ISMIR 2008 Student Travel Grant Committee

As a Peace Corps volunteer (PCV), I was one of the few with consistent internet access. For this reason, it fell to me to collect music for the occasional PC soireé. The opportunities to hang out with a bunch of Americans were rare enough that all parties were special, but the best events were those with a little dancing. So, I set about trying to do my service to the partygoers and give them a bit of fun.

Like any good computer scientist, my work began with a systematic contemplation of the problem. I eventually noticed that traffic would come and go on the dancefloor, but only so long as there were people out there… An empty dancefloor can take hours to fill. So, I wanted my algorithm to minimize empties instead of perhaps aiming for a higher average.

Ideas filled my head, and slowly I tracked the pattern to "you have to know your audience." I could pick the songs for general dancability, but what really kept the people on the floor were songs that they knew. Since there were only 150 Americans in the country, and 70 of them were PCVs, I was only sampling from the 55 of the 70 that were well-suited (or learned to be) to the drunken debauchery of a Peace Corps party. I eventually knew the favorite songs of almost every single volunteer in the country in my attempt to get them on the dance floor.

It worked great. If the point of music is to leave people with a memorable evening, I helped a couple parties see the sun come up.

I started to think of how I could take this idea and make it work if the sample population was ten or a hundred thousand. I dreamed of opening a bar in DC where folks could try foods and songs from all over the world. I'd outfit it with RFID tags, touch sensitive pads — big LEDs with slick graphics. They could use these screens to answer questions from the bar. Did I mention the bar would learn how to make the people as happy as possible? The environment would be a reinforcement learning algorithm. Automate as much of it as possible. Set a general theme and clientele, but once you've built up a regular crowd, give more and more control over to the machine. Let people pour their own beers and order drinks from their seats if they don't want to fight to the bar. Let the simple benefits of losing their privacy pay them back for giving it up.

I would let the dancers tell the computer what they wanted to hear. I've frequently been on the dancefloor and wanted to give the DJ a piece of my mind for playing crap. My ideal bar would let them do that. I started to think then about how the computer would decide what to play, and realized many of the questions don't have answers yet. It's an open science and one that we are just starting to have the tools to address. What an awesome experimental setup this bar would be.

If I play a series of mushy love songs will it reduce churn on the dancefloor/bar barrier?

I would think it would make pairings between couples more frequent. Perhaps the young men and women would get a bit braver and talk to the person they've had their eye on, but were perhaps a bit reluctant to approach. I would expect pairs to have a lower likelihood or movement since the party has to come to agreement. The extent to which I analyzed the potential patterns in damned near every situation put a strain on my relationships for a bit.

I want to build that bar. This is why I want to go to ISMIR. Music informatics is the cutting edge of research on the relationship between the codifiable characteristics of sounds we enjoy and human behavior. I'm going to be awash in data: who and where people are and what they do. This is no simulator. You get to run questions on actual groups of people.

They might actually start coming more if you tell them you're studying the science of how to make them happier.

So, I'm in recommender systems, learning to guess what people want. ISMIR will provide me a broader exposure to the field than I have gotten to date — the entirety of my formal MIR training being a self-directed literature survey for a general AI class. There's also the possibility that I will make useful contacts that might help me find a route from my current computer science Ph.D. to one in a school with a solid sociology program. I don't know anyone with a CompSci undegrad who has gone into a sociology doctorate program. I hope to also to talk to people from various schools and find out if this would be a potential path for even the brightest of students, let alone someone with my mottled academic career.)

I am, despite the largess of Sun Microsystems furnishing and paying me handsomely to work on Project Aura this summer, not a man of significant means. Any charitability that you might feel would certainly be appreciated.

Thank you for your time,

Will Holcomb
4 July 2008