Grand Theft Adaptation

Grand Theft Adaptation
Thinking Space Invaders

Joe Blancato | 16 Aug 2005 12:03
Grand Theft Adaptation - RSS 2.0

Adaptation is what ensures survival. Everything adapts to its environment, or it dies. In an environment as socially Darwinian as the free games market, this is especially true. This truth is why many sequels are bad; the games do not change, even though we, the players, have. Many of the big companies, the industry's version of the great white shark, pump out high budget sequels so efficient at being fun, we're still not tired of them. But many developers are working on techniques to make our games adapt to us, and they're using techniques billions of years old.

Genetic algorithms, they're called. They're a relatively simple process: Send out a group of subjects, give them a goal, and the ones who achieve it survive to pass their methods onto the next generation of subjects. It's an incredibly popular design method in real-time strategy games and first-person shooters. Bots learn to adapt to your habits and theoretically "grow" to be able to predict what you do, zig when you zag, and turn you into a rag doll.

But we're not quite there yet. Since the concept of genetic programming was dreamed up in the mid 1960s, innovation has been staggering across the board. Still, computers are pretty lacking when it comes to problem solving on a broad scale. They do not have the distinctly biological ability to, well, adapt on the fly without any programming help from a human. But as a species, they're only 50 years old; we're about 500,000. They have a distinctive future edge, though: intelligent design. It's a topic that's up for debate in the real world, but digitally, everything has a goal.

Take Endless Fire for example. Tom Betts' creation is a simple, psychedelic version of Space Invaders with a Pi-like soundtrack that ranks in at about 0.5 on the Fun Scale, when I'm sober. But somehow, I can't stop loading the game whenever I have five minutes.

Betts built the game on the basis that your enemies would "learn" as they died, recording how you kill them and bestowing that information to the next generation of attackers. Little aliens prance back and forth to the music as I rhythmically dash in the opposite direction across the screen, firing the spread shot, taking out entire groups of invaders before they ever make it half-way down the screen. That lasts about five minutes. Suddenly, in whips a bogey from the right side of the screen, firing a targeted laser and shooting a guided missile at me, simultaneously. Luckily, I bring up my Smart Bombs, a glorified shield, in time, nullifying the attack. I roll right and take out the alien with a precision blast, head on. I start humming Highway to the Danger Zone and make everyone call me Maverick.

Betts' aliens didn't bring their A-game. No, the little guys tried their best, but they encountered a vastly superior intelligence in my primate brain. But again, if Betts' code is solid and I had enough time on my hands, we might reach a point where those things could beat the hell out of me. Chances are, I'll reign supreme for some time; the game resets itself every time it's restarted.

Enter Tron.

The Dynamical and Evolutionary Machine Organization (DEMO) put together a Java version of Tron which attempted to learn from every person to ever play it online. It's currently in its 830th "generation," and wins roughly 93 percent of the time. Intrigued, I hopped into DEMO's Tron for a brief encounter with a blue-toned AI opponent. Fifteen minutes later, the score was 67 to two, in the machine's favor.

One match in particular stuck out. I was zooming along, leaving red vapor trails behind me, in a completely different quadrant than Blue. Somehow, I managed to juke my way toward It, cut It off. It was boxed in, and I had more room; just survive, stupid - you have more real estate. I was watching it wither on the vine, zipping back and forth looking for a way out of its grim fate. In fact, I was so engrossed by what it was doing, I crashed into a wall of my own creation and lost.

This thing has played enough games, evolved enough times, to rank as the digital Bobby Fischer of light cycles. The massive sample size is proof that we can be outsmarted by our own drive to create something similar to us. It's only a matter of time before philosophical debate is replaced by real world litigation. It's only a matter of time before a program calls its author "Daddy."

Joe Blancato is a Contributing Editor for The Escapist Magazine, in addition to being the Founder of

JOHNATHAN GREY CARTER | 7 Dec 2010 15:00
JULIANNE CAPPS | 16 Aug 2005 12:00
BOB "MOVIEBOB" CHIPMAN | 17 Dec 2010 17:00
ANDY CHALK | 3 May 2007 19:07

Comments on