Ed Beach believes that the Tactical AI for military units in CIV 6 was working very well... Do you agree ?

[tedhebert]

In this interview,

, (thanks @Eagle Pursuit for posting it in the media reviews thread), starting at around 3:00 in the video, Ed explains that he was in charge of guiding the team in discussions about what NEEDED to change VS what was actually working pretty well in CIV 6 while designing CIV 7,
in order to stop the team from wanting to change EVERYTHING.

His take on military combat was that it was working pretty well once all the forces were put in their proper place, that the tactical interplay was working pretty well....

That's where the decision to create the commander unit comes from, they felt that the problem was actually getting your units to the city you wanted to attack, and that since CIV 5 and 1UPT, that part was very difficult and so they wanted to fix THAT part of the problem.

Do you agree with his take on things ?

Personally, I wholly agree with the decision and the reason for the commander change, I think it's fantastic, but I'm not really sold on the first part where he states that tactical AI was working well.... Seems to me it means they did not work on that too much

 

[Professor Phobo]

The actual tactical rules I think were fine. Not too complicated for that layer of the game, deep enough you had positioning/composition decisions that mattered.

The AI is wholly incapable of managing a unit carpet and it was a chore for a human, but I don't know if I'd count that as part of the "tactical" game as intended, more a consequence of other design. I don't think he's wrong to focus on solving the traffic problem before doing any more tweaks to how combat works.

 

[Eagle Pursuit]

I think the Army Commander is going to dramatically help to make combat more enjoyable.

It remains to be seen whether that improves the tactical AI.

Caveat: everyone should remember that Firaxis's standard for good AI means making a moderate challenge for median skill players at Prince difficulty. Not for high skill players at Deity.

 

[pokiehl]

It's hard for me to untangle exactly what the fault with the AI in combat was...I can see both points of view (that tactical AI was fine and it was 1UPT management that was the issue, vs that tactical AI was also bad)

 

[Siptah]

Do you agree with his take on things ?

Personally, I wholly agree with the decision and the reason for the commander change, I think it's fantastic, but I'm not really sold on the first part where he states that tactical AI was working well.... Seems to me it means they did not work on that too much

I mean, what should he say? The AI sucked, so we had to have some changes?
It's just a clever way of phrasing it: you admit that the AI was poor in some ways, and then present your solution for that particular problem.

And I can totally see that this helps the AI (and my fingers) quite a lot, actually. Unless it is easy to trap the commander during maneuver.

My big concerns with the tactical AI are elevations and district walls. In my experiences, the big problem with the tactical AI is that you can chop off units 1 by 1 with ranged units. I fear that elevations and complex wall formations create even more benefits to do this, assuming that they block melee attacks (btw, pls make Musketeers have at least 1 attack range). On the other hand, HK eventually got the AI to cope with elevations and fortifications, so hopefully, FXS can achieve it as well.

 

[SupremacyKing2]

I think that Ed is mostly correct. A big challenge with 1upt is moving the units into place. Units can block each other's paths. It can be an ever changing sliding puzzle. And getting your units into a place to take a city could be challenging because depending on terrain, siege units would sometimes need to be adjacent to a city center to be able to range attack it. If you skip straight to putting units already in the right location then you dramatically simplify a big part of the challenge. I think that is what the army commander mechanic attempts to do since the game will automatically deploy units around the commander in logical formations. So the AI does not need to worry about that part anymore and can focus on the actual combat part. Of course, the actual combat part could still suck if the AI is not very good. So army commanders do not automatically fix the problem, they just make it easier.

 

[TheGrayFox]

Couldn't really tell you if Beach is correct without actually playing against the AI.

I'd imagine the new army system dramatically increases its effeciency in moving units around which is probably the largest part of the battle but that alone won't make them good at tactical combat or sieges.

 

[Quintillus]

I can see the argument that if the AI were able to get the right forces to the front, it may do a decent job. In my experience the problem is the AI almost never gets the right forces to the front, so it doesn't have a chance to show whether it would be decent if it had a good balance of forces present. Instead it just sends units in as they arrive, and against a human adversary who has planned, that's not going to end well. The AI also seems to be incapable of sieging down cities so their land units have a chance - again, the question becomes, does it just not understand the concept of battering down walls, or is the problem that it doesn't know how to position its troops to get its artillery close enough but still protected?

