Brooks (brooksmoses) wrote,
  • Mood:

QWOP, learning, and small amusements.

So, because it came up in conversation earlier this evening, as these things do, Andres introduced me to QWOP.

QWOP is mostly the reduction to triviality racing game: You have a runner, and a 100-yard flat track, and you want to get to the end as fast as possible. There is no competition except yourself, the track, and the clock. The catch: You basically get to play the motor neurons in the spinal cord. You have four control keys -- Q and W to control the thighs, and O and P to control the calves. (Thus the name.) No further explanation is given, except a reminder that it's not about whether you win or lose....

The result is a lot like the little cars in the various "genetic algorithm" machine-learning demo programs, where a bunch of randomly-generated simulated "cars" are placed on a bumpy track, and most of them flip over or otherwise spectacularly fail to go anywhere, and a few make it a little ways down the track before spectacularly failing. There is much flailing, and flipping over, and generally failing to make forward progress.

I found it very interesting observing myself learning how to play it. The first step was mostly conscious -- what, exactly, do these keys do? (My immediate guesses were wrong. The answer, which I will not spoil in case you want to repeat the experiment, seemed rather clearly the right thing for them to do once I got there, though.) And then there was figuring out what to do with them. Some of it was watching what my subconscious brain came up with and seeing whether it worked or not.

I noticed myself getting very good at the "this has gone pretzel shaped, hit the spacebar to retry" recognition and action. And starting to become good at the "don't hit the spacebar until the runner has actually hit the ground" timing on that (because it doesn't do anything until he has actually finished falling over). I didn't have to think about that one much, and generally I was already hitting the spacebar by the time my conscious brain processed what just happened.

Somewhere in there I managed to run 3.2 yards backwards. I also came close to Andres's best (in the demo) 3.0 yards forward, but didn't beat it.

Also, at some point I had a distinct insight of, "oh, right, that's how running works," at which point I got about 3.5 yards almost immediately. I still had more false starts and flailings and such than I had good starts, and most of the good starts went only a couple of steps before the spectacular failure, but it was something.

Then was a stage of an interesting back-and-forth between unconscious and conscious learning. I had a reasonably good run of 7.3 yards pretty quickly, and then started getting worse again and had trouble getting past 3 or 4 yards. It seemed like what happened was that my unconscious had temporarily figured out something that worked, but then as I started consciously trying to understand why it worked, I was throwing off my timing and also interrupting the pattern and thus getting momentarily stuck at a key moment. And eventually and more laboriously, I started to both understand consciously a little of what was working and not working, and -- a little ahead of the conscious understanding -- I started being able to generate patterns that worked some of the time.

At the end, I was working on reliably getting a good start (mostly by conscious observation) and getting a recoverable one maybe one time in three, and a good one maybe one time in nine; on mostly-consciously developing a sense of what a good pattern looked like once the running got started; and pretty much entirely unconsciously working on getting the timing right to keep that pattern going once I got there.

The conscious/unconscious interplay became even more intricate; I was noticing that most of the recognition of a good pattern was unconscious -- the critical stuff was happening too fast to really do any conscious processing in time to do anything useful -- but I was trying frantically to overlay conscious processing on it, first to understand what had gone wrong, and second to figure out what I needed to be looking at to keep it from happening. Of course, this was also throwing off my game a bit; I kept breaking the subconscious patterns by focusing on different things. But occasionally I got a decent pattern going that didn't fall apart too fast, and didn't interrupt it too quickly with conscious thinking about what was making it work.

My best was a run of maybe 10 steps or so, making 8.5 yards.

It's remarkably addictive for such a simplistic game.

Crossposted from Dreamwidth (original here), with comment count unavailable comments. Comment here or there; comments here will eventually be duplicated to there.
  • Post a new comment


    default userpic

    Your reply will be screened

    Your IP address will be recorded 

  • 1 comment