Other Ten Percent 2/19/14

Feb 19 2014

First off: Yay! Nikki fixed the blog so I’ve spent the day reposting all the stuff I’ve done so far this year to the archives and I once again have some proof I’ve been doing this shit for coming up on five years now pretty regularly and so my enthusiasm for OTP has returned somewhat. It’s still muted a bit by the fact that we’re moving in like 2 weeks and our lives are crazy as woah but it’s actually seeming like a thing that’s worth the effort when I have any additional effort to expend.

– – – – –

But let’s talk about Twitch Plays Pokemon though because you’ve been watching it for 30 seconds now and honestly it seems to have broken the popularity barrier where even non gaming nerds are talking about it. If you have not heard about it the general bit is this: a guy is running a game of pokemon on Twitch which has a chat channel running next to the video stream and thanks to a chat bot he’s programmed in any commands that are yelled in chat are executed by the game which means however many people are in the chat channel (and it’s currently hovering around 80,000 people) are all effectively playing the game at once.

Now there are interesting things to say about trying to organize 80,000 people so they can perform simple tasks but I don’t want to talk about that both because it isn’t really what interests me here and because it’s already been written about really eloquently all over the place at Kotaku and Ars Technica and Polygon and everybody that covers this stuff.

I wanted to talk instead about Twitch Plays Pokemon as it relates to Artificial Intelligence because it was the first thing I thought of when I started watching the stream. To be clear: this thing isn’t an artificial intelligence it’s just a system for fulfilling a ton of user commands at once but the output of that system, what it actually looks like for a computer to try and obey 80,000 Pokemon players at once, looks a lot like other Machine intelligences that have been popping up in the real world and in fiction a LOT lately because it’s clearly an intelligent system able to perform tasks (albeit haltingly and stupidly) but it’s “decision making process” is so foreign to the one humans use I’m not sure it qualifies as having one. Twitch Plays Pokemon is made up of humans but it doesn’t think like a human or operate like one at all.

Overnight last night, after the noise from 80,000 people trying to command the game at once became too much to progress through a simple maze, the stream’s creator shut it down for a few minutes and brought it back with a new mechanic. In addition to being able to input button  commands users could now vote between the original “anarchy” mode (which is a weird name considering it was basically an expression of constant pure democracy) and a democracy mode where the machine gathers up all the votes for 30 seconds and then performs the most popular command (again this is actually more accurately a description of republican government but whatever I’m being a pedantic dick). From what I gather the change is unpopular because it’s ruined the “purity” of the original experiment but to me it’s made the whole thing even more fascinating. The Gestalt intelligence that runs Twitch Plays Pokemon can now choose between two decision making modes, one that moves more quickly but seemingly randomly and another for slower but more accurate control So the constant guessing is not just about what the intelligence will decide to do but how it will choose to make that decision.

One of the things that’s coming up a lot here in the year of our lord 2014 is that the amount of data that just pours through a machine intelligence means it’s going to end up thinking differently just by necessity. Human beings are really bad at managing large data sets. We can really hold only so many things in our minds at once so even if we built a hard AI that thought like a human we’d have to figure out (or more likely it would have to figure out for itself) how to make it NOT think like a human if we wanted it to say, organize the amount of data Google has for us. Which means one way or another if we’re going to build complex decision making machines that sift through shit tons of data (and we are because we already have) we’re going to have to learn how to deal with thoroughly alien intelligences in the next decade or two whose systems of logic are totally opaque to us but which we’ll still have to interact with faily. All of which makes watching Twich Plays Pokemon a weirdly effective preparation for the future.


No responses yet

Leave a Reply