I get inside the taxi and tell the driver to go to the Burger Shot in Broker. After two minutes of safe driving, the taxi comes to a busy junction. Despite the red light, the driver pulls out and smashes into the side of another car. He begins to correct the error by reversing, only to meet the headlights of a police car. Five minutes later and the barely roadworthy taxi is still messing about at the same crossroads, so I decide to put the hapless driver out of his misery with a swift shotgun blast to the head. Of course I’m talking about GTA IV’s acclaimed artificial intelligence, which begs the question ‘what is good AI?’.
Perfect AI would be unfair – consider an FPS bot that could headshot from anywhere, they would just be too good. What games need is flawed AI, as to emulate the behaviour of humans. Enemy stats and abilities are hindered as to balance the gameplay, to make it possible for the player to win. Bond would have been overwhelmed had the bad guys actually been any good, and the player is expected to exploit their weaknesses in order to succeed. It is however, unlikely that professional henchman would genuinely be so stupid as to return to patrols immediately after the alarm is turned off, or so intensely distracted by an old erotic magazine left on the floor. Herein lies the difficulty in emulating human actions: human flaws are too random. Player-controlled actions cannot be predicted or manipulated like AI can. Although this manipulation of AI is key to single player victory, designers are constantly striving to blur the systematic approach taken by AI into a more unpredictable pattern of behaviour.
Handicap methods have long been used in gaming. You can expertly throw your kart around a course in first place, setting record lap times in the process, before being triple red-shelled by Donkey Kong five seconds before the finish line, spluttering you into third place. The same AI can also be beaten in the final lap of a race in which you have been slow and wobbly for the first two laps, as this is how the handicapping works. There have been far too many situations like this that have caused me to fling my controller across the room in rage. It just seems so unfair that the computer can beat you like that. Donkey Kong is such a damn cheating ape.
Another problem with AI is when and who you are allowed to kill. In the Call of Duty series, certain squad mates can’t die. In the GTA series, you can’t kill certain targets until a certain trigger is activated. I’ve shot countless NPCs in a countless number of games over and over again, just to establish who you can and who you can’t kill. The rules dictate that you cannot kill somebody that will shape the story at a later time, and it is obvious to see why this is so.
Perhaps the best type of AI is when the computer does things the designers don’t expect. Oblivion and Fable have so many moments of comedy, just through the bizarre behaviour of the NPCs. I’ve witnessed violent brotherly murders in the aftermath of an accidental mid-battle collision, only to discuss the local rumours and gossip with the nonchalant murderer seconds after the crime. Guards are equally amusing. In Assassin’s Creed I’ve stabbed guards in broad daylight in front of their colleagues, only to ‘hide’ and watch the clueless detectives shout “who could have done this?”.
As we look forward to future releases, namely Farcry 2, Fallout 3 and Fable 2 we must consider what we want from artificial intelligence. Do we want to be able to kill our friends? You won’t be able to complete missions if you choose to do so; some games grant you this liberty, some do not. We need to be able to understand the unwritten laws of AI behaviour, in order to manipulate it to complete the mission. There will always be ridiculous mannerisms displayed by enemies, and there will always be stupid ways in which to exploit some of the less believable character quirks. The vital factor in this issue is whether we choose to. It is in the very nature of gaming to adapt to these minor flaws. By actively playing games, gamers must use their imaginations. Instead of approaching games expecting to be able to find the flaws in the AI and game design and to use them to complete the game, gamers should attempt to immerse themselves in the virtual environment. When I die in GTA IV, I pay the fines and continue playing, whereas instead I could reset and re-load the game. I pretend that I miss the invulnerable targets until I’m allowed to kill them. I like role-playing and I deliberately absorb myself in the game’s world. The AI in games of the present and future should be assessed within the framework of immersion, but only if gamers are willing to indulge themselves in the fantasy that games create: which is surely the reason why games exist in the first place.