In a weird way, if you can capture enough of the light field over time (an insane amount of data by the way) you can "go back" and get whatever picture you want. Which has the potential of allowing complete novices to grab the raw data with some local sensor, and to out source the 'picture taking' to someone who can look at it and return the picture that they should have taken. I know, its crazy talk and off the wall, but my friends in computational circles say that its no more crazy to them than a 1TB hard drive possibility was to me back in 2000.

It does not take 41 megapixel pictures, only 5 megapixel pictures. It used a 41 megapixel sensor, then down-samples it. This results in a higher quality picture.

"Another neat trick: the Lumia 1020 actually creates two images every time you press the camera button in Nokia Pro Camera—a super high-res version for editing and archiving, and an oversampled 5-megapixel copy for easy sharing via email or social networks such as Twitter and Facebook."