On rare occasions I will see the AI attack walls; Rome battered down the walls of a Mayan encampment in my current game. So... maybe fixing the transport challenge would help?

1 UPT certainly makes the logistics of path-finding much more complex. Pre-Civ V, you could use standard pathfinding algorithms. If I invade the neighboring AI and they need to move troops to the border, there are textbook formulas for how to do that, which take into account things like forests costing more movement than grasslands. Small changes might occur like if a road is built or if a forest is cleared, but by and large, "getting troops to the front" is a solved problem in Civ I to IV.

Whereas in V and VI with 1 UPT, the rules of how to get there are changing all the time because every time a unit moves, it's blocking off a potential path that other units may have been planning on using, and that path may or may not become available again by the time the unit farther back gets there. There are no textbook formulas for dealing with that, so Firaxis would have pretty much had to come up with something on its own (or based on knowledge of how earlier games like Panzer General may have solved it), and the result left something to be desired.

Hopefully this will improve the situation for VII... and I would be curious to try V or VI with a "multiple units per tile" mod and see if that makes the AI noticeably better.

 

[Professor Phobo]

Do we know how Commanders behave in combat as a unit? Are they the priority archer target?

 

[Mr Jon of Cheam]

One concrete tactical thing that the AI did not seem very good at was targeting a single unit in order to kill it, they tended to spread the damage rather than concentrate it. Sometimes ok, often not.

However, I would say that manoeuvring the units was the biggest issue.

 

[SupremacyKing2]

I'm far from convinced that commanders will help that much. They have the potential to, but that doesn't mean the actual implementation will realise that potential.

What's more likely is that removing promotions from regular units will. The AI was never good at keeping its units alive long enough to get the juicy promotions, or how to make the most of those promotions while they had them. Something that human players find easy. That means that humans could use their strong uber-promoted units to chew through multiple times their number of weak AI units (even at the same tech level). By removing most promotions, the playing field will be level, and the AI will be able to make th most of its production advantages (at the higher levels). Commanders promotions might pull the other way, but probably not as strongly.

That is a good point. if the AI can win a few battles and level up their army commander, the promotions will help all the units in the stack, so it may help the AI stay competitive against the human player.

I wonder how effective it will be to snipe army commanders. I could see the human player adopting the tactic of sniping the AI army commander to weaken the AI and then finishing off the units. Sniping commanders may become the new version of this, where the human has leveled up army commanders so their units are buffed, and the AI does not because they all got killed early.

 

[King of Prussia]

I wouldnt say the tactical AI was good. Things like archers not shooting when they could and should. Ai loving to hunt down workers over other units. Ai not attacking out of cities when it should.

but i can see where Ed is coming from. Ai can play chess "well", but isnt capable of getting more than a few pieces on the chess board to play properly.

 

[Solver]

Yes, the tactical AI in Civ6 was alright, certainly by the end of development. Even Civ5's AI, which I think overall performs worse, was tactically better than it seemed.

I've said many times the Civ6 AI underperforms in combat, by a lot, compared to its capabilities. The main problem with the AI's combat performance is actually getting the units there. It has difficulty navigating the traffic jams, it ends up with some assigned units not making it to the target area, it fails to move them in a manner that reacts to battlefield changes, etc. It's not even something the AI engineers should be blamed for because pathfinding is hard and the mechanical design of Civ5/6 is very unfriendly to the AI. The game's maps have 1UPT while being small, units have low movement range and the already small maps are further constrained by water, mountains, small distances between cities and then neutrals like city-states. It's hard.

I found the encounters where things work out just right for the AI to be memorable - if it manages to get a decent force to the front lines, it performs pretty well. Not perfectly - the Civ6 AI is IMO too reluctant to focus down individual units - but it's a good performance. But most of the time it loses the battle before even getting there. For that reason I'm very happy about the Civ7 commanders, it sounds like a change made directly in response to traffic jams Civ5 and 6 suffer from so much, which can help the AI and reduce player annoyance at the same time.

