Geometry

The mic/speaker assembly are in the middle of the window but the window has substantial size relative to the dimensions of the room and the distance the listener is from the device. What this means is that you have a multipath problem; the original sound can come through the whole window (even the wall!) but the canceling sound can only eminate from the speaker.

If you draw a straight line from the original noise source, through the device and from there it happens to head towards your ears, this can work. If not things will get jumbled up and it might cancel or it might actually reinforce the noise, rather than cancelling it.

http://en.wikipedia.org/wiki/Loudspeaker#Directivity

Physics

A speaker is a device that works as a derivative of the original signal. That means it needs some amount of "advance notice" of the sound in order to be able to create the proper pressure wave to cancel out said sound. The "advance notice" is usually done by means of some distance between the microphone and the speaker but the device is very thin. Worse, the amount of "advance notice" depends on the frequency of the sound which generally makes it an intractable problem to solve for the transient case.

http://www.customanalogue.com/elsinore/elsinore_19.htm

So what people do is focus on periodic noises like hum, hiss, buzz, rumble, what-have-you. In this case you don't need to make sure that any one peak has it's inverse reproduced perfectly, you only have to ensure that ONE peak is cancelled out by SOME OTHER out of phase peak. If it takes a couple of cycles for the computer to catch on, no big deal that's just a bit of start-up noise. Bose headphones don't IMMEDIATELY make the plane quieter, they just do it so fast that you don't notice the amount of time it takes for the system to warm up. It seems instant even though it's not.

But what this means is that a bunch of the sounds being cancelled out probably wouldn't be cancelled out quite so nicely as a lot of them are transients like the bottles breaking or the construction site or the park.

I'd really love to be wrong about this. There's a joke that goes "An engineer will tell you it's impossible until you show him how to do it. Then he'll say it's obvious." It's entirely possible that this guy has figured something out that the rest of us haven't. But I will remain skeptical until I can actually put my hands on one and benchmark it for myself