How much our brains are computationally structured is still an area of active debate ( https://en.wikipedia.org/wiki/Computational_theory_of_mind#Criticism ). There is no clear evidence at the moment for or against this at the moment; hence being in the state of open debate. ( Also, people struggle in almost every debate with having a consensus about the exact definition of intelligence and learning, since we do know not yet to define it properly ). Any way if you are interested in a bit of history of machine learning, then I will recommend the first 20 minutes of this talk : http://videolectures.net/mlss2012_lawrence_machine_learning/

The input of the machine learning system is a new image in my case. How well the trained model can predict correctly determines the quality of model. Important thing to note is that a good model should extrapolate well, that is work in unseen surroundings, by using past information it has seen and making new associations/strategies from them based on the current scenarios. That is not to say that models will be perfect ( It very seldom will be!). But it can handle a huge amount of cases with high precision. Models will get better with more data ( experience ) , and stronger algorithms ( better reasoning). These are all very akin to human learning process also.

Whether this is a metaphoric association or have a more literal correspondence, is the open question. It sure is a scientific and philosophically question that is fascinating a lot of my peers and me to no ends, and expect it will take quite a bit of work and years before we have some sort of understanding about it.