Last year LMFAO’s "Party Rock Anthem" was the most downloaded MP3 on iTunes UK. If this isn’t conclusive evidence that our civilisation is in terminal decline, a research team of AI geeks from Bristol University now have quantifiable proof: we’re stuck in a period of unprecedented musical homogeny. Basically, our pop charts have become one giant inbred circle-jerk, with a gene pool smaller than the Royal Family.Their computer algorithm, Score A Hit, can predict how high (or low) a tune will score in pre-computed music charts with an accuracy of 60%. That figure gets higher from the year 2000 onwards. The team identified 23 different universal audio features; qualities like tempo, time signature, loudness and danceability. By assigning each feature a "weight" to signify its importance within the tune, they could break it down to a mathematical equation.They studied 50 years of UK top 40 charts, totting all the top five and bottom 10 hits. This allows the algorithm to rate a new, uncharted tune, by comparing its weights to the top and bottom echelons of the chart.Depressing, yes. Surprising, not really. I got empirical with head researcher and senior A.I. lecturer at Bristol University, Tijl De Bie, to find out more.Noisey: So, is there a way to distinguish if a tune is more predictable than others?
Tijl De Bie: What it measures is essentially how similar a song is to recent past hits and how dissimilar to recent past non-hits. A high score may indicate it's too mainstream and too "formulaically made". But really, I wouldn't want to say that high score means bad song, all it means is that it's more likely to go to the top of the chart than to stay at the bottom.According to your research, music has gotten more predictable since the turn of the millennium compared to the 70s and 80s. Could you elaborate on that?Mostly in the late 70s and a bit into the early 80s, our equation didn't do well at all. On the other hand, in 2000-2010 it was fairly easy. Also in the first half of the 90s it was easier than average. I think this means that music in the times when it was easier to predict hits using our method was more formulaic, to use the same term. There were less surprises, whereas in the late '70s for example, hits were different to other tunes.So, it suggests producers in those periods were looking at what other people were doing?Perhaps, that's one possible explanation.Do you think this trend of predictable music since 2000 might suggest that the major labels have their own algorithms?Perhaps, but I can only guess! I personally wouldn't think so. I think it's more likely that music goes through fast evolving times and then through times when it's more stable. My suspicion is that there will come another time when predicting hits becomes more difficult again. But that's only a gut feelingWith the app, you’re putting this technology in people’s pockets. Are you afraid we’re potentially going to see a deluge of really predictable music?
I don't think it's a risk, for several reasons. First: the accuracy is still limited, second: although it can analyze music, it can't generate it, so it can't be used to make hits. And third: I think it would be self-defeating if everybody started using this. If music would become all the same, people would long for something new, and the equation would start to fail because people would get fed up with the high scoring songs. Essentially, we shouldn't underestimate music listeners. They will never be told what they should like.That’s reassuring. Score A Hit could potentially be a pretty big cash cow. Have you had any contact from the major labels?
Well, we've had quite some interest from various corners, yes. We haven't made any plans yet, but we are looking into options. Remember this really is an academic project, and initially we thought it would always stay that. But given how popular this has become, we are considering if we should do something more with it. We'll see!A friend of mine in the industry, said you simply couldn’t account for that "x factor" in a tune, that indefinable thing that just makes it work.Some AI enthusiasts would say that whatever a human can, sooner or later a machine would be able to do as well. I tend to agree, but I think it will be later rather than sooner. So there is a possibility that for the foreseeable future the x factor in a song is too complex for a machine to quantify, while some humans can do it. But again, I wouldn't know, I think the only way to find out would be to put those humans to the test as well.Do you think something like Score A Hit might replace people in music industry A&R jobs? Is this one step closer to the matrix?
I don't know, it depends on the answer to what we discussed earlier: I don't know how good a human is at this same task. Score A Hit may well beat them, but maybe not. I personally hope and think that there will always be humans in those jobs too, as they will be able to identify new interesting trends. So they may still work when the equation fails (like e.g. in the late 70s). Something like Score A Hit could be a support tool though, like as a first selection, or as a validation. But it all depends on how good Score A Hit is relative to those people.Well, you've made me feel less threatened by the machines.Great! You can always switch them off too.What if they develop an algorithm to predict when I'm going to switch them off and they kill me before I get to the switch?
