A long MUD-Dev thread on intelligence of NPC's, and some extremely faulty arguments...
For a very long time now it has been assumed the definition of computer intelligence would be answered using the following conjecture:
It is proposed that a machine may be deemed intelligent, if it can act in such a manner that a human cannot distinguish the machine from another human merely by asking questions via a mechanical link.
That isn't a very difficult task using today's computing power if you consider a human player's willingness to believe that the NPC might actually be human. I offer this under novel pretenses, that certain games have achieved a level of interactivity by which a subset of the gaming population already fails to distinguish between a human player and a computer player. Certain chess engines I think offer the most convincing proof, since you often can't tell the difference in play between what a human might do and a computer. Of course we don't say that the chess computer is intelligent, but within it's domain it truly is very intelligent. In fact many chess engines can be ranked and given a score of how well they do, similar to the score given to human players. We may not like this idea, but yet I offer it as proof we are approaching intelligence in computing.
Let's take a different form of intelligence than the ability to play a game. Let's instead take the ability to interact in a social environment. Some recent games have started offering convincing interactions between NPC's and human players, namely the Sims. Those that play the game religiously don't necessarily agree with the concept that the NPC's are convincing, however, I believe if an equivalent game were created where there weren't any NPC's, then we'd see a lower level of social interaction between the real human players than exists in a game of Sims. How does this happen? Well, human players aren't that human when they enter a game. Human players are goal driven and often very predictable. Half the time you wouldn't be able to tell the difference between a human player and a computer player until one or the other opens their mouths and says something stupid (obviously the human player would clearly be the one saying something stupid).
All hands down, I think the line we draw on computer intelligence is faulty. Computer AI is already intelligent, and we missed it, because we think that intelligence is us. In many cases the computer AI is better at something, call it a domain ability, than we are. However, if the computer can't call your mother a name, become pissed off at your ignorant conversation, or otherwise behave irrationally it can't be intelligent in most peoples' minds. My ranting is mainly on the audacity of individuals to imply that the future focus on improving NPC AI in games is a task that provides little return on investment. The split in the game industry tends to be split 50/50 with groups supporting better AI and other groups supporting more games that are human player interactive.
Now, I don't think that even a 50% split in the industry can derail improvements to convincing AI in games and other software packages. Even a 99% split against it wouldn't kill the trend, since intelligence becomes easy to improve once we constrain our domain in which the intelligence has to operate. Many of the reasons we don't see more improvements is because of the broad scope we force on intelligence. A bartender in real life isn't very likely to do more in a night than listen to conversation, convey stories, and provide drinks. We know that the bartender is a person outside of this role, and that he'll be able to respond to a large number of other events outside of the domain of being a bartender. However, on a given night there is great possibility that he'll only NEED to be a bartender, as he'll be too busy to be anything else. Under rules for human intelligence, the AI bartender isn't intelligent unless he has some reaction to every possible phrase you could throw at him. However, there doesn't exist a single bartender that could meaningfully respond to any possible physical or verbal approach, so by that approach, bartenders are unintelligent (and so is everyone for that matter, since not a single person could meaningfully reply to any situation).
If you could walk up to a human and ask them the value of the 200th prime number, they spend a few minutes and finally give you an answer. The 200th prime isn't a joke, so you probably declare that person fairly intelligent, however most humans would probably give the wrong answer since the 200th prime isn't something within their domain. At the same time, you ask a computer to behave using a set of rules defining a bartender and give him a sufficient knowledge base with which to respond to various inquires. It does a great job of keeping the player entertained for nearly an hour as they buy drinks and chat with people around them. When asked a sexually oriented question about their mother, the computer player fails an appropriate response, and thus we call it unintelligent? Good grief, I think the lack of response is probably the most intelligent part of the bartender program you could possibly point out.
[Editorial Note: I've found a very interesting conjecture that identifies even a human's lack of ability to respond intelligently given ANY circumstance]
Can a reasonably intelligent human, discern another reasonably honest and
intelligent human from a similar culture, apart from a computer?