My view of true artificial intelligence is one that is completely adaptable to all situations, one that has no pre-programmed instructions for various tasks and that tackles all situations using logic and anticipation. A true AI will be unbeatable (or as unbeatable as possible) at any given logic task where time is irrelevant (an AI running on faster hardware isn't any less intelligent than the same AI running on slower hardware) and will not even require special programming to perform any task.

The AI should be able to take any situation and act accordingly in such a way to maximize the probability of the most ideal outcome. The only way to do this is to accurately anticipate how other actors in the situation would most likely behave, to take into consideration all of their possible actions and to give weights of probability on each one. Initially this would probably be based on assuming the other actors are also as intelligent and calculating as an AI, and over time build a profile of each type of actor to better weigh probabilities of their actions. The AI would calculate the actions of the actors in very small increments of time so that to calculate all the possible actions an actor could take the AI would need only to cycle through the possibilities one moment at a time and note all the motions of an actor that result in something meaningful. I.E. The AI will not need any inspiration or guidance to be able to see every possible action an actor may take.

With a base engine somewhat like that it should be possible for the AI to ultimately act 100% logically and as best as possible in any situation and applied to any function from controlling a physical body to orchestrating a ruthless extermination of mankind (but why?).

In the course of an AI's advancement to it's goal it will likely find that improving itself physically (i.e. speed of computation) will greatly shift the odds to the favour of it's goal. Depending on the set goal the extent of its improvement may vary. For example if the AI had the goal of beating a person in a game of chess, it may improve itself to a sufficient level that to improve itself further would be a waste of time because it is already impossible for a human to defeat it. But if the AI was asked to defeat another AI with the same abilities it would result in both AI's constantly improving themselves in order to outdo the other AI. Chess may be a bad example because there is probably only so much intelligence needed to master the game, but you see the idea.

A problem I can see is how to give the AI a goal in the first place. I don't know why an AI would find it necessary to perform any action. Unlike humans, AI won't have the same drive to gain success and achievement. This drive in humans is hard coded into our personalities and we don't have a choice but to continue forward for no reason. Any ideas on this would be appreciated, it's been bugging me recently.

Dylan McGannon - peltium -at- gmail.com