I am more curious about the “compute” of an AI system like this. It must be extremely complicated to do real-time video feed auditing and classification of targ

I am more curious about the “compute” of an AI system like this. It must be extremely complicated to do real-time video feed auditing and classification of targets, etc. How is this even possible to do without having the system make a lot of mistakes? As much AI talk there is on HN these days, I would have recalled an article that talks about this kind of military-grade capability. Are there any resources I can look at, and maybe someone here can talk about it from experience.