hi Hari4sh
I like the art of movies, and I think you are a lucky guy to have an opportunity to participate it in a film being made.
Let me pop in some data I find interesting, before the final thought, just to have a base we can build up story upon. In general, operating systems like Windows consist of core that does all the job and other parts that the core synchronizes. If you build up an AI system, you can go along the same path, you can make the core that is not mutable and connect it to other parts that do specific jobs. Jobs may utilize inference rules, something like logic rules that transform input to conclusions and decides what an output will look like. To change inference rules, you can do it without shutting down the core, but to change the core (i.e. for self improving system), you have to reboot the system in order for changes to take the place. In other words, you have to shut down core, then replace it by improved details, then boot it up again.
Multiple machines can run multiple algorithms that can communicate over network. You can even run multiple instances in the same time on the same computer, and that is being called multitasking or multithreading (look up wiki on these terms). Core might be used to synchronize these threads. When running each thread instance on each machine, the processor consuming task of concluding knowledge can be divided and distributed among the instances. I.e. one machine can be in charge for logic, other for language understanding, third for imagination, and so on, and they all work in parallel as a separate processes, sending or receiving communication as internal events occur, just like I imagine the actual brain works.
As for logical inference rules that are carriers of a logical thought, we have deduction by which we can conclude informations implicitly contained in a starting set of premises (like solving a math problem using math formulas, but instead we use logic rules). There is a whole vast of logical inference rule systems by which we operate on premises and transform them towards conclusion. In each system, usually there are three or so rules that can be used to deduce all the other rules, and those most basic three or so rules are called axioms. The thing with axioms is that they are independent, meaning we can't deduce one axiom from another, and that is what makes them special among other rules. Axioms are given prior to any other conclusion and we use them for deduction.
Then we have induction that is really finding new rules (formulas) that hold on given data. Induction might be time consuming and may pollute incorrect results if the data is incomplete, if it pollutes any result at all, if there is a lot of data. It is like gambling, you run the induction process in the background and pray for it to find some useful rules like "all birds have wings", or similar stuff. Also, axioms may be induced from semantic tables of basic logical operators: and, or, if-then, equals.
As for data being smart by itself, may I present a thing that is particularly interesting to me? Computer programs are basically sets of ones and zeros, and they have a finite length. If the length of a program is finite, then there is finite number of combinations of these ones and zeros, and each combination represent another program, some may work, some may contain errors. Given enough length, while iterating over combinations, in some point of time we will have an AI algorithm, but the problems are 1. we won't recognize it and we will step over it; and 2. number of combinations exponentially grow as the length grows (2^length combinations).
And another thing: the same set of zeros and ones are interpreted differently when we use them as say picture, sound, text, data or algorithm, so it is possible that there exist a bitmap picture out there, that is actually an AI algorithm, but we don't know for it , as we never run it.
I hope this helps, let us know of your thoughts.
