Russ R. has an interesting podcast this week with Pedro Domingos on Machine Learning. Part of their discussion surrounds how knowledge intensive companies develop algorithms that either send you advertisements or help you choose things to purchase. The most common examples are the Ads that run in Google that are instantly customized to the end user and the “recommended” book selections when you are browsing and buying books on Amazon.
I have long complained about buying books on Amazon, though I have spent thousands of dollars on books there in the last decade, because I can’t “browse the shelves” to see what else is around. Part of my favorite thing about bookstores (and beer stores!) is having the ability to wander around, see if something new or different piques my interest, or to compare what else is on the shelves in the categories I am looking at. There have been moments where I have started down the paths of entirely new disciplines by doing this sort of thing.
The downside of Amazon is that you can’t easily browse the stacks to see what is near the book you are looking at. They have computer algorithms that examine the books you have browsed and the ones you have bought and compares it to thousands of other customers and it makes all kinds of suggestions for things you might like. There are millions of books to possibly choose from, and since you cannot obviously self-browse this massive stack, something needs to help parse it.
I used to think this was good, then bad, and then good.
Compare this to an old bookstore. My former bias was that Amazon was inferior because I couldn’t see the other books, and to be honest, there are more books “in my line of sight” at a bookstore than grab my attention on a computer screen.
But is Amazon really inferior? I used to push back because “a computer” made selections for me, and did not allow me to see what the computer did not pick out. How silly? For sure, the computer did narrow down what came to my eyeballs, but if I start searching on other books, then the computer will start sending me different selections. But ultimately how is this any different than what is happening at the bookstore? Booksellers have to select which books to buy and stock, then where in the store to display them, and then how to display them in those particular location. And since there is obviously a dearth of actual physical space, the bookseller is not able to offer me very much in the way of options. At most, they may have 1,000 possible books that I might ever consider buying.
Is what Amazon does any different? Is the “computer” really “picking” what I look at? Of course not. The algorithms are developed based on what other real human beings actually like and do. And my best judgment is that physical booksellers are doing exactly the same thing. They do need to stay in business, right? So while I might wish that they pulled the little read classics and featured them on shelves for me, or that they offered in depth tracts on arcane subjects, those things just don’t sell. So the motivation of the bookseller does not seem to be very different than Amazon. The same is true for when we see “ratings” and “recommendations” posted in bookstores, and featured shelves – is that really any better than what Amazon offers?
My simple point(s) is that no matter how you shop, there is always going to be some institution that parses what is displayed to you. What styles a brewer likes to brew limits the available beer choices you have when you show up to a brewery. What pens are most popular limits the offerings when you head to the stationary stores. The same is true of media content. There never was a time when you had the world at your fingertips and had an easy time parsing through it. Walk into the Library of Congress and tell me if that experience is the best way to choose what to read. When you have everything available to you, in no particular order, that is not much better than having little available to you – you will quickly develop your own internal algorithm to help you navigate your way through the millions of volumes.
Now, can Amazon do better? I am sure of it. I would like it if I could tweak the algorithm better. Just because I want to buy book X does not mean that my tastes in books are like other people who like book X. We may have lots of different reasons for liking or wanting to read a particular book. With better computing power maybe algorithms could delve the depths of the thousands of books on our wish lists and compare them to purchases and also enable us to edit and discuss the features of works that are important to us. Maybe the algorithm already does that. And the potential for machine algorithms to do better than human algorithms is its advantage. The correct comparison is not whether the machines are going to get it perfect, but whether they do even a little bit better job than the people.
Of course, I am on my way to … the Fairport Library right now.