Imagine a computer model of a stream of cars on multilane highway or city street grid.
Arbitrary waypoint objectives are assigned to each vehicle at program start, and when each waypoint is attained.
Cars always move forward and never stop or reverse, but speed, turns, and lane changes are up to them.
Cars will not collide with other cars.
Traffic densities vary over time.
Cars have neural-net rules for speed and turns which are initially random, but which are genetically mixed in each generation.
Cars know these variables for themselves and others near to them:
gap ahead to next car
50% guess about the gap ahead of car ahead of themselves and others nearby
when last lane change or turn was, and the nature of it
Cars in this evolutionary environment are rewarded with offspring based on how efficiently they can get to their waypoints.
What will happen? I conjecture that the cars will exhibit behavior rather like those seen human traffic.
Most if all, it would be fun.