It’s thoroughly evident by now that the project of Artificial Intelligence (AI) has turned out to be a lot trickier, to say the least, than it was perceived around 50 years ago. It seems that genuine, creative intelligence is something of a mystery even for today’s scientific inquiry, and as late John McCarthy, responsible for the coining of the term Artificial Intelligence, famously put it, “We understand human mental processes only slightly better than a fish understands swimming.”
So, even if we are nowhere near to replicate, or even understand, human intelligence per se, we can in certain cases mimic some of the most basic cognitive functions more or less convincingly. Of course, video games are a major arena for this kind of an exercise, in which the need for AI has been of importance from the very beginning of the medium. For instance, each ghosts in Pac-Man has its own separate AI behavior following own logic of how to get to the player. Unsurprisingly those AI routines didn’t fool anyone, even at the time, to believe the ghosts were actually strategizing their patterns, but regardless made the game that more interesting, or even a classic, for all I know.
Of modern games, the Portal series is now famous, among other great things, for its peculiar “AI” character GLaDOS, who is, interestingly, realized by not employing something that could be traditionally described as AI routines at all, but merely by scripting. The thing is, GLaDOS is nothing but a plot device through which the overarching, prewritten narrative is conveyed, so the scripting is understandable, and even unavoidable. She isn’t thus interesting from the point of view of actual AI that is under discussion here.
However, in addition to GLaDOS there were these cute little turrets that were also supposedly controlled by AI according to the fiction of the games. And in contrary to GLaDOS, the turrets actually possess genuine, although extremely basic, AI behavior, in that they react dynamically to the player’s actions.
Besides acid pools and abysses, the turrets are the sole enemies in the Portal universe so far, and what differentiates them, in my mind, from other AI antagonists in other games is the exceptionally credible nature of them. As said, human intelligence is still beyond the grasp of today’s most sophisticated AI systems, which means every simulated human thought process in any video game is a far cry from the real thing. This isn’t, of course, the case with our Portal turrets.
Indeed, because the turrets are, according to the Portal narrative, controlled by “AI”, the simulation of such a turret using real AI routines (which are admittedly at a poor state) is so much more feasible, and thus believable, than simulating behavior of any human(oid) character. And that is why the Portal turrets fascinate me, as they are, in a way, as credible as AI characters can be in a video game. It’s really not that much of a stretch to imagine something like those turrets actually becoming real at some point in the future, with a similar AI behavior.
So it seems that simulation of a certain type of fictitious “AI” is far more within the reach of current state of actual AI technology than simulating genuine intelligence, which said out loud sounds quite obvious. But still, it struck me how such low-level AI can be at the same time the most convincing AI, when wrapped with a cunningly designed context, like a sentient turret in this case.