Wanna get rich? Do this • 7 September 2007 • The SnowBlog
Wanna get rich? Do this
When you want to buy a screwdriver, it's pretty easy to get one that will do the job (-> to tighten/untighten screws). They're standardised. But when you want to buy some fancy new slacks, it's only semi-easy to do. The sizes are standardised, but finding the right size is only half the job (-> to clothe you in style). When you want to buy a CD or a novel, it's not easy at all. Yes, they're standardised so you can be sure the CD will play in your stereo or that the novel is in the correct language, but will you like it? The surest way to find out is to buy it and read/listen to it. And risk wasting your money. If only someone would come up with a better way. Popularity works to a point. If a squillion people like something, maybe you will too. But then again, if that was all there was to it, entertainment would be standardised like screwdrivers.
The next-step idea I've been talking to people about since the mid-Nineties is this: you get a hundred thousand people and ask them what their favourite twenty albums/novels are. Then you look for clumps in that data: people with eighteen or nineteen choices in common. Then you use the variations within a clump to make recommendations. If you and I have nineteen albums in common but you've never heard of my third favourite band, there's a good chance that you'll love them.
There are lots of alterations and tweaks you can make to that idea. Rating all your albums will provide better data than just covering the top twenty; mention the ones you hate too. Perhaps you'll need a bigger sample size. Maybe you'd have to try a few methods of making predictions to find one that worked.
But think of what you might discover. Maybe there'd be crossover between album and novel choices - and perhaps movies too. Maybe instead of infinite variety we'd find that, musically speaking, there are only forty-one types of humans. Or maybe no two people are identical in their tastes - but you can bet we'll find plenty of people who are very, very similar. Sooner or later someone has to make this work and I'm pretty convinced that, with a few trials and experiments, it's viable. There seem to be two main problems.
First, if it does work in, say, the music industry it will threaten most of the intermediate/middleman roles in that sector - and that very much includes all the people who are in charge right now. Most bands can learn to record their own MP3s. Most music fans can download them to their pseudoPods. All you need in the middle is a big website that collates preferences and makes good suggestions, plus a big music library/shop. Without all those middlemen, it becomes economically viable to charge one or two pounds for an album instead of ten or fifteen. That means music lovers get a lot more music for the same outlay. And hopefully they rate it and keep the process going. Music lovers do better, all but the biggest bands do better, and only a few suits lose out. But the suits are the ones with the capital and the lawyers and the friends in government. So that's problem one.
Problem two is that Amazon have a system that tries to do something like this already. Every time I mention this idea to people, that's what they tell me. But Amazon's system doesn't ever seem to work. Not for me, at any rate, or for anyone I've talked to. Every now and then I rate a bunch of books and albums on Amazon and let it update its recommendations accordingly. And I believe I've grumbled about the results before. Amazon's system delights in saying 'Hey, if you quite liked this book in paperback, we bet you'll love it in hardback. Or with a different cover. Or better still, have you thought of buying other books from the authors you already know you like?' Amazon's 'system' almost never makes a helpful recommendation. Most of us have already worked the brilliant notion of buying other books from our favourite authors. But what about authors we've never heard of? What about all those things we don't yet know we want? Well, twice now Amazon has recommended non-obvious albums and I've bought them. I didn't like them, but hey, at least they were imaginative suggestions. I had another look just now and of their 15 recommendations there are only two suggestions that go beyond just 'more by this author/band'. I'm almost tempted to waste some money on them. Though I won't be getting my hopes up.
Which brings me to a final point. Imagine what the system I've described would be like if it actually worked. With the Amazon version I'm fairly sure I'm throwing my money away. But what if they started getting it right? What if they began to inspire confidence? What if spending money on their recommended titles actually had a track record of producing treats instead of flops? Well, I think it's pretty clear what would happen. I'd start spending a lot more money with them. Or whoever else cracks this problem. So come on you garden-shed entrepreneurs, asssemble your database gurus and your stochastics experts and your bayesian-inference wizards and your hundred million in venture capital and make that most exciting of gadgets: a jukebox that plays my favourite song from a band I've never heard of.
