The idea of "I am a Strange Loop" is that we develop a sense of self from self reference.
The central thesis is that one can explain how consciousness and self-awareness can arise from previously inanimate systems by the creation of tangled hierarchies where there is no top or bottom because moving up or down in the hierarchy enough times will return you to the level you started at. The author calls these hierarchies strange loops and the title of the book refers to his belief that every conscious "self" is a strange loop.
The act of believing it had an I, or self, would create one out of the loopiness of the logical process of self reference. By believing it had an "I" or self, it's actions would be influenced by this belief/ "I" thus reinforcing the concept and letting it grow into something more developed(Hofstadter likes to reference his experimentation with a camera that turns it's focus on the tv image it is recording and begins to make very unusual recursive images). The question then becomes, how can you get a computer to believe something? Wouldn't it need an "I" to do that?