Here are a few things.

Ranked in order of importance by groups (Ie: I think fixing diplomacy issues is more important then the faction specific issues)



Diplomacy:

1) AI only ever declares war on 'weak' factions. You never see multiple factions declare war on a stronger one.

2) AI doesnt care if one person is conquering the entire world.

3) AI is extremely transparent - always uses close borders before war.

4) AI gives resources for "treaty" early game. (They shoudnt)

5) AI is still exploitable using Luxury resources, dust, and research (eg: In summer/winter certain luxury resources get big boosts when traded, also in the link below the BL will give me 120k gold to end war)



Research:

1) AI sometimes doesn't research strategic resource mining technologies (Unanswered bug thread with saves)

2) If an AI isn't at war they sometimes wont research weapons technology- and if they pass that tech tier they will not go back. So if the AI cant find hyperium, you can be fighting against units that are equipped with glassteel when you have palladium weapons (even though Endless AI is higher then you in technology)(Unanswered bug thread with saves)



Buildings:

1) AI doesn't use dust to hurry units/buildings. If they did, in the save above I should still be fighting the BL off. They shouldn't be down to 1-2 cities with no units and 120K+ gold.(Unanswered bug thread with saves)



Battle:

1) AI does not take into account unit formations (see Abmpicoli post)

2) AI will send units away from other units to be surrounded by enemies

3) AI sometimes focuses attacks on one city without clear reason (city is not always closest, easiest to take, or biggest)

4) AI will still focus on taking out militia instead of other units in certain situations (You should force AI to target them last).

5) AI still sends questionable unit mixes against human player (eg: sends pure wyvern and necrodrone armies against armies of Rangers/Harmonites)

6) AI pulls garrison out of city to fight losing battles.

7) I can get good results on large open battle maps by sending ranged units in all directions. AI will doggedly send entire army after one unit at a time allowing me to pick them off by range.

8) AI still focuses on weak/wounded units too much allowing for easy exploits (IE having a wounded hero or wounded wyvern run around the map)



Units:

1) AI seems to put more emphasis on later era accessories- when in reality titanium rings/pendants are better then any other.

2) AI doesn't use heros as much in unit stacks as human players do.



Faction Specific:

1) Roving Clans hardly ever uses mercenaries.

2) Broken Lords routinely miss/don't use dust related techs.

3) Drakken never "force-peace"

4) Factions don't play to their playtype

I've said this:

/#/endless-legend/forum/5-general/thread/2890-your-feedback-on-ai

I *know* this can be something that will probably be painful to do... But would it be possible to expose an interface for a third party AI? I'm sure that there are lots of geeks and nerds over there (hey, maybe I'm even one of them ) that would love the oportunity to develop their own AI for the many aspects of the game, such as combat , strategic decisions, etc... Then we could even place matches between the different AIs and see who is the winner..

To which you have answered this:

Manu wrote:
...

You can enable AI tools, and even switch between empires to see what the AI does, by following instructions from this thread: /#/endless-legend/forum/15-modding/thread/5778-accessing-debug-modes

...

I think you are missing the point... I'm talking about being able to outsource the AI, by exposing an interface to it, and let interested third parties develop alternate AIs... I don't know how difficult that is , and it will basically depend greatly on how you have designed the AI's architecture... If you already have a well encapsulated AI, then it will be only a matter of exposing the interface and allow some communication protocols to hook into it (and yeah, I know, it's easy, any MIT Ph.D team can do that... )...



However, I think you are suggesting this as an answer about my other comment : /#/endless-legend/forum/5-general/thread/2890-your-feedback-on-ai



So, yes, we can run at debug mode and make some inspections, however we don't see the internal decision making that made the AI take one decision or another... We can only see the result as a black box... It is also difficult to track turn by turn all the AI decisions, by swapping through all the empires in debug mode, since everything keeps going on... That is what I've said... We don't have even a major event log, it seems... to see when each player have built what... Or am I wrong, and this is hidden somewhere? Are there any logs I can activate?

For me the AI seem capable of building large and somewhat mixed armies often led by heroes. However they seem to mass produce such armies but rely only ever use them in defense and I rarely see them even escorting settlers with these, or garrisoning new cities quickly. Another issue is that the AI doesn't seem to ever invest in strategic resource equipped armies even when they have plenty of resources and uses only iron and dusf gear making them far weaker to player armies of a similar era. Another issue is that while the AI is constantly expanding horizontally, it never seems to build up its core cities with many districts. Usually for any given city they have a maximum of four districts while most only ever have two. Well these are the main problems I see with the AI so good luck improving them!

