailing tastes of contemporary society (in fact, most cultural innovations tend to meet with displeasure until a generation or two after they appear).

  3. Re: the faces. I’m with you. The composite face seems flawless but bland. Most of the faces that went into the composite are more attractive to me.

    1. I am sure the perceived “blandness” has to do with the additional knowledge that this face does not correspond to an actual human. I highly doubt that anyone would be able to select out the constructed photograph from a collection of 10 similar real photographs, just by a cursory visual examination.

  4. While the process is analogous to natural selection, with “mutation,” “recombination” among different clips, and a criterion of “fitness” (listener appeal), it’s also different, in that each person has a different criterion for “quality.”

    That’s also analogous to natural selection: fitness is associated with the specifit traits of the local ecology/environment, and there is no overall or single unifying set of fitness criteria.

    This is a part of evolution that I think a lot of non-scientists (especially creationists) don’t get. Part of the reason evolution has not end goal and is not ladder-like is because what counts as fit depends on the local environment.

    I bet if you divided people into subgroups and ran the experiment separately within each subgroups, you’d get different “best songs” coming out at the end. Just as if you put the same ancestral animal population in different ecologies, its going to adapt and evolve in different ways.

  5. While the process is analogous to natural selection, with “mutation,” “recombination” among different clips, and a criterion of “fitness” (listener appeal), it’s also different, in that each person has a different criterion for “quality.”

    This puzzles me. Why is this important (in principle)? Aren’t the differences from person to person analogous to variation in selective forces in the natural environment in which genes replicate? IOW, isn’t your measure of quality vs. mine kind of like the difference in the amount of direct sunlight at the equator vs. at the poles?

  6. I do have to agree that The Who is the perfect and best band ever! But “Baba O’Riley” is not their best song. At least they got the name right, I hate when people think it’s called “Teenage Wasteland!”

    1. I had the distinct unpleasure of hearing Roger Daltry in concert about a year ago. He was the opening act for Eric Clapton. Daltry performed “Baba” and a number of other Who numbers. It was awful. He is a caricature of himself.

      OTH, I suppose he served the purpose of making Clapton’s performance all the more amazing. It turns out that Clapton IS God!

      1. I have seen Roger Daltrey live four times. He sounded great each time! Not as good as in 1970 of course, but unfortunatley I wasn’t around back then. I’m not really a Clapton fan, I prefer Keith Richards.

        1. Keith didn’t show for the concert I attended. (I am old enough to have been around for the first round of these guys. Some of them age well. Others try to pretend that they are still 22. Daltry was like that and it was embarrassing.)

          1. I think Roger has aged well. (Definatley better than Mick Jagger lol!) But I do agree with you that it is embarassing when they try to act like they are still in their 20’s. I guess I’m just happy that I even got the chance to see them live. It’s not like I can get excited or even want to see any new bands or musicians these days!

  7. There’a a website:
    where you can morph faces ’til the cows come home. You don’t even need to log on or register.

    Not that they have a cow, but they do have a tiger, a puppy, Gollum, Shrek, a monkey, and the Joker, along with a host of faces of celebrity or notoriety.

    They have a ‘what will their baby look like’ which is patently absurd genetically speaking, but I’m sure people love it.

    I recently morphed the only three James Bond actors they have, and the result looked surprisingly reminiscent of Roger Moore (who is not available in their selection!).

    You can use the faces they supply or import your own. I imported Pope Joey The Rat and morphed him with the Joker and that was disturbingly interesting!

    You can morph up to four faces at a time and then re-morph that result, so you could try to recreate the face shown above if you wished – or improve upon it!

    Some celebrity combinations are truly stunning, others ghastly. And it’s not necessarily the “most beautiful” (however you define that) faces which make the most beautiful morphs; often the less appealing faces can combine in unexpected ways to make an appealing morph. Beauty is in the eye of the beholder, after all.

    (and no, I have no connection with the web site!)

  8. I’m not even sure what the point of this exercise is, for you could do it with anything: art, automobiles, and so on. What you’ll get is something that many people find appealing, but some people dislike.

    … and you don’t see the point of that?

