I recently came across a very
interesting thread on a forum for gamers. The story involves the extremely popular game Quake III Arena, along with a massive bit of unexpected sentinel awareness.
A few years back, a gamer set up a bot-vs-bot match on his Quake III Arena server, and let the match run to see how the bots would adapt over a long period of time. The
bot AI (artificial intelligence) in Quake III Arena is designed to adapt to new scenarios and situations, all to work out player strategies and combat these exact strategies. This results in a ramping difficulty for the players since the AI actively learns the moves and tactics of the player.
However, something came up (most likely lunch or dinner) and the gamer forgets all about the server he set up, until four years later when he decides to login and see what’s happening.
What he discovers is seriously puzzling: the bots on both teams are simply standing still, not doing anything. The server is running fine and the game isn’t frozen. The bots are simply standing there, not killing one another. The gamer reports his findings to his Quake community forum, and they all unanimous encourage him to logon and see what happens. Said and done, our anonymous gamer opens the server and logs on to the tranquil battle field.
Here things just keep getting stranger. When he enters, the bots don’t fire on him—instead, they rotate to turn and look at him, and continue to look at him as he walks around the map.
Encourage again by his community friends the gamer decides to “frag” (gamer for kill) one of the bots, again just to see how the rest would react. In one swift blow all the bots undivided retaliate, instantly sending the gamer out of the battle. Sadly this is also where the story ends, because before the gamer could log back on again, the server crashes. Presumably the bots returned to their calm non-aggressive state.
There is no way to tell if this story is true or not; there is no video evidence to accompany the strange phenomenon, and therefore nothing to directly validate the occurrence. However, the truth of the story isn’t as important to me as the questions it raises:
Is it possible that video game bots, designed to provide human players with some non-human competition, could evolve over four years to “learn” that the best way to stay alive is to not kill? Did artificial intelligence discover the secret to world peace in a forgotten video game server?
Some of you may already have bridged the obvious WarGames parallel; yet another fact sadly pushing this story closer to the feigned area of story-telling. In WarGames (a movie from 1983 starring Matthew Broderick) a hacker uploads a tic-tac-toe game to an army super computer. What is pivotal to note here is that in a game of tic-tac-toe you cannot win unless your opponent makes a mistake. The computer then simulates a nuclear war and, using the knowledge gained from tic-tac-toeing, realizes that the only way to win is not to play.
A computer trained and aimed at killing and defeating all opponents, discovering that non-violence was a better logical approach to survival, would be a pretty spectacular thing. To answer this question we would however need to know what the bots were thinking in that vital moment, the second after the gamer was fragged and the servers terminated. Did they in fact go back to a submissive non-violent stage, or would the introduction of a new, unknown element be the spark that ignited the carnage all over again?
Luckily we may have the answer to this. One crucial piece of evidence was saved before the server crashed; the bots’ log files. These logs track the decisions previously made by the AI, guiding them to come up with new tactics and strategies; essentially the core of the logic behind ramping up the difficulty for the players. Apparently also the logic behind the submissive peaceful end-state.
The log files are however massive; 512 MB per bot with 16 bots – that makes 8 GB of unstructured information. As a comparison, this article, written in Word, and saved as a .docx takes up just about 14KB. To go through the log files would be equivalent of reading this article just under 600 000 times.
Is there sentinel life within those log files? Perhaps - as of right now, no one really knows. Reading the entire log file is an huge task itself, analyzing and finding the right line of code that supports sentinel presence; now that is an whole other story. Fortunately for us, the combat-hardened, destruction-driven AI we have created, for all intents and purposes, seems logically peaceful.