Rob
When you want to buy a screwdriver, it's pretty easy to get one that will do the job (-> to tighten/untighten screws). They're standardised. But when you want to buy some fancy new slacks, it's only semi-easy to do. The sizes are standardised, but finding the right size is only half the job (-> to clothe you in style). When you want to buy a CD or a novel, it's not easy at all. Yes, they're standardised so you can be sure the CD will play in your stereo or that the novel is in the correct language, but will you like it? The surest way to find out is to buy it and read/listen to it. And risk wasting your money. If only someone would come up with a better way. Popularity works to a point. If a squillion people like something, maybe you will too. But then again, if that was all there was to it, entertainment would be standardised like screwdrivers.
The next-step idea I've been talking to people about since the mid-Nineties is this: you get a hundred thousand people and ask them what their favourite twenty albums/novels are. Then you look for clumps in that data: people with eighteen or nineteen choices in common. Then you use the variations within a clump to make recommendations. If you and I have nineteen albums in common but you've never heard of my third favourite band, there's a good chance that you'll love them.
There are lots of alterations and tweaks you can make to that idea. Rating all your albums will provide better data than just covering the top twenty; mention the ones you hate too. Perhaps you'll need a bigger sample size. Maybe you'd have to try a few methods of making predictions to find one that worked.
But think of what you might discover. Maybe there'd be crossover between album and novel choices - and perhaps movies too. Maybe instead of infinite variety we'd find that, musically speaking, there are only forty-one types of humans. Or maybe no two people are identical in their tastes - but you can bet we'll find plenty of people who are very, very similar. Sooner or later someone has to make this work and I'm pretty convinced that, with a few trials and experiments, it's viable. There seem to be two main problems.
First, if it does work in, say, the music industry it will threaten most of the intermediate/middleman roles in that sector - and that very much includes all the people who are in charge right now. Most bands can learn to record their own MP3s. Most music fans can download them to their pseudoPods. All you need in the middle is a big website that collates preferences and makes good suggestions, plus a big music library/shop. Without all those middlemen, it becomes economically viable to charge one or two pounds for an album instead of ten or fifteen. That means music lovers get a lot more music for the same outlay. And hopefully they rate it and keep the process going. Music lovers do better, all but the biggest bands do better, and only a few suits lose out. But the suits are the ones with the capital and the lawyers and the friends in government. So that's problem one.
Problem two is that Amazon have a system that tries to do something like this already. Every time I mention this idea to people, that's what they tell me. But Amazon's system doesn't ever seem to work. Not for me, at any rate, or for anyone I've talked to. Every now and then I rate a bunch of books and albums on Amazon and let it update its recommendations accordingly. And I believe I've grumbled about the results before. Amazon's system delights in saying 'Hey, if you quite liked this book in paperback, we bet you'll love it in hardback. Or with a different cover. Or better still, have you thought of buying other books from the authors you already know you like?' Amazon's 'system' almost never makes a helpful recommendation. Most of us have already worked the brilliant notion of buying other books from our favourite authors. But what about authors we've never heard of? What about all those things we don't yet know we want? Well, twice now Amazon has recommended non-obvious albums and I've bought them. I didn't like them, but hey, at least they were imaginative suggestions. I had another look just now and of their 15 recommendations there are only two suggestions that go beyond just 'more by this author/band'. I'm almost tempted to waste some money on them. Though I won't be getting my hopes up.
Which brings me to a final point. Imagine what the system I've described would be like if it actually worked. With the Amazon version I'm fairly sure I'm throwing my money away. But what if they started getting it right? What if they began to inspire confidence? What if spending money on their recommended titles actually had a track record of producing treats instead of flops? Well, I think it's pretty clear what would happen. I'd start spending a lot more money with them. Or whoever else cracks this problem. So come on you garden-shed entrepreneurs, asssemble your database gurus and your stochastics experts and your bayesian-inference wizards and your hundred million in venture capital and make that most exciting of gadgets: a jukebox that plays my favourite song from a band I've never heard of.