 

[Gedemon]

Couldn't really tell you if Beach is correct without actually playing against the AI.

I'd imagine the new army system dramatically increases its effeciency in moving units around which is probably the largest part of the battle but that alone won't make them good at tactical combat or sieges.

yes, need to test it to judge, but the new commanders combined with possibly more spaces between cities are giving me some hopes for for the AI tactical performance.

 

[ggalindo001]

I think there are three layers here -- and one (my #3 below) not discussed here that IMO has been the biggest failing.

1 -- getting the units to the battlefield properly -- I think we can all agree this was a problem.
2 -- the tactical "who do I shoot at", etc. -- as mentioned, I found the AI often spreading damage around versus concentrating on inflicting maximum damage. I cannot tell you how many times the AI had OVERWHELMING advantage over me, but could not deliver the knockout punch -- often because of......
3 -- the strategic combat AI -- where do I concentrate my focus on battle -- one turn it was going after one city/battle group, next turn, it would focus on a different city -- and giving me room to breathe. Often once I introduce other units (often with inferior numbers) into the battlesphere, it would confuse the strategic combat AI.

#3 was BY FAR IMO the most frustrating in the AI. The basic of walls generally also throw off the strategic combat AI, or the random nature of changing from turn to turn. Often then due to this lack of strategic focus, I could easily counterattack and mop up the AI when they moved out of position.

I'm hoping #3 is helped by the commanders -- so to answer the original question, if we are just focused on "tactical" in turn AI -- I guess it is OK. However, the strategic doesn't hold a candle to games like Old World that can punish you.

 

[Teutonic_Tirade]

I do think sometimes combat and war would be more fun if they switched the AI that handles missionaries/apostles. They usually come in a non-stop onslaught from other religious Civs to the point that I declare war to keep them out, or make tons of my own to block them from coming into my territories through passes and stuff on my border.

 

[bv98132454]

One concrete tactical thing that the AI did not seem very good at was targeting a single unit in order to kill it, they tended to spread the damage rather than concentrate it. Sometimes ok, often not.

However, I would say that manoeuvring the units was the biggest issue.

Beyond the maneuvering of units to the frontline (and I think the army commander feature is the best thing we've seen yet from Civ 7, by far), the AI bonuses are calibrated to allow it to achieve a numbers advantage. The problem is that 1UPT makes it very difficult for the AI to use its numbers advantage. This is particularly true against human players who are more likely to make use of choke points and defensible terrain to nullify any numbers advantage the AI may have.

In real life military strategy, the general rule is that you need a 3:1 advantage to attack. With 1UPT, it is extremely difficult for the AI to achieve such a ratio.

I think 1UPT is superior to deathstacks. But, I don't think the option is 1UPT or deathstacks. I think there is a middle ground here that I'm somewhat surprised hasn't been implemented yet.

To me it makes sense to have some sort of supply limit mechanic. Maybe at the very beginning of the game the most units you can have on tile is 2 units (perhaps you'd have lower limits on desert/tundra/rainforest tiles, maybe allowing certain civs, i.e. Arabia - desert, to have a higher limit on certain terrain). Various technologies throughout the game could raise the supply limits on the different types of terrain.

 

[Ashbery76]

He has to say it of course. A bit like comical Ali would.

 

[SupremacyKing2]

I wonder what happens if you send your army commander with a stack and try to attack an enemy army commander with a stack of units before they deploy. Do the units in the stack automatically deploy or does the game resolve combat with the units still stacked?

 

[bv98132454]

It has been in other games. Civilization: Call to Power is an example of a hybrid system. So are all the Paradox games. Even the TW games if you chose to auto-resolve every battle. In all cases, stacks - but with a hard-ish limit on the number of units in the stack. And the stack fights as a whole rather than the individual units doing sequential individual units. Why the main civ series never adopted something like this is a good question because it seems like the natural thing to do. 1UPT is more fun and involves more player agency, but it has always felt like it was never quite on the right scale for Civilization.

I think it would work even better in Civilization than it does in the Paradox games because the different terrain types are far clearer in Civ than they are in EU or HOI (which often have terrain modifiers that you can't see at a glance).