01.30.07
Double Agent Blooper (or: How To Recognize Great AI)
To answer the question in the title: You know it’s great AI when it’s so responsive that it sabotages itself in certain situations.
Of course that’s a massively crude way to say it, but I arrived at the conclusion about a week or two ago when I was playing Splinter Cell 4: Double Agent and experienced a hilarious situation due to a combination of great AI and horrible timing.
I was hiding in hostile territory, my eyes on the door to a server room I was looking to enter as part of my mission. Inside the server room was an enemy. Outside it, in the antechamber where I was hiding, was another enemy, looking for a screwdriver. The guy on the outside finally found the screwdriver and made to enter the server room; but at the same time, the guy in the server room was planning to leave.
They reached the door simultaneously and as the entering man tried to open it, it bumped against the man on the inside and was blocked. The guy outside proclaimed “Something’s blocking the door!”, drew his gun, and warned: “This door’s going down.” Then he kicked in the door with the predictable result that it burst open, hitting the other guy directly in the face, and throwing him back into the server room, unconscious. The man on the outside then entered the room carefully, discovered his limp ally and shouted horrified “We got a man down!”
My cover blown, the whole facility on the lookout for me, and all my trust with the terrorist organization that I was infiltrating gone, I loaded and tried again. This time the timing was different, the man in the server room had time to leave before the man outside found his screwdriver, and the game continued.
This would have never happened with a pre-scripted AI. A scene such as this would’ve been meticulously timed in a game such as Half-Life 2 or F.E.A.R., but in Splinter Cell, the AI is actually scripted to respond realistically (within the scope of the game world) to the environment. If a door is blocked, he kicks it down. If the light in a room is suddenly turned off or a door left open, he carefully goes to investigate (but never carefully enough, because it is impossible to be careful enough when opposing Sam Fisher). If an enemy in a Splinter Cell game finds one of his allies rendered unconscious, he will wake the fallen man up. The man on the outside in the scene described above acted correctly to overcome the hindrance he encountered; unfortunately he had no sense of context, so a problem arose.
That, and it was kind of annoying that he saw fit to blame his idiocy on me, but that didn’t make the situation any less hilarious.
It goes without saying, the more advanced - the more intricate, responsive, and flexible - an artificial intelligence is, the more prone to error it becomes. To Ubisoft’s credit, this is the only example of AI complications I have encountered in the game, and I am currently on the last level (which is a piece of pure narrative brilliance, but I won’t get into that here, just buy the game!). I will not hesitate to name the AI of Splinter Cell 4 the most impressive artificial intelligence I have ever seen in a game - if not the most flawless and immersive. Stealth games place high demands on the AI, and SC4 delivers.
It also looks beautiful.
