Let's talk about AI

[Jotari]

I love this series. There's not game in it I've played that'd I'd say I didn't enjoy. Certainly some I have problems with, but the gameplay overall has always entertained. However, there is one aspect of the series, a core and crucial aspect really, that occasionally bothers me. And that is how completely and utterly perdictable the AI is. I know exactly how every single enemy will react in every single scenario for 90% of the maps in the series. And more often than not it's charge recklessly for the unit they can damage the most/kill without any care for their own well being. It feels less like I'm fighting humans and more like I'm fighting robots. Even in Gaiden/Shadows of Valentia where enemy units are meant to have some semblance of self preservation and seek out heal tiles, they do it so robitcally, always after reaching a certain threshold of HP loss and once that happens seeking out recovery becomes such a massive priority they completely stop being any sort of a threat. Maybe I'm the only one that wants this, but I'd like the enemies in the game to be able to surprise me more. Whenever I mess up in Fire Emblem, it's rarely because the game has done something that I think is intelligent, it's useually either because I've done something stupid, or I miss a 70% hit chance a few too many times. So I propose a revision of how Fire Emblem handles enemies. Something that would make these fleshbags we're fighting seem more like actual people. I call it, personality.

Every enemy in the game has a personality type. Ideally this should be something visible to the player on a stat screen, though with Three Houses new aggro detector it doesn't necessarily need to be, as enemies personalities could be figured out based on that. Basically instead of having one universal AI for almost every enemy in the game, there are several. For example.

*Pragmatist: Typical Fire Emblem AI, goes for the enemy they can do the most damage to.

*Confident: Also prioritizes attacking the enemy they have the best chance of actually hitting (good for enemies with powerful but inaccurate weapons).

*Coward: These enemies prioritize their own survival and thus target your units that will deal the least amount of damage back to them.

*Glory Seeker: These enemies have no care for their own life and merely seek out glory in battle. So they're coded to actually go for your strongest units (these enemies would typically be slightly more powerful than regular enemies so they don't essentially become the Palestini Suicide Squad from Life of Brian. Think early game promoted enemies).

*Vengeful: Goes after units that have attacked them first. Like the way gambits and giant enemies work in Three Houses.

*Guerrilla: These are the units that wait for you to come to them and lay traps. They can get individual hard coded actions to take to best make the trap they lay in the map have maximum effect.

All enemy types will still go for a kill if it's freely available to them. Combined this would mean glass cannon units are at less risk of immediately dying. Tanks can actually tank things outside of the singular instance where they're the only unit you have in the enemy's range. It would also lean better towards games where you fight enemies on equal footing with your units instead of fighting hoards upon hoards of easily dispatched soldiers who can barely scratch you.

And to add to this idea, I'd bring back the idea of Commanders seen in previous games. Only this would be a new twist. The enemy Commanders have the ability to override the personality of other enemies in their range. Enemies under the commander will act in unison and more able to prioritize on a single goal, but take out the enemy commander and suddenly they're without leadership, so they revert back to their own AI and things get more chaotic.

I guess ultimately this might mean that instead of a singular AI that's easy to cheese and predict we'd get multiple AIs that are easy to cheese and predict. But I think that could potentially make things more interesting. What do you think about shaking up the AI formula of Fire Emblem?

Edited December 12, 2019 by Jotari

[henrymidfields]

I wouldn't mind if the AI acts a bit more like, as you said, more human, with them actually "thinking" about minimizing their losses and maximizing their objective fulfillment. I did notice, though, that at least for Fates Conquest, baiting the enemy was generally more difficult, as certain groups now act together. Another thing to note is that, some enemies just did not bother to initiate attack if they deemed it ineffective (eg most melee units against Effie). These two, I think, are a step in the right direction to discourage turtling?/baiting? play. It may be worth focusing on improved AIs (along with enemy weapons, stricter victory objectives etc) on higher difficulties for future games, instead of mainly having to face enemies with inflated stats.