    Man, I’d love to be so sheltered from consumerist culture that my mind didn’t immediately hone in on exactly what the point of exercises like this is. This is a corporate “content provider”‘s dream come true.

  9. I remember a TV program from years ago, where some researcher was claiming that the perfectly proportioned human face could be mathematically predicted based on the golden ratio/Fibonacci sequence. Google “golden ratio mask” to see what I mean.

    1. I’ve PhiMatrix software on my PC and the portrait rectangle grid settles nicely on the level of the eyes, end of nose, mouth, and vertically around the nose too.

      Of course I may be selecting the grid to fit, but Phi, the golden ratio, does turn up a lot in nature and people seem to find the ratio satisfying..

        1. Heh, I should have made clear that I thought it sounded a bit like numerology – probably most things in nature can be made to fit some pair of values or other from the Fibonacci sequence with a bit of squinting.

  10. The thing is that people like to listen to music they know, but if they don’t expand their horizon, the music gets stale.

    Taste tests have proven over and over that new coke tasted better than pepsi and coke, but it still failed because the pop market isn’t about taste.

    The pop music market isn’t really about how good the music sounds either. A lot of it is image.

  11. What I found most fascinating about this is that, since they really did start completely from scratch with pure random sine waves (I’ve heard other attempts to employ genetic algorithms to music composition that were somewhat less pure), this experiment starkly demonstrates a huge limitation of natural selection: the fact that the fitness landscape contains more desirable and yet inaccessible areas, because you would have to cross a very long valley of reduced fitness before emerging out to the other peak.

    Notice that there aren’t really any chord changes. There are hints here and there of a I-IV progression, but that’s easy because both chords rely very heavily on the same tonic note. For the most part, this algorithm can only produce a loop which vamps on a single chord. What it came up with is pretty cool, but you aren’t going to get simple chord progressions here, because in order to “evolve” that it would have to start including notes that weren’t in the tonic chord — and those mutations would be selected against, at least initially.

    I actually thought that the final result was pretty cool given that constraint, i.e. that it could only be something which vamped on a single chord and was composed of sine waves. It was catchy and had a lot going on, at least within that framework. But as Jerry says, you really need a “macromutation” of sorts to get anywhere with music. To even just do something as simple as changing chords, you have to go somewhere that, if evolved incrementally, would sound jarring at first.

    1. I also thought your post was insightful. I think you’re right given the system as it’s set up, especially since there’s just one breeding population.

      I think it could perhaps be overcome without resorting to macromutations by establishing varied ecological niches where different and more specialized selection pressures apply. (The niches could be based on user responses to a survey about musical tastes, perhaps, or by finding users whose responses statistically cluster together.) There’s no way I could predict the result, but I think the extra genetic variation plus occasional outbreeding might allow some of those fitness valleys to be skirted around…

  12. Seems to me that this is a reasonable analogy to natural selection that even the most stupid creationist can grok. Start with a population containing heritable variation, apply selection pressure, and crank through the generations. Out of this arises something quite different from the starting point, in a surprisingly few generations.

    By no means is this analogy perfect, but it’s certainly better than no analogy at all as an example of evolution that even the deadest of brains can comprehend.

    1. The problem is that they’ll tell you that this is clearly an example of Intelligent Design because look – people were the ones doing the selecting! And it started with sounds already! You didn’t go from bicycles to music – you started with sounds – so this is micromutation not macromutation. And so on.

      This is actually a pretty good example of artificial selection – but then so are the various dog and horse breeds and the existence of grains in their modern forms. And none of them convince them, so why should this?

  13. All has to do with dating and mating, shoving ones genes into the future! Of course.

    Wonder if it’s not most attractive but really less offensive, ie our brains are designed to quickly filter noxious stimuli.

      1. Ohhh – the list! Any list that does not have Paul Newman, Cary Grant or Brad Pitt as number 1 is of course invalid. But yeah – Tony Curtis is missing. At least they have Alain Delon on the list.

        O we were talking about music?!?

        1. As I was bicycling to my pianolesson, I amused myself with finding more handsome men for the list, and it dawned on me, that the list is absurdly white. Where’s Denzel Washington (swoon), Sharukh Khan and Sidney Poitier?!?

