It is important to note that humans, and other mammals, are very hard-wired to process vision and other inputs. For example, the retina is more than just an organic lens; it also encodes information about the motion of objects seen within its field of vision:
http://www.sciencedirect.com/science/article/pii/S0896627307...
So in addition to post-processing, there is probably significant pre-processing done by the sensory extensions of our brain. Similarly, the physicist Georg Zweig studied the cochlea and found how it mechanically separates sound into its frequency distribution. Zweig's research on the cochlea also resulted in the discovery of the continuous wavelet transform, whose discrete version may be familiar through its use in JPEG2000.
I have a couple of questions that I would love someone well-versed in AI to answer:
1. Does the difficulty of systems like driverless cars arise because we haven't been able to replicate the feedback loop mechanism that is largely hardwired? Is it some limitation of control theory (I'm just speculating). How is this so fundamentally harder than the ability to exponentiate one 1024-bit numer to another mod a third 1024-bit number (which is done in microseconds)?
2. With regards to aspects of AI that we might want to interact with, is human vision and visual post-processing done in the brain the hardest to replicate? Is it a matter of unknown algorithms or rather massive parallelism that gives humans a large advantage? If not, are other senses, like hearing (voice-recognition) or haptics harder to replicate?