I think you guys are missing the point a bit with the pure UCAV vs manned fighter debate.
That is a far broader, longer term and more theoretical discussion that the one at hand.
The Dark Sword is supposed to be an unmanned wingman to the J20 (and potentially other legacy aircraft, but let’s just stick with the J20 for now).
That immediately gives the J20 a lot of advantages and options compared to pure manned enemy 5th gens.
Some obvious examples off the top of my head:
- the DS could range ahead of the J20, and either use active radar and/or passive IRST to hunt for enemy stealths while broadcasting their sensor data back to the J20.
Either opfor detects the DS first and engages it, thereby giving away their own location, or they get detected by the DS first, allow both the DS and the J20 to attack first.
With the DS up front, that should far extend the normal detection ranges, which will give the J20 far more manoeuvring options it might otherwise not have against enemy stealths even if it detected the enemy first.
- the DS could significantly expand the number of missiles available to the J20 pilot.
- in a fight, current AI should be able to easily handle BVR. In WVR, there are still a wide range of roles the DS could perform.
Even with a basic dogfighting AI that is normally no match for a manned pilot one-on-one, it is still a dangerous threat the enemy pilots could not afford to ignore.
If they are chasing or being chased by the DS, they are not doing the same to the J20.
An enemy pilot might be able to shake a DS on its tail, but doing so will likely cost him/her a lot of attention and energy, thus potentially making them a much easier target for the waiting J20.
On the flip side, an enemy stealth chasing a J20 may have to perform defensive manoeuvring itself to shake a DS snapping at his heels, giving the J20 time and opportunity to break away and turn the tables.
If you want to get into the more advanced options, you can have a hybrid man-in-the-loop control system, whereby a human pilot is looped in and could prioritise targets which the DS would then go attack/defend.
You could do all of that in training/simulator, and throw in machine learning and have a constantly improving AI in the DS.
If you can build a good enough simulator programme, you could load that, and the machine learning AI into a supercomputer and after a few million, ‘games’, the AI might actually surpass the best human pilots much like how it has happened with Go recently.