They can't unless you hook it up to a gun, so let's not go there!Although Tijl seems to have vindicated our sneering attitude towards MTV and radio, I felt that I needed answers from an industry insider.The Notorious RIG was until recently the A&R manager at one of the bigger labels. You’ve definitely heard of them, and if you’re a blazer-and-jeans wearing cheese monger, you’ve probably ‘put your hands in the air’ at BCN in Mega-luf to one of the tunes he’s signed. We ran Tijl’s findings by him, but he didn’t want to burn any bridges, so we’ve given him a clue-laden pseudonym.Noisey: To what extent are hit-predicting algorithms used in the music industry?
The Notorious RIG: Algorithms are good for armchair A&R but there's no substitute for human perspective. Or taste. In the commercial chart domain however, where neither of these things exist, it seems robots are fair game when it comes to predicting the penchant of the people.My own experience of using such methods is more as an evaluation tool for records that have already been made; a shit-test, so to speak. I’ve never personally seen any such witchcraft in action in the recording process; however, the mind of a producer is perfectly capable of running its own algorithms to fathom the workings of its audience’s aural tastebuds. Even more so if they also play out as a DJ, such as David Guetta, and get to gauge first-hand the impact of their music on their customers, I mean, fans.How much of chart success is down to the actual music compared to things like marketing?
The music is very important, as it often is the marketing. People hear previews on radio, clubs, YouTube and soundtracking commercials advertising their favourite jeans, and the music is designed to stay in their heads. Marketing is merely the Trojan horse, in my opinion it’s about 25% marketing and 75% music. But in pop music, the actual music itself also contains about 75% marketing content. Does that make sense?Totes. According to Score A Hit, chart music has become a lot more predictable in the past 10 years than it was in the 70s and 80s. Why?
Pop music is becoming more predictable because it has carved itself out as an actual genre rather than a statistic group. Popular music, to use its unabbreviated form, has certain structures, hooks, melodic progressions and vocal content that are easy to digest by the masses. This is something that the major labels have perfected into a science over the last thirty years. The sneaky bastards know that "brick wall mastering" a track will make it sound louder on radio, repeating certain phrases in a certain syncopated phrasing will trigger the brain’s reward centres. they also know that the "danceability" of a record is now an important factor due to heavy club promotion, so they have ways of making sure their tracks make you want to cut up a rug or bust a robot.A more realistic take on the dichotomy between the 70s/80s and 90s/00s is that TV and Radio programs directly influence the charts by their choice of which artists to feature on their shows. This choice back in the 70s and 80s was often influenced by practices such as "payola", basically large of amounts of top shelf rack (see cocaine) in exchange for airtime. This apparently doesn’t happen anymore as DJs can now afford to support their habit.Ha! So do you see this continuing or has something eventually got to give?
The internet has done some damage to the major label framework. They are no longer the gatekeepers. The people are empowered by their own free will and are rejecting their media-fed pseudo-reality and replacing it with their own iPod playlists. This will result in many charts instead of a few central charts and they will spawn and multiply and become diverse and interesting.A new dawn will arise and sub-genre after sub-genre shall resound from the earphones of the people. Music needs a messiah. A Julian Assange, Charlie Sheen or John Connor, sent back from the future to set the record(s) straight. The world of entertainment awaits the Anti-Bieber.Inspirational. So, to what extent can the music industry tell people what to like?
To the same extent that other industries can tell you what to eat, wear, drink and so on. At the end of the day, the music industry is a vehicle for a product. That product, like any other will be manufactured, packaged, marketed and sold.However, even if Simon Cowell were to subliminally force his way into your home and market a psychologically tuned algorithmic hotdog straight into your mouth, it’s always your choice whether to swallow it. I’m certain I wouldn’t. It’s nothing against Simon Cowell, mental intrusion or even algorithms. I just don’t like hotdogs.Right, keep your hot dog out of my ear, pal.
Advertisement
Tijl De Bie: What it measures is essentially how similar a song is to recent past hits and how dissimilar to recent past non-hits. A high score may indicate it's too mainstream and too "formulaically made". But really, I wouldn't want to say that high score means bad song, all it means is that it's more likely to go to the top of the chart than to stay at the bottom.According to your research, music has gotten more predictable since the turn of the millennium compared to the 70s and 80s. Could you elaborate on that?
Advertisement
I don't think it's a risk, for several reasons. First: the accuracy is still limited, second: although it can analyze music, it can't generate it, so it can't be used to make hits. And third: I think it would be self-defeating if everybody started using this. If music would become all the same, people would long for something new, and the equation would start to fail because people would get fed up with the high scoring songs. Essentially, we shouldn't underestimate music listeners. They will never be told what they should like.
