What a fascinating topic!! I had no idea this was possible, so when I got to the PDF link I took the time to read their whole paper. The whole time I was reading it I was wondering what I would see if I simply applied light to a typical 5mm LED and measured it with a DVM. So when I read your last post above I decided to do just that. I chose a blue LED (just arbitrary color choice) and measured about 50mV with it standing vertically under my fluorescent lighted bench magnifier, about a foot above it. I then laid the LED on its side and used an LED flashlight which has 109 bright white LEDs, and got as much as 230 mV from the LED! I had the flashlight about a foot away from the LED. In the rather bright incandescent-replacement fluorescent lighting I have in this room I can read 3 mV from the LED simply oriented vertically in my lap. So if the circuit being used with such an LED as an input could distinguish the difference of a few mV ( I know a comparator could do it) the LED does indeed make a good light sensor. I wonder why this idea never went to market on a massive scale? The ideas they present in that paper seem like good ideas, and very inexpensive to implement.
I am very new to the Arduino, and still working through beginner level books experimenting with it. Can the Arduino differentiate single mV readings and have their programming react on them?