Plus, if your proposal above facilitates an more even footing footing of enemy units vs player units in chapters then more power to the dev team. This, I think should be the norm, as Three Houses was explicit about each units resembling battalions. And then, any battle which has one-sided number differences should be justified from the dictating plot in the chapter, and the objectives also defined accordingly. It could be an ambush chapter where the heroes need to escape to friendly territory, or a stealth chapter where the heroes need to elude the enemies, or a defend chapter where the heroes need to survive until the cavalry comes to the rescue.

Edited December 12, 2019 by henrymidfields

[Elephantus]

Interesting idea. Personally, I'd say yes, but only if, as you mentioned, the player has a way of predicting what the opponent will do. Maybe that's just me, but I tend to play in a style that minimizes risk while still being as efficient as possible, and if the AI is too unpredictable, that playstyle goes out the window.

[Eltosian Kadath]

This seems like a cool idea, but they would have to be careful with the implementation of it.

 

3 hours ago, Jotari said:

 

Every enemy in the game has a personality type. Ideally this should be something visible to the player on a stat screen, though with Three Houses new aggro detector it doesn't necessarily need to be, as enemies personalities could be figured out based on that. Basically instead of having one universal AI for almost every enemy in the game, there are several. For example.

Making it visible to the player would make it far easier for the player to manipulate the AI, but than again if you can't easily distinguish the AI types

 

3 hours ago, Jotari said:

 

And to add to this idea, I'd bring back the idea of Commanders seen in previous games. Only this would be a new twist. The enemy Commanders have the ability to override the personality of other enemies in their range. Enemies under the commander will act in unison and more able to prioritize on a single goal, but take out the enemy commander and suddenly they're without leadership, so they revert back to their own AI and things get more chaotic.

this incredibly cool idea will feel as empty as it did when they tried it in Genealogy of the Holy War, when some commander would give allies clever mode AI, as it was too hard to notice.

[Jotari]

8 minutes ago, Eltosian Kadath said:

this incredibly cool idea will feel as empty as it did when they tried it in Genealogy of the Holy War, when some commander would give allies clever mode AI, as it was too hard to notice.

Potentially. It would really depend on the execution. Clever Mode AI wasn't really all that well implemented, though I do appreciate the fact that they tried to stir up the enemy AI at least once (it is what inspired this post, though honestly I think it's pretty much accepted that the standard AI in Genealogy is harder to deal with as it's slightly more esoteric. Neither setting never really seemed to make a difference to me).

Edited December 12, 2019 by Jotari

[NinjaMonkey]

6 hours ago, Jotari said:

*Pragmatist: Typical Fire Emblem AI, goes for the enemy they can do the most damage to.

*Confident: Also prioritizes attacking the enemy they have the best chance of actually hitting (good for enemies with powerful but inaccurate weapons).

*Coward: These enemies prioritize their own survival and thus target your units that will deal the least amount of damage back to them.

I've seen all three of these in Radiant Dawn to some extent, so this is nothing new.

6 hours ago, Jotari said:

*Glory Seeker: These enemies have no care for their own life and merely seek out glory in battle. So they're coded to actually go for your strongest units

Which would usually be your Lord unit i.e. what normally happens anyway. Also, how do you define "stronger"? Is Soren with capped magic (40) stronger than Ike with capped Strength (37)?

Edited December 12, 2019 by NinjaMonkey

[eclipse]

Hmmm. . .

The thing about ANY AI type is that it becomes predictable, once we know what to look for.  I like the variety, but perhaps an enemy or three per map should have randomized AI - just enough to make a player stop and think, but not enough to overwhelm.

1 minute ago, NinjaMonkey said:

Which would usually be your Lord unit i.e. what normally happens anyway. Also, how do you define "stronger"? Is Soren with capped magic (40) stronger than Ike with capped Strength (37)?

Highest level?

[NinjaMonkey]

6 minutes ago, eclipse said:

Highest level?