Caotico has a good list.



I'd add that the AI doesn't appear to pay any attention to nearing victory conditions:

• Does not end treaties or cease diplomatic trades when a player approaches a diplomatic victory
• Does not wage war when a faction is nearing a victory condition faster than the AI will be able to
• Does not defend or target capitals when nearing supremacy victory

I've started a game in debug mode, and switched to the 2nd player AI to see what it is doing:



They have this unit design (at turn 67: Roving Clans - Turn 67.zip)



What I see as wrong with it:

1) No accessories whatsoever! Not even basic iron ones... Only by placing the iron tier 2 necklace with improved movement would raise hell to ranged units, due to the increased movement and initiative... Add the tier 2 ring and, besides the extra viewing range, you have protection from ranged attacks... This unit would be a nightmare to stand against! Is this made on purpose (for not making the AI too d* difficult )? Not talking about improved movement for eventual interceptions or even a good old blitzkrieg...

2) Not a single equipment with titanium or glassteel, even with the empire having a good quantity of it... I know, maybe this is something not to be placed in the bulk units... But they don't have extra designs for elite builds! Only the plain unit types... Maybe it would be wise to create an evolution plan for the units, and have different designs according to different assigned roles (garrisons with less mobility, but more hitpoints , damage and initiative; attacking armies with high mobility to reach far, quickly ).

I think the AI should never ever trade away Victory techs. Currently it does so even on endless, even if you have to give them everything you have. But it should realize you don´t need everything you have if you can win in two turns. I don´t think i´d have won scientific without using this yet.

BPrado wrote:
I think the AI should never ever trade away Victory techs. Currently it does so even on endless, even if you have to give them everything you have. But it should realize you don´t need everything you have if you can win in two turns. I don´t think i´d have won scientific without using this yet.

You mean the VI era techs? To tell you the truth I'm the kind the player who have difficulties finishing a game, because I keep get some kind of impulse in the mid game and I end up starting another game all over again... If that is true, yes... This is one to place definitely at the records ... But I think that Victory tech per victory tech, if the AI is chasing a victory techs and has more victory techs than you is a valid trade...

Add a button to battle UI - > AI on/off. It turns the AI anytime for me. Sometimes I start a manual battle but after the first round I find myself overwhelming advantage so I want AI to finish it for me.

abmpicoli wrote:
You mean the VI era techs? To tell you the truth I'm the kind the player who have difficulties finishing a game, because I keep get some kind of impulse in the mid game and I end up starting another game all over again... If that is true, yes... This is one to place definitely at the records ... But I think that Victory tech per victory tech, if the AI is chasing a victory techs and has more victory techs than you is a valid trade...

That´s called attention deficit I haven´t finished a civ game in months.



And yeah, era VI techs. They never do it cheaply, not even a victory tech for a victory tech - but when you´re still far enough from winning, some techs and some luxuries are enough, while when you´re really close (and nothing should be enough) you have to give every single luxury and every single strategic and some 5 or 6 techs, it´s a lot, but it shouldn´t be doable. I´m not winning scientific victories, those are actually pretty diplomatic victories instead.

This is another thing I've found... Once the AI picks a target, it wants it *dead*... Even when it is impossible... Even when there is an easy pick unit that can be shot within 1 turn and at least harass the player with a battle of attrition...

In trade negotiations, the AI seems to over value it's own techs way too much and under value the player's weapons and armor techs. This causes some really weird trades at times.

Troy_Costisick wrote:
In trade negotiations, the AI seems to over value it's own techs way too much and under value the player's weapons and armor techs. This causes some really weird trades at times.

@ developers

This is such a case where we have a hard time to prove or illustrate our acceptions, because we don't have that information in the save game to actually see what deal was made... We don't have any kind of transaction logs...



@Troy:

I'm not sure about that, Troy...

This is one such case of hard to prove acceptions, because we don't know the context ... Do you have a concrete example? With a screenshot?



As for overvalueing the techs, I, personally, tend to overvalue my techs, because the techs help era advancement. So even a trivial "search party" tech, provided a deal sweetener could make the adversary reach a new era, research full blown new mineral armors equipment, and ruin my day.

Here's an example of the AI not being very smart when negotiating:

Vicarious wrote:
Here's an example of the AI not being very smart when negotiating:

You realize that the 591 Spices if sold on the Marketplace would be worth ~2000-3000 . In any case it's worth a lot more than 271 , not even considering the booster value.