Advertisement
Well, we've had quite some interest from various corners, yes. We haven't made any plans yet, but we are looking into options. Remember this really is an academic project, and initially we thought it would always stay that. But given how popular this has become, we are considering if we should do something more with it. We'll see!A friend of mine in the industry, said you simply couldn’t account for that "x factor" in a tune, that indefinable thing that just makes it work.Some AI enthusiasts would say that whatever a human can, sooner or later a machine would be able to do as well. I tend to agree, but I think it will be later rather than sooner. So there is a possibility that for the foreseeable future the x factor in a song is too complex for a machine to quantify, while some humans can do it. But again, I wouldn't know, I think the only way to find out would be to put those humans to the test as well.Do you think something like Score A Hit might replace people in music industry A&R jobs? Is this one step closer to the matrix?
I don't know, it depends on the answer to what we discussed earlier: I don't know how good a human is at this same task. Score A Hit may well beat them, but maybe not. I personally hope and think that there will always be humans in those jobs too, as they will be able to identify new interesting trends. So they may still work when the equation fails (like e.g. in the late 70s). Something like Score A Hit could be a support tool though, like as a first selection, or as a validation. But it all depends on how good Score A Hit is relative to those people.
Advertisement
They can't unless you hook it up to a gun, so let's not go there!Although Tijl seems to have vindicated our sneering attitude towards MTV and radio, I felt that I needed answers from an industry insider.The Notorious RIG was until recently the A&R manager at one of the bigger labels. You’ve definitely heard of them, and if you’re a blazer-and-jeans wearing cheese monger, you’ve probably ‘put your hands in the air’ at BCN in Mega-luf to one of the tunes he’s signed. We ran Tijl’s findings by him, but he didn’t want to burn any bridges, so we’ve given him a clue-laden pseudonym.Noisey: To what extent are hit-predicting algorithms used in the music industry?
The Notorious RIG: Algorithms are good for armchair A&R but there's no substitute for human perspective. Or taste. In the commercial chart domain however, where neither of these things exist, it seems robots are fair game when it comes to predicting the penchant of the people.My own experience of using such methods is more as an evaluation tool for records that have already been made; a shit-test, so to speak. I’ve never personally seen any such witchcraft in action in the recording process; however, the mind of a producer is perfectly capable of running its own algorithms to fathom the workings of its audience’s aural tastebuds. Even more so if they also play out as a DJ, such as David Guetta, and get to gauge first-hand the impact of their music on their customers, I mean, fans.
Advertisement
The music is very important, as it often is the marketing. People hear previews on radio, clubs, YouTube and soundtracking commercials advertising their favourite jeans, and the music is designed to stay in their heads. Marketing is merely the Trojan horse, in my opinion it’s about 25% marketing and 75% music. But in pop music, the actual music itself also contains about 75% marketing content. Does that make sense?Totes. According to Score A Hit, chart music has become a lot more predictable in the past 10 years than it was in the 70s and 80s. Why?
Pop music is becoming more predictable because it has carved itself out as an actual genre rather than a statistic group. Popular music, to use its unabbreviated form, has certain structures, hooks, melodic progressions and vocal content that are easy to digest by the masses. This is something that the major labels have perfected into a science over the last thirty years. The sneaky bastards know that "brick wall mastering" a track will make it sound louder on radio, repeating certain phrases in a certain syncopated phrasing will trigger the brain’s reward centres. they also know that the "danceability" of a record is now an important factor due to heavy club promotion, so they have ways of making sure their tracks make you want to cut up a rug or bust a robot.
Advertisement
The internet has done some damage to the major label framework. They are no longer the gatekeepers. The people are empowered by their own free will and are rejecting their media-fed pseudo-reality and replacing it with their own iPod playlists. This will result in many charts instead of a few central charts and they will spawn and multiply and become diverse and interesting.A new dawn will arise and sub-genre after sub-genre shall resound from the earphones of the people. Music needs a messiah. A Julian Assange, Charlie Sheen or John Connor, sent back from the future to set the record(s) straight. The world of entertainment awaits the Anti-Bieber.Inspirational. So, to what extent can the music industry tell people what to like?
To the same extent that other industries can tell you what to eat, wear, drink and so on. At the end of the day, the music industry is a vehicle for a product. That product, like any other will be manufactured, packaged, marketed and sold.However, even if Simon Cowell were to subliminally force his way into your home and market a psychologically tuned algorithmic hotdog straight into your mouth, it’s always your choice whether to swallow it. I’m certain I wouldn’t. It’s nothing against Simon Cowell, mental intrusion or even algorithms. I just don’t like hotdogs.Right, keep your hot dog out of my ear, pal.