That still doesn't help things much. Is a unit promoted at level 20 considered to be at a higher level than another unit promoted at level 10? Or are they both treated as being the same level?

[Boomhauer007]

Variable AI is a neat idea but would probably take a very long time to implement into every map. 

I remember the creators of wargroove talking about how it's basically impossible to make AI that can even come close to competing with a person in games like these, and that realistically all you can do is make the situation difficult or give the AI an advantage that you have to overcome. IIRC it was largely that programming dynamic AI is something that was either impossible or extremely difficult. 

I'm sure someone else can talk to it better than I can.

[Flere210]

If they programmed Deep Blue, they can Program FE Deep Blue. It's just that playing againist FE Deep Blue would be hell on Earth.

[Light Strategist]

Heck apply it to all units and auto-battle might be a more viable option.

[X-Naut]

I'll echo the point about making the AI somewhat human. At the very least, give it enough common sense to not attack foes it can't hope to harm and back up archers that are up against a unit they can't attack. Preferably, give it the sense to think a move or two ahead so that it can work around simple bait traps.

That being said there's a limit to how human an I can be, and I'm not talking in terms of programming. Having the enemy bunch up into a singular group every map would not make for a fun game.

[Light Strategist]

36 minutes ago, Light Strategist said:

Heck apply it to all units and auto-battle might be a more viable option.

Y'know, after thinking on it a little more, I realize it might be wiser to implement certain strategies for the AI to use and have them assigned to a combination of the stat values and their personality. If you have the enemies with a certain personality in a position you think you have him, he could have a different strategy based on his stats and perhaps remaining health. For example, a Pragmatist AI might switch to a Coward's strategy of retreating if its Defense or Resistance is lower than its Speed while its HP is lower.

1 hour ago, NinjaMonkey said:

That still doesn't help things much. Is a unit promoted at level 20 considered to be at a higher level than another unit promoted at level 10? Or are they both treated as being the same level?

Fates treated promoted units as 20+ current level. So a level 5 Hero is considered level 25. Maybe do that?

1 hour ago, NinjaMonkey said:

Which would usually be your Lord unit i.e. what normally happens anyway. Also, how do you define "stronger"? Is Soren with capped magic (40) stronger than Ike with capped Strength (37)?

Base it around the offensive stat of the currently equipped weapon. In this case, yes as Soren's most offensive weapon being a tome would put him higher than Ike's most common offensive weapon. Things could be shifted by giving Ike a Sonic Sword though. That shifts your opponent's strategy and makes you have to plan around those unit types when using your strongest units.

[Flere210]

I think the best is to program a coherent strategy, for example in an escape map the red units can move around trying to encircle your units, or in a kill boss chapters the bulkiest units can move in a block around the boss and would only break the formation if they have 100% chance to kill something. Armor knights in general could be coded to bodyguard frail ranged units, and archer may not move toward your army unless covered by melee guys, so they don't get killed by pegasus knight in an embarassing way.

Edited December 12, 2019 by Flere210

[Jotari]

4 hours ago, NinjaMonkey said:

I've seen all three of these in Radiant Dawn to some extent, so this is nothing new.

Which would usually be your Lord unit i.e. what normally happens anyway. Also, how do you define "stronger"? Is Soren with capped magic (40) stronger than Ike with capped Strength (37)?

Take you pick. I'm just throwing stuff at the wall to see what sticks.

Although I don't think in the current system if you put your cleric and lord in range of the same enemy, they will go for the cleric. Except maybe in Thracia. I think every enemy is hardcore's to go for Leif in that game.

[eclipse]

9 hours ago, NinjaMonkey said:

That still doesn't help things much. Is a unit promoted at level 20 considered to be at a higher level than another unit promoted at level 10? Or are they both treated as being the same level?

There's multiple ways this could go, with no right answer.  So, uh, choose what you think is best!

[Jotari]

Internal level has been a thing since at least Awakening. And three houses doesn't even bother to level reset on promotion. It's really not an issue.