A-a-and this is
literally what I was writing about. What you brought, besides the first video (although it has so much less to do with Swarm AI than any news about processors for AI) are projects that have been going on since the beginning of the 10s, and still do not give clear results... like any fundamental research at the beginning of its journey.
These are still drones controlled by an operator from a base, only now the operator does not control one drone, but leads a group with him. And yes, this is still a multi-node relay of the base signal, only now not along the chain, but directly to the slave group. This is very characteristically seen in the video, where we do not see any group actions of drones other than movement - which has long been “hardwired” into the drone itself and requires the operator only to indicate the direction and parameters of movement... which is demonstrated in the video.
This is indeed why drones continue to be used in Ukraine.
Drones continue to be used because 100% electronic warfare coverage (which is economically impossible) will leave their own units without any communication at all (which is absolutely unacceptable in conditions of active combat), despite the fact that even the best electronic warfare installations do not have a very large range of action, by war standards... despite the fact that in working conditions they are relatively easy to find with any electronic reconnaissance station. And the short range means that they are in the artillery range...
However, there are already “trench” electronic warfare installations that create a “dome” over the squad’s position, models mounted on vehicles and even individual “backpacks” that prevent a grenade from being dropped on you from above or give a chance to evade the FPV drone. They are not cheap yet, but they are comparable to good drones at this time.
And this is only the beginning of active counteraction to “shield and sword.”
Remember that one has to jam the entire swarm simultaneously in order to neutralize the entire swarm simultaneously, and this more complex and expensive than you make it sound.
This is wrong. Precisely because the Swarm does not have a base or operator, it is limited by its
own resources. Each node is a part of its resource, and knocking nodes out of the network wastes the resources of the Swarm as a whole, and with them its efficiency.
I repeat, we do not operate on Neyman’s Swarm; in our case, it is limited by the available resources and the scope of the task.
It's becoming clear that you do not understand how swarm AI works.
No, this means that it’s just you who don’t understand how it works. First of all, the Swarm's network IS NOT PHYSICAL! Real-time distributed computing is powerful, but only in a stable network environment. But a stable network “in the field” is a myth, as anyone who has worked in the field will tell you. This means hello to data loss, desynchronization-reconnections and other “joys” of field communications... And if for a human operator this is only an inconvenience of orientation, experienced pilots manage to fly drones almost blindly and hit targets by adjusting the reconnaissance drone operator sitting next to him, hovering further away - then for Swarm AI this can literally mean a lag in detecting and locking on a target, as well as a drop in accuracy.
Secondly, with the fact that the Swarm AI drones will be “like regular drones, plus a processor” - you’re pretty hiperbolised. Modern drones are able to independently return to base when the signal is lost or the charge is depleted, and also avoid obstacles (although absolutely everyone disables this function to save charge...), but constructing a combat route and hitting a given, but not previously captured, target in a floating location - this task is several orders of magnitude more difficult.
In addition, the operator does not just press the sticks and make the drone whiz-whiz. He has the terrain in his head, a map of the combat mission in front of him, reconnaissance drone operators and a wing commander, whose task is to respond to emergency situations, are sitting nearby. Shifting all this to the Swarm AI itself will require connecting them with the same intelligence officers and training it to read intelligence data at the level of the same operator. And it’s also long and tedious to train his reactions to emergency situations, up to the level of a wing commander... and I haven’t yet mentioned the need to load defragmented map and combat mission data onto the Swarm's vehicles, coupled with the invention of a way without dumping all the data on each vehicle eliminate the risk of losing important fragments in the event of a node destruction or communication failure.
All this will require either releasing a really huge number of drones for any trivial task, which in the case of the traditional model could be done by three people and a couple of drones... or providing each node with sufficient power so that they can work effectively even in small quantities.
So far there is not a single hint that at least some of these problems are nearing resolution.
