Dazed on a haystack
Bounty can be a blight
My Instagram timeline is punctuated with ads that would have seemed bizarre to my twentieth-century music-loving saving-up-money-for-weeks-to-buy-physical-media self: Musicians offer to send me CDs of their professionally recorded and produced and packaged songs for free, if I’ll just tell them where I live. Often, their ads either play a clip of their music in the background or open by playing one in the foreground1. Sometimes, the artist making the offer performs one of the tracks in an official music video.
Most of these videos are pretty slick. Most of the tracks aren’t bad. But most of the music is also not so good that I take the trouble even to follow the ad’s link to request a CD. It’s not an original observation—we reached this point years ago—but music is so cheap now that musicians literally cannot give it away.
But I know what I would pay for: I would pay for a service to recommend music to me that I like so much I would hand over my email address to listen to it.
Closing the box
Something not a galaxy far, far away from this has existed for years: Pandora-dot-com. I even used to use this service in the early oughts. Since then, against current fashions, Pandora has pivoted away from its original AI-based model—it hoped to license its proprietary algorithm to music companies—and towards offering a smart streaming service to consumers, and it did so over twenty years ago.
The way it worked when I started using Pandora is pretty much the way it works now, as Wikipedia says:
Listeners can tune into established genre stations, other users' stations or create their own stations based on their musical interests.[41] The user can use thumbs up and thumbs down buttons to declare whether they like a track or not, which determines whether similar songs should be played in the station.[42] A second thumbs down to the same artist will ban that artist from the selected station.[43] A thumbs down immediately skips a song, but the number of times a user can skip tracks is limited unless they are using one of the paid subscription plans, or opts to watch a video ad.[44][45]
The “artificial intelligence” lay in Pandora’s encapsulation of your musical tastes in a clever enough form—Pandora’s aforementioned proprietary algorithm—to make good recommendations. This algorithm encodes the characteristics of millions of tracks to cluster them so that songs and recordings with similar musical properties are placed closer to each other in computational space. If you like one track in a cluster, there’s a good chance you’ll also like one of its Pandoran neighbours.
Pandora didn’t just recommend established artists, but ambitious up-and-coming ones. Not only would their tracks be added to your stream; but, if you liked them, you’d also get the opportunity to do more than Like them: you could add your Pandora-registered email address to their mailing list. And I often did that. To a point when I’d get several emails a week from wannabe stars, usually telling about their tours of small venues in Minnesota that I was never going to visit in a hundred years.
This isn’t why I stopped using Pandora; I stopped because the full service was no longer available in my territory (the UK) for copyright reasons. But a UK-based/Europe-based service that operated in a similar way—so I had a realistic chance of attending the gigs of bands it recommended to me and rewarding their efforts by handing my cash to them directly—would have won my patronage in an instant, in exactly the way that Spotify’s feeble attempts to recommend music to me failed to. (I do, in fact, pay for YouTube Premium, but not for its bundled YouTube Music service, which does at least have the advantage over Spotify of offering quite a few tracks that record labels have suppressed on other streaming services.)
Why is Spotify so rubbish at recommending artists to me? (I use the word “me” intentionally, because I have at least one good friend who is very happy with this aspect of the service.) This leads to my supplementary question: Why is what I believe to be a poor service good enough for most other people?
I want less of More Of The Same
The best commentators can distinguish how they personally feel about something from how other people feel about and distinguish how other people feel about that thing from the truth about that thing. Everyone jokes about how, immediately after they buy an item from Amazon, The Algorithms fill their browsing experience with ads for exactly the same product. Because it’s obvious that, immediately after buying a particular product, we aren’t interested in buying another one, isn’t it?
But there’s a gap between what it is “obvious” in our opinion and obvious in other people’s opinions and what other people whose living depends on it find to be true. Amazon’s living depends on its making sales. Every time I boggle at the current clown-pants layout of the company’s website, I check myself and silently ask: “But what if this shitshow works? What if Amazon devotes millions of person-hours to measuring exactly what level of user-experience chaos its customers will click through to hand over even more of their money to the Bezoids?”
Because what really is undeniable is that Amazon shifts a lot of product. At the risk of my sounding like an evolutionary biologist telling a just-so story about the past emergence of some contemporary human anatomical feature, maybe Amazon shows people more of the same because people do, in fact, want more of the same. And that applies even to products other than the inherently-repeat-purchase lines that Amazon now aggressively pushes “Subscribe & Save” discounts on.
Indeed, Amazon Prime Video takes More-Of-The-Same marketing to an extreme. As you scroll down its home screen, it offers you an entire row of films you have already watched, and invites you to watch them again; it shows an entire row of movies it recommends precisely because people watch them over and over again—and few people would be able to get through that Often Rewatched row without seeing something that they had rewatched; and it shows you an entire row of films that belong to the same subgenre as films you have watched before.
It’s this last kind of More Of The Same that interests me. Or rather doesn’t interest me. Because, yes, there are certain cinematic genres that I am drawn to, but the problem with popular genre content—sorry cineastes, cinema is a popular medium, not an artform2—isn’t that it’s limited by the expectations of its genre; it’s that people who like that genre are prepared to lower their expectations in exchange for conformity with its tropes, tropes they know appeal to them. “Yes, this film is ruined by Steven Seagal’s appalling acting and unnatural physical movement; but I would like to watch ‘Die Hard on decommissioned battleship’.”
