Thursday, September 21, 2006

Random is subjective

I read a cool article on the randomness of numbers.

Wall Street Journal - How Random Is iPod's Shuffle Feature?


The concept and theory of truly random numbers is very interesting. I enjoyed the article. I danced nicely along that boundary between interesting and too geeky. There are certainly a lot of random number nerds out there. Amazing.

I dig the math behind it, but I don't feel like it addresses the real issue.

My real issue is with my iPod's random shuffle not seeming random.

The key word there is "seeming." The problem being that a human's perception of what is playing is truly the most important thing when talking about a user's experience. It really isn't pertinent at all if an iPod plays mathematically random. What matters is the customers perception, and thereby enjoyment of the random.

I believe what iPod users really want is a weighted shuffle, that plays songs with certain definitively non-random caveats:

1. It should be more likely to play songs it hasn't played, than ones it has played frequently. For instance, I have just over 1,000 songs on a "My Favorites" play list on my iPod. I've owned it around 18 months, and I play that play list a LOT. I listen to it for hours on end at work, and for my two hours of commuting time 5 days a week.

As I look at my list right now, there are 15 out of 1058 songs that have NEVER played. That violates the rule of true randomness, and also frustrates me as a listener. What exactly is wrong with Ramshackle by Beck that my iPod deems it unworthy of play? Conversely what does it love so much about Feeling A Moment by Feeder that caused it to play it 120 times?

I believe a good algorithm would allow for all songs to be played over time, in a weighting system that favored songs below the median play count and disfavored those over. Something like “no song’s play count should ever be X more than any other.”

2. It should favor playing songs by different artists instead of the same artist consecutively. I can’t tell you how many times I’ve heard a 3-pack of one artist on my iPod. Often, two of those songs are from the same album. It should give a temporary count to all the songs form a played artist that will expire after a few songs to prevent this from happening.

3. It should learn what songs you listen to and which you skip. Even listening to my favorites, I skip some songs. There are some in there that quite honestly I don’t like as much as I did when I added them, and a couple that quite frankly I thought were some other songs. Being that the iPod doesn’t allow you to do much of anything until you get to iTunes, I can’t de-list these songs while I’m in the car. It has little to do with randomness, but it sure would be a nice feature if it created a list of songs you’ve skipped several times, in the first 15-20 seconds, and have it ask once you got back to iTunes if you want to get rid of them.

All in all, my point is this: users don’t want true mathematical random. They want perceived random. They want to hear all the music from a large playlist eventually, with the fewest repeats, and less multiple-song, single-artist blocks.

That’s what MP3 player manufacturers should really be doing, instead of looking at noise from a lens cap.