Does the AI really use tactics?

[aw89]

ive seen it try to make me intercept "flanking" units, they just hold several units ready (cav) and try to make you attack the bait, quite clever actually...

[Red Harvest]

It uses a few ploys, but the tactics are weak and AI strategy once in battle is also weak. It will charge the archers/skirmishers in front of your line...even if doing so is suicidal. It also likes to charge its archers as if they were melee at times. It seems to be either overly aggressive, or overly passive (when it stands there and lets you cut it down.) Given the choice it will march it's slow phalanx units at your fast light skirmshers instead of engaging melee units.

When the lines meet, the AI often turns many of its units trying to find the supposed best tactical match up at the last moment. This creates mass disorder in its ranks (and yours if you try to do 1 vs 1 match ups rather than a "march through" order.) It should be arranging its battle line ahead of this rather than waiting until the lines meet. "Intercept course" headings really cause chaos when this happens.

[Dutch_guy]

if they deploy, they always do it the same way. en never use their skirmishers
the correct way ( they always skirmish the wrong way when charged, always away from the main battle line. at least the numidians do that alot in my gamews. )
and they only flank with javelin cavalry, if they flank at all.
but I have never ever seen the AI do any hard tactical maneuvers (?)and also I've never seen them deploy in any other way than the 3 line formation.
this al is on the medium / very hard (battle/ map )

[Puzz3D]

When the lines meet, the AI often turns many of its units trying to find the supposed best tactical match up at the last moment. This creates mass disorder in its ranks (and yours if you try to do 1 vs 1 match ups rather than a "march through" order.) It should be arranging its battle line ahead of this rather than waiting until the lines meet. "Intercept course" headings really cause chaos when this happens.

You'll see the AI do this even if all the units are of the same type. I don't know what kind of matchups it's looking for. I can march my line of Roman swords up close to a line of AI phalanx units, and when I get close the AI units go in all different directions with some even marching away from my line. I have no flanks exposed, and all the matchups are the same. I can't make any sense of what the AI is trying to do unless it perceives that a frontal fight is not to its advantage and is trying to use indirect attacks, but units that turn and expose their flank are ignoring the fact that an enemy unit is right in front of it and can charge into its flank. If it's trying to hit the flanks of my units, it can't because all my flanks are covered except for the two ends.

The AI in original STW v1.12 is more intelligent than the AI in RTW. If you get close to an AI unit in STW v1.12, it will turn into and charge you frontally even if it has the weaker unit. Initially, it will maneuver for an indirect flank attack until the enemy unit starts moving toward it. It does this frontal assault because it knows that it's the best choice given that the human saw the attempted flank move and came forward to confront frontally. If the AI unit doesn't turn to fight, it will get hit in the flank, and the AI knows that's worse than charging the stronger unit. Once that unit is engaged, it will attempt to flank the enemy unit with a second unit if it has one available within the activation distance. I've never seen the AI in STW v1.12 try to hit a flank that's covered by another enemy unit. In RTW, I see AI units exposing their flank to enemy units as they try to charge into unexposed enemy flanks.

Basically, the AI in RTW cannot defeat the human player with tactics. It beats you only if it has the stronger units.

[Red Harvest]

Puzz3D,

That's exactly what I see as well, especially compared to STW/MTW. I suspect they have tried to add some features with intercepts and the like, but have not had time to tune the AI to do things at the right time. As a result is self destructs like this. It's like half the new approach has been coded, while the other half has not yet been coded/updated.