Photos Spare Cycles MythBusters

Anthropomorphic iPod

My post frequency is down, so I'm going to cheat and mine a post from an e-mail thread.

This New York Times has an article on people and their iPods, and more specifically, how people attribute a higher level of intelligence to their iPods than actually exists. For example,

The iPod "knows somehow when I am reaching the end of my reserves, when my motivation is flagging," Mr. Greist insisted. "It hits me up with 'In Da Club,' and then all of a sudden I am in da club."

People also seem to think that the iPod favors certain artists, and point to the fact that the songs by the same artist will frequently play in proximity to one another. Often this artist will be someone the person likes, so they think that the iPod has learned their music tastes.

Personally, I think this viewpoint may be a result of how humans have a hard time comprehending random.

There is a problem that math/CS majors study called the Birthday Paradox, which asks "given N people, what is the probability that 2 have the same birthday?" It only takes 23 people for the probability to reach 50%. When we did this in class it only took ~15 people before we had two of the same birthdays. (Rubin reminds me that birthdays are not actually distributed evenly throughout the year, so the probability of having two people with the same birthday is actually much higher "since people in certain weather areas always seem to get randy around the same time").

This problem has applications to the iPod shuffling problem. Assuming that you had an equal number of songs from 100 different artists, then you would need 12 songs for there to be a 50% probability of at least two songs by the same artist (100 different artists). This doesn't mean that the songs by the same artist are 12 songs apart; it just means within that span of 12 songs there are at least 2 songs by the same artist, which means on average they will be a lot closer than 12 songs apart. If there are only 50 artists, then it only takes 9 songs, and for 200 artists it takes 17 songs.

However, like the Birthday Paradox, these assumptions are unrealistic: there are definitely artists that we have a lot more songs of, and soundtracks also inflate the number of artists. We also, as the article points out, buy more music of the kind we like. Putting this all together, even if the iPod is being completely random, it should be the case that you frequently hear songs by the same artist close together, and that artist will likely be someone you like. Thus, through complete mindless randomness, the iPod has 'learned' all about your preferences.

(I didn't verify any of the math I used in this entry)

Comments (3)

I have had this discussion with people before who were "in the know" and assured me that the iPod tracks what genre you listen to most often and, while working in random, will bring more and more songs from that genre together. Then they went on to discuss how doing this will completely bias an iPod since the songs most heard are picked from a group of most heard songs.

So ... there really is nothing but randomness and coincidence?

kwc Author Profile Page:

The iPod does track what you listen to, however, I personally don't believe that it tries to harness this effectively. My reasoning is simple: it takes too much work. The less work the iPod does, the more battery life it saves, and the more iPods they sell. I'm more likely to believe their director of iPod marketing (quoted in the article), because as an engineer it makes more sense that the iPod isn't trying to be fancy, and if the iPod did have this feature, Apple would try to promote it with as much hyperbole as possible.

In the above entry, though, the point I'm trying to make isn't that the iPod is random; what I am trying to say is that people don't understand randomness very well, and it leads to them believing that random behavior is actually non-random, and vice-versa.

For example, if I were to write a new shuffle feature where the algorithm was "don't pick any song where the artist has already been played, and order the songs by their second-to-last letter," and I then showed people the playlist that was generated, they would say, "wow, that's really random," even though that's not random at all.

To give another example of this, a friend of mine uses this to always win rock, paper, scissor. Her strategy, IIRC, is to chose whatever would have been beaten in the previous round. i.e. if her opponent choses scissors, then the next round she choses paper. This works because people think being "random" means that their choice has to change, so the person will usually choose paper or rock the next round, giving her a tie or a victory.


I am unconvinced that the birthday paradox has but passing relevance to the particular case I was decribing (artist pattern: ABA). Saying that there's a 50% possibility of the same artist in n songs represents a big superset of the case I'm interested in.

Even if there were a 50% possibility of a pattern ABA every 23 songs, why I am I seeing that pattern almost 100% of the time?

I'm still guessing that there may be some combination of poor randomization perhaps with incorrect reliance on the pseudo-random hashing of songs into the ipod's 20-or-so directories. Indeed, that hashing seems to be at the discretion of whatever program you use to upload to the ipod -- perhaps one of the alternate clients is doing something simple-minded or simply incompatible.

Another possibility is that this might be some misguided attempt at battery conservation. Since spinning up the drive is probably far more expensive than seeking, I don't think track ordering would actually help much.

As for battery power needed to track genres, that's a pretty weak counterargument. 1: the power required to actually spin up a platter is far greater than to move electrons around. 2: smart playlists already update in real-time...

Post a comment


related entries.

what is this?

This page contains a single entry from kwc blog posted on August 27, 2004 10:05 AM.

The previous post was A pleasant DMV experience.

The next post is Bull riding (not me).

Current entries can be found on the main page.