Even though I anything but a pop culture snob, I am pretty sure that this is not true of me (and of at least some other people—I’m not that special). I enjoyed Die Hard hugely the first time I saw it. I have rewatched Die Hard many times since. But I have never watched an entire Steven Seagal film. I have no desire to watch a Steven Seagal film. And my feelings about that wouldn’t change however much a Steven Seagal film resembled Die Hard.
The Problem and The Tragedy
This is the problem. Or at least it’s the problem for a minority of people like me. Almost all non-Pandora-style recommendation algorithms are more or less sophisticated versions of:
“You watched films A, B, and C, which you ticked the Thumbs-Up icon to Like. People who also Liked A, B, and C also Liked films X, Y, and Z. Here are X, Y, and, Z as your recommendations. most of these films also happen to belong to what our clustering recommendation algorithm has labelled ‘Genre Alpha’.”
Yes, it might well be true that A, B, and C have fans in common with X, Y, and Z; and it might well be true that X, Y, and Z are of the same genre as A, B, and C. But it is also often the case that A, B and C are good films/albums/products of Genre Alpha and X, Y, and Z are bad films/albums/products of Genre Alpha. I don’t care if other people agree with me about the, admittedly relevant, question of genre; the problem is that they disagree with me about the, more important, question of quality.
This might be the problem, but, for my twenty-first-century music-loving self, it’s not the tragedy. The tragedy is that there isn’t a strong enough incentive for any large corporation to solve this problem, because lots of people, like my Spotify-loving friend, are (almost by definition) happy with this kind of recommendation3—not least of all because there probably isn’t enough good stuff in any given genre for any algorithm to recommend new good stuff they haven’t already seen aren’t already familiar with.
This latter issue is, I suspect, where movies separate off from music. Because, at least for now, feature films still mostly cost large sums of money to produce. Popular music recordings do not. Pandora can work for music because there are lots of good musicians with limited resources who can still make impressively produced recordings. But why would record companies want to promote them when the record companies’ back catalogues can make them more money with less marketing input?
Twentieth-century record companies were just banks whose employees snorted (even) more blow and hired (even) more prostitutes—and who lent money to degenerates by design, rather than in error. Perhaps a better analogy is venture funds, because they would lose silly money on nineteen losing prospects and still make a profit thanks to one massive hitmaker. But, today, not only is the ratio of ambitious music-makers to successful money-makers higher; the wannabes don’t need record-company-size sums either to produce studio-quality recordings or to release it to everyone on Earth with money to give them, for those consumers to play in high fidelity directly into their ears or their homes.
The economic result of this technological revolution is that the artists hustle online themselves, like the record companies would have done once-upon-a-time on their behalf (and at the expense of the artists’ advances), shaking their metaphorical moneymakers out of the metaphorical brothel windows of the Internet.
But no one pimps like a record label
Here’s a perfect example of the kind of talented unsigned artist trying to give me a CD. I watch this sponsored Instagram clip. It’s downtempo; it’s not performed in English; but it’s gripping and cool. I’m intrigued.
Then, I visit the maker’s website and I fear I’ve been redirected by malware to buy a 1990s collection of “Your Secret To Restful Sleep” self-hypnosis cassette tapes. The faux-marble background! The emojis! The filtered headshots! You tempted me into a dark spy-movie subway station with icy, monochrome central-European art-pop; now you’ve got me down there, you smash me in the face with exclamation marks and retina-scorching cerise pink.
This whole layout is a reminder that singer-songwriters should not be doing their own marketing. This website should be being put together by a posh, starving intern, suffocating under the thumb of a cynical, drug-addled music-industry boss. Singer-songwriters should be writing songs and singing them. But God help them doing that fruitfully in a world where streams are the measure of chart success and AI-generated songs can collect tens of thousands of streams.
Even if you don’t think pop music matters, variations of this problem, which had been around for decades everywhere from academia to public policy to journalism to literature to photography to biotech, are getting worse, as the technologies that amplify it have become more powerful. AI as long-refined as Pandora isn’t quite up to the job of stopping this—and it will anyway, increasingly, be in an arms race with other AIs.
We live in an attention economy. And it feels like attention is drifting further from those with talent and closer to those with a talent for self-promotion and armies of machines to fight their campaigns.
I don’t have a solution. But I do have an intuition: The future of curation is human. When recommendations are mostly the result of blind overlaps between your patterns of consumption and those of complete strangers, when most Web traffic is bots, when a large fraction of online reviews are paid-for, the gold standard, the most precious resource, is the active, disinterested discrimination of a real human being whose judgement you trust. We live in a world where computers now have the power to create, but we will probably all be dead before computers have taste.
CODA: I Am Not A Robot
As a public service, here are some other artists I’ve enjoyed that I “discovered” over the past decade for myself—a couple were already well known to others at the time—purely via the Internet, with no real-world word-of-mouth intercession4.
The Allergies
BOVIY
Alma Cook
Dirty Loops
Dodie
Frost*
Ghost Coach
K/DA
The Main Squeeze
Seaforth
Betty Who
The weird ones consist solely of an, often badly recorded, monologue to camera by the songwriter, talking about how “important and personal this collection of songs is” to them. I immediately scroll past these. I don’t know who you are; why should I care about your personal life? If your music is good, I’ll listen to it; if you’ve had a rough year, tell your therapist.
One day, I’ll write about how popular music and cinema can only be Art-with-a-capital-’A’ by accident, and why that’s A Good Thing.
It’s Good Enough for most people in the same sense that they often-questionable sound quality of low-resolution MP3s is Good Enough.
One of the first acts I thought of including in this list, a band I had been introduced to by Pandora, I immediately remembered had gone from raw, soulful inspiration to bland, autotuned hell, so there’s that.