> To assume the creation of AI with "a positive feedback loop that goes nobody knows how far" without humans first understanding how seems more of a belief in magic to me.

Not really. This is pretty much a definition of a positive feedback loop.

To give a very simplified example, imagine that a mind of IQ N is able to create, at best a mind of IQ N+10. So say, the smartest human alive has 150 IQ. He goes and creates an AI that has 160 IQ, which then goes on to create a 170-IQ AI, ad infinitum.

Of course you could argue the relationship is different. Maybe the ith mind can create at best an N+(1/2)^i mind, at which point the whole series will hit an asymptote, a natural limit caused by diminished returns. But it would be one hell of a coincidence if humans were close to that natural limit.

So basically, what we need to do to potentially start intelligence explosion is to figure out how to make a general AI that is just a little bit smarter than us. Which seems entirely possible, given that we can use as much hardware as we like, making it both larger and faster than human brains.