Stefan Schuster works on archerfish: those are the fish which spit water at insects and other small animals close to the water such that they fall into the water and can be eaten by the fish. The animals can even learn to shoot down moving targets. At close distances, the fish just waits until the target crosses it's line of fire. In contrast, for larger target distances they employ a leading strategy, aiming slightly ahead of the target. The power with which the fish shoot at their target matches the size of the target such that larger objects elicit a stronger shot.
The fish face the difficult problem that the dislodged target can basically land anywhere on the water and can then quickly drift away. The fish has to be right where the target lands or other animals with steal the prey. How can the animal predict the point where the prey will touch down? Important variables in this calculation are the initial height of the target, horizontal speed, direction and vertical speed. Once this calculation has reached a result, the fish has to map its behavior very closely to it. This mapping includes potential obstacles between the fish and its prey as well as choosing one of two in case two prey were hit simultaneously. Of course, all these variables of course also depend on the shape and weight of the prey species and from which substrate it was dislodged. All this complicated processing takes only about 40ms!
It is possible that the animal achieves such short latencies by doing some pre-calculations before the shot goes off and uses some information from the shot itself to do these calculations. It turns out that even if no shot happens and the experimenter dislodges the prey, even then the fish are just as fast. Given the latencies in the involved networks, the animal has about 10ms to sample the trajectory of the prey. So all of these calculations have to be done with a very small circuit in order to be fast enough. Thus, high enough selection pressure can lead to incredible performance gains and computational power even in very small networks. This certainly is something to keep in mind when studying invertebrate nervous systems.
What a fantastic and entertaining talk this was!