Memory, gradients, and cartography
Notes:
Each particle is blind, quite stupid, but globally the swarm has already a small memory, for if it had no memory at all, it could be lost in a no-hope/re-hope cycle.
But this is indeed a very rudimentary one. If we increase it, and if we improve the swarm ability to compute gradients, the swarm, seen as as a whole, could have quite quickly a pretty good idea of the "shape" of the search domain.
Now, again, it is not at all costless, so it is not clear whether it is worthwhile or not.