Removing humans from the cockpit frees designers up to do a lot more than save weight. The big restriction with human pilots is a ceiling for how much G you can pull before your precious human cargo blacks out. Silicon has no such restrictions; you can turn as hard as your fuselage will allow.
In the real world ability to pull more G's is of limited value. It mostly only matters when defending against a missile after all else has failed. Building an airframe strong enough to handle that load incurs a huge weight penalty.
The UCAVs built so far are actually stressed to lower G loads than manned tactical aircraft.
Also, imho AI fundamentally changes how machines of war are made. Currently, a machine must be made expensive and sophisticated since a human is inside it, and if it's already expensive, may as well add some more expense to make it a bit better, and a bit more, etc.
By contrast, with AI, the question becomes a matter of performance per dollar. If I can swarm an enemy with hundreds of cheap fighter jets, then it's okay for me to lose half of them.
This is existential to the US imho, since the US generally always has the best equipment period, but at a very high cost.
There is no way to build a cheap AI controlled fighter jet while still retaining any degree of usefulness. They can perhaps be made marginally cheaper but building airframes, engines, sensors, and weapons will still be extremely expensive. MQ-25 unit cost is in the $150M range.
Also endurance. Now you're only limited to hardware endurance and refueling capabilities rather than pilot fatigue --affording much better performance.