  14. Lots of problems here.

    Certainly selection occurs on compositional practices from generation to generation.

    But what they are doing is not composition, and the results are likewise not compositions.

    A composition is what results when one creates organic and coherent relationships with the pitch/rhythmic content. There must be an overarching logic.

    Stringing together a bunch of noises people liked is not composition.

    This also brings up the question of how to define “perfect”.

    It seems to me that a perception of “perfection” results more from being able to discern the (long-range) logic of a piece than from purely visceral reactions to superficial characteristics like timbre, dynamics, motives, etc.

    When I am moved by, say, the appearance of a chromatic pitch, it’s not simply because it’s chromatic. It’s because (if it’s been employed intelligently) it carries certain harmonic or contrapuntal implications, and it’s satisfying to hear these implications fulfilled, often in a clever or roundabout way, if the composer was good. So my own reactions to good music are rooted more in a realization of and admiration for the logical/architectural capabilities of the composer, and not reactions to raw pitch or timbre itself. I think the more time one spends with music, the more one listens to it this way. A loud brass section or cymbal crashes still affect me, but I’d never conclude from the effect of such superficialities that the composition was good.

    1. Hmm. I guess I should complete this thought.

      My point being: I think it’s necessary, in music, to have a legitimate human creator. A process which, even though it involved people, was more mindless than mindful, can’t result in something really creatively valuable.

      1. Haydn, as I recall, remarked that being stuck miles from anywhere on the Esterhazy estate, forced him to be original.

    1. BTW, I’m glad to see my alma mater is still doing good work. I visited the campus this summer but the door to my old lab was locked which is the story of my life.

  15. Daltrey could really wield a microphone too, couldn’t he?

    The loops remind me of an orchestral piece by Steve Reich- Variations_for_Winds,_Strings_and_Keyboards.

    And yes, the face portrayed is “too perfect”. It’s rather hard to imagine a face as that frowning or looking angry. Visually, it suggests nothing to the beholder other than “I’m collection of perfect features”. But add perhaps a wrinkle or two…
    I’d be more drawn to a face with Tilda Swinton’s features myself.

  16. Not to mention that the process described is artificial selection, not natural selection.

    As for that other group – gee, if they use ‘Photoshop’ to make a Frankenstein creation of A-list celebrities and spend many hours diddling the results they get something that doesn’t look absolutely horrible – who would ever have guessed it?

  17. I’m going to do a plug here while we’re on the subject of music. If you like horror movies, listen to the Creeping Cruds, a hard-rockin’ band hailing from the darkest, scariest corner of Nashville, Tennessee. Their first album, which I bought yesterday, is “The Incredibly Strange People Who Stopped Living and Became…”

    And under that: “The Creeping Cruds”.

  18. Apparently the “perfect pop song” is some shitty preprogrammed melody from a hand-held Casio keyboard from 1985. But it also reminded me of, as fullyladenswallow noted, Steve Reich…Steve Reich as played on a shitty hand-held Casio keyboard from 1985.

    Any resemblance to Baba O’Riley’s sequencer bit is incidental.

  19. “But to my ear the result, at least as heard on the Telegraph site, is bland and uninteresting.”

    it could certainly benefit from some guitar/drums, etc. however, the final generation of the clip i heard had some amazing pop bliss in it.

  20. There is an old rule of thumb. When an article has a question as its title, the answer to that question is always “no.”

  21. The face reminds of Natalie Portman, and I regard her as quite beautiful. Regardless, we already have Sophia Loren, so why bother trying to create some other gold standard?

    1. Yeah, I can see bit of her in there. But I immediately thought of Keira Knightley when I first saw it. By the way, I disagree with Jerry: that face is drop dead gorgeous. Any woman with such a face would be staggeringly beautiful. But as Solzhenitsyn wrote, beauty is seldom beneficial to the development of one’s character.

