NASA has a open source A.I. software called Livingstone2 that you can start with, then we can add more modules. If you are ready, I can help.
Let me check this out a little more in depth, I'd heard of it but haven't gotten my hands on it yet...
For input devices, frankly I don't know. Like guru was saying I think we could just add on "modules" i.e. sight, infrared signature recognition, it really just adds up to what we can find for apps in opensource that can be modified to the original "brain".
My final goal would be for the A.I. to activley search out information to learn. For example, if you input an unknown parameter into a computer now most of the time your gonna crash or get some error message. Even a simple A.I. might be able to check and see if the "unknown parameter" is actually a valid command that it just hasn't associated with it's previous programming. Then like what was stated earlier, if it learns other proggraming languages, codes, scripting, etc, then it should be able to adapt other technologies to itself... and possible assimilate us all into it's mothership (you never know).