Propbuddha wrote:
You realize that the 591 Spices if sold on the Marketplace would be worth ~2000-3000 . In any case it's worth a lot more than 271 , not even considering the booster value.

Of course I do. And I'm surprised that the Roving Clans value 591 Spices less than 219 .

Vicarious wrote:
Of course I do. And I'm surprised that the Roving Clans value 591 Spices less than 219 .

LOL. Sorry was reading that one backward....

Propbuddha wrote:
You realize that the 591 Spices if sold on the Marketplace would be worth ~2000-3000 . In any case it's worth a lot more than 271 , not even considering the booster value.

I think the main issue here is another, entirely: How come that the Roving Clans had 591 spices in stock? I avoid keeping more that 2 x [boostercost] at stock if I have autosufficiency . I simply convert everything every surplus beyond that value to dust...

About the AI to value dust more than a luxury resource, there can be reasons to value dust more than spices: if they are at a deficit... However , they *shouldn't* have a deficit, if they have such a great spices surplus... We should take a look at the current situation of their treasury (run in debug mode that turn and take a look).

I have made a mistake when posting evidences at this thread: we should always post, together with the screenshot, the savegame from the turn where the screenshot was taken: this way, we can open the file in debug mode and take a good look at an eventual rationale that made the AI reach that decision.

Thank you for soliciting feedback on this. Endless Legend is a very complex game and I don't envy the AI programmers' job on this...



Your initial focus should be on defensive strategic military play. I believe this is the #1 area the AI needs to improve on. It does not matter if the AI can build a strong economy, act as a savvy diplomat or roleplay a faction when a human player can conquer it at will. Tactical battles are usually won before they start based on who brings the bigger stick. Survival needs to be the #1 concern and when the AI can defend itself, the rest of the game will open up.



One general thing is that you need to continue to balance the game play as balance is intertwined with AI performance. If the AI uses a strategy that is weak, the AI will be weak. You have the choice of strengthening weaker strategies that the AI may use (or can handle better) or telling the AI to ignore them. In either case it's important for the AI (and the AI programmers) to recognize these. One area is the shortage of defensive advantages that make aggressive play too rewarding.



I'll go through a few topics, focusing on AI Survival. In general, the AI needs to be paranoid and overcompensate to it's lack of creativity by focusing on military.



The Arms Race



The AI should never skip important military technologies (Mill Foundry, Alchemists Furnace) in the early game. Military technologies and strategic resource extraction should be priorities throughout the game. Unit designs need to focus on what is important in the combat system (Initiative, Damage) and the AI should prefer Ranged units over Melee. The AI should be in a position to dramatically increase output to produce units if needed and drop whatever it is doing to go on a war economy if needed. All armies should be upgraded to the latest tech level if possible.



Survival Diplomacy



The AI needs to do a better job of recognizing threats and reacting sensibly. The AI needs to constantly guesstimate whether or not it is stronger than other Empires. If the AI is weaker it should cooperate with other Empires to fight the threats (mutually beneficial trades, declaring war together) or pander to the stronger Empire. It should look for opportunities to strike at a stronger empire that may not be in a position to defend itself (ex. troops are away fighting another empire)



Army Deployment



One big sin is that the AI get's into fights it should know it won't win. The AI likes to stack units in garrisons and wait to to die via Sieges. When an invasion comes the AI needs to concentrate forces to destroy the invader or fall back to fight later. If the AI cannot defend a non-Capitol city (i.e it wont be able to muster enough units to destroy the besieging army + reinforcements before the Garrison does), it should abandon the city (possibly Salting the Earth) and save the forces to defend it's capitol.



Faction-specific Considerations



The AI needs to start using faction abilities. Drakken Force Truce and the Vaulter's Teleport are two very powerful tools I don't believe I've ever seen the AI use.

I believe the AI is missing a grand strategy directive that overarchs all other considerations. A good example for that is how AIs deal with the cultists.

In any game of cultists vs AI I played, so far, some AIs tried complimenting their way into my good graces, then burned down any converted village they came across, if allowed to do so.

The AI should make up its mind whether it wants to play nice or evil and then have all its subfunctions act, accordingly. If it wants to be nice, it should do positive diplomacy (compliments, fair 1:1 trade offers, alliances), take this into account for expansion plans (at peace/alliance expand towards the ally, before that expand anywhere but towards the potential ally to prevent border conflicts) and use it for its military disposition (defend less towards territory protected by allies that do not give foes the right of passage).

One more thing I regularly notice: Single settlers moving into territory without even a scout to accompany them. The AI likes to do that, repeatedly, even if they just lost a settler in that very region, before.