The truth about AI

[The Former]

Once again, I find myself having to explain the economics of video game development. Please don't take this for a personal attack on the intellect of those who disagree with me, I merely mean to elucidate so I can make a further point.

When you develop a game, you spend money. You're paying man-hours for labor. Concept, art, design, programming, bugtesting. Then you've got your overhead. Office utilities. Do you have any idea what the power cost is for a studio with even several dozen computers that are running hard for most of every day?

The point is, making a game can get expensive real fast. While the game is being made, you're not bringing in new money. This is a major reason that development timeframes are established. In theory, a game could keep developing for as long as it needed to in order to become damn near perfect in every way. But the studio would likely go broke by that time.

Normally I use this to explain why DLC exists, but today I'm going to use it to talk about AI. A common complaint every time a new Total War title rolls around is that the AI hasn't improved. But the truth of it is this: Either we can have a new Total War game, or we can have the same Total War game we just got with better AI.

What do I mean? You see, when you make a new game from the ground up - new units, new mechanics, new terrains to contend with, what have you - your AI team has to account for all those new things. They have to make sure the AI knows how to use every unit against every other unit. Which terrain features to contest and when. Which paths to take through new city layouts during a siege. All that good stuff.

That consumes development time. See where I'm going with this?

If CA gave us Rome 2, and then four years later gave us the EXACT game, say... Rome 2+. In this case, the AI would have an opportunity to improve, because they'd be able to dedicate four years of development time to refining that AI. But how many people would be happy about getting the exact same game with better AI?

"They should just upgrade it for free." And burn through their operating budget for no monetary returns? Financial suicide. Then there are NO more Total War games. Or worse, they become a subsidiary and have less control over how their games are made.

The point I'm trying to make is this: Game development is more costly and more difficult than you might think. Open your mind to what the developer has to go through to get this game to us, and you might see it's not as bad as we think. They're actually doing a great job on AI, all things considered. For one thing, let's look at the fact that Total War is one of the FEW franchises that has completely generated maps. These are GREAT! They make every battle new and fresh.

But when you don't have set terrain features that are always "good" and "bad", programming the AI to make use of the terrain is a challenge. You can't just tell them to "go to A and hold", you have to try and "teach" the AI what types of procedural terrain are favorable to make a stand on, and hope that the AI decides one specific hill is better than another on a map with two big hills.

Just... My two motes on the matter.

[Aeternull]

You forgot to mention two things:
-AI will never be better than a human (unless it cheats)
-Sometimes it's really un-fun to playing against a very smart AI. Despite what many believe, people play games to win.

[Omegabigb]

Originally posted by Aeternull:

You forgot to mention two things:
-AI will never be better than a human (unless it cheats)
-Sometimes it's really un-fun to playing against a very smart AI. Despite what many believe, people play games to win.

Additionally while I partially agree with the OP's post it does make it sound like the AI never improves, the AI has indeed improved, it just still isen't as smart as a human (because no AI is).

[The Former]

Originally posted by Aeternull:

You forgot to mention two things:
-AI will never be better than a human (unless it cheats)
-Sometimes it's really un-fun to playing against a very smart AI. Despite what many believe, people play games to win.

Very true. Play against a machine that's literally programmed to outmaneuver its opponent and does so perfectly well, and you'll lose every time. Computers compute faster than humans.

[The Former]

Originally posted by Omegabigb:

Originally posted by Aeternull:

You forgot to mention two things:
-AI will never be better than a human (unless it cheats)
-Sometimes it's really un-fun to playing against a very smart AI. Despite what many believe, people play games to win.

Additionally while I partially agree with the OP's post it does make it sound like the AI never improves, the AI has indeed improved, it just still isen't as smart as a human (because no AI is).

It absolutely does, I was a bit unclear on that. What I meant to say was that we'll never see remarkable overhaul-level improvements in AI unless we get a "copy" once or twice. Over time, the AI absolutely as improved. I still remember it trying to charge in Medieval 2...

[Omegabigb]

Originally posted by Sheriff of Nothingham:

Originally posted by Omegabigb:

Additionally while I partially agree with the OP's post it does make it sound like the AI never improves, the AI has indeed improved, it just still isen't as smart as a human (because no AI is).

It absolutely does, I was a bit unