As a child, I used to dabble in robotics. Even then it was apparent to me that the body of the robot was an engineering problem, whereas the mind was a scientific problem and outstripped the other in difficulty. Artificial intelligence as a subject has many layers of complexity. Even a seemingly simple thing like a rule-based system involves complex issues like epistemology. Another factor that makes Artificial intelligence difficult is the many diverse disciplines that come together in it. To have an overview of the field, it is almost imperative that along with computer science, you also be acquainted with psychology, mathematics, philosophy, neuroscience, just to name a few. Of course, that is part of the reason why Ai is so pursuable, it's never boring.

It is sometimes opined naïve to consider a computer program intelligent, that it is merely less or more 'complex'. This is principally a philosophical question. I would be inclined towards thinking that there exists a threshold at which, or there exists a set of qualities at the possession of which an artificial system may be deemed intelligent, or at least indistinguishable from a 'real' intelligent system. Then of course, there is the age-old debate on weather it is at all possible to have an intelligent machine. In my opinion, many of the arguments given for and against the possible existence of the thinking machine are a little premature, as we cannot even settle on a satisfactory definition of concepts like intelligence yet. The most prevalent idea being that intelligence is 'theoretically inconsistent', psychologically speaking; we would be forced to settle for a practical interpretation in order to make any progress at all. Drawing upon the Turing test, a program indistinguishable from a human may be considered intelligent. Thus behavior is the ultimate benchmark for intelligence (Skinner would concur). It is reasonable to assert that a particular behavior is triggered by finite set of preconditions, the rationale being that humans behave purposefully (thus the triggering from preconditions) and have finite life spans (hence the finiteness of the preconditions). Ah, I would allow for divine intervention, but providence is even more elusive than intelligence. Moving on, we can say that the set of all possible behaviors and preconditions is countable in a countable environment. Thus intelligence can be achieved by just a projection function on a set of possible behavior using the preconditions as a key. Simple, but difficult to argue against! The question is not of theoretical possibility, but of magnitude. Dreyfuss once concluded that a computer would never defeat a human at Chess. He overlooked the fact that the set of all possible Chess states is daunting, but is finite. Given enough raw computing power, it is possible to generate the entire Chess tree. Of course, an Ai constructed purely by enumeration is not only infeasible right now, but would also be unsightly enough to forfeiture the term ‘intelligence’.

There are two main schools of thought in Ai, symbolic and connectionist. I also worked in Ai for gaming for a while. Even though it is not academically very interesting till the gaming environment becomes complex, it is illustrative.