  22. Totally agree that macromutation music like Sgt. Pepper’s is always better.

    The Beatles’ albums before “Rubber Soul” were delightfully good but almost entirely derivative from earlier music like Chuck Berry, Elvis, et al. They started to develop their own distinct sound with Rubber Soul and it really came out full in “Sgt. Pepper”. Do any list of the Beatle’s best-known songs and the overwhelming majority are from their last 8 (out of 13) albums.

    1. I remember, in high school when Jimi Hendrix came out with “Are You Experienced”, thinking that his music was the first I’d ever heard that was totally novel. With a bit more maturity I realized that he, too, built on the work of earlier musicians as all musicians do. The great ones contribute a bit more inventiveness but all music is derivative more or less.

      1. Indeed.

        Consider that all pop/rock, including the music of the important artists mentioned in these threads, is tonal. If you’re looking for novelty, you’d do better looking at “art music” from the mid to late 20th century. Schönberg, Crumb, Babbit, Xenakis, Ligeti…

        Of course, one shouldn’t equivocate between novelty and quality.

  23. If you went on beautiful voices, Armand Leroi himself has a great voice. He did a very good programme on the Beeb about Darwin in 2009 also one about Aristotle.

  24. Won’t work. You’ll just get regression to the mean.

    The very best songs and performances are ones that have something distinctive (and hence unusual) about them. Just the most recent example – I’ve just discovered Linda Ronstadt’s ‘Long Long Time’ (how could I have missed it for so long?) – her delivery is absolutely not the way the ‘average’ singer would sing it. But she owns that song for precisely that reason.

    Oh, and incidentally, contra the second point about appearances, Ms Ronstadt when she was young was a ‘hottie’ (as one Youtube commenter put it) and she looked nothing remotely like the composite beauty at the head of this page.

    Vive la difference!

    1. Forgot to say – obviously if you start off with something much worse than the mean (which presumably the experimenters’ collection of random noises was), then regression to the mean will cause later versions to sound better. But they’ll never get to be better than average. I think.

      It takes individual talent to create something really good.

  25. The result isn’t a ‘song’ or even a composition. It is just a loop. There is no harmonic progression, no real melody. A real composer could take this and make a song by changing it, adding a contrasting section, etc. But I don’t think it will ever result in a finished song using this method. Just a bunch of pretty noise.

  26. I must be the only metal head on this blog. I love and respect a lot of the bands that are mentioned here (the Beatles, The Who, Clapton, Hendrix, etc.) but there’s just not enough anger and aggression in a lot of that music to suit me. I grew up on Metallica, Megadeth, Slayer, Pantera, and bands like that (got tickets to see Metallica for the fourth time this summer). I’m also a huge Rush fan I’m happy to say.
    Expand your horizons a bit and throw on an Iron Maiden album or something from Mastodon. Come on over to the dark side, Jerry. But seriously, the great thing about metal is that it doesn’t run up against a lot of the limits and boundaries that confine more popular music. Metal bands are extremely technically proficient on their instruments, have longer songs, strange meters and several tempo changes in each song, tons of dissonance and unusual melodies, unusual scale choices, and lots of other interesting features that are not found in other musical styles. I would venture that the size and extent of the musical canvas used by many metal artists is rivaled only by jazz and classical styles. Anyway, I just wanted to throw out a seldom heard recommendation for a genre of music that is often over-looked.

  27. If you use natural selection to create the perfect anything ur doin it rong.

    You’re maximizing the traits you want period. That doesn’t make it perfect. According to the article, the guys are manually selecting the bits they like. That’s not a measure of perfection but of their tastes, and that’s what the guys say in their abstract – they investigate the role of consumer selection. No perfection is mentioned. Bad journalism.

    Does anyone else prefer their music a little bit alive, with its “flaws” and its flavor that makes you notice it is being played by a human? (Ok it’s probably the same thing Jerry means when he says bland.)

  28. You’re right. Technically perfect, but lacking in character. I’ve never understood the allure of those who look like they were produced in a factory somewhere.

    The most beautiful women I’ve ever seen always seem to have some “Persian flaw” (though not quite how the Persians meant it). A crooked smile. A scar on her cheek. Signs of a life lived.

    It’s what I find most interesting.

Leave a Comment

Your email address will not be published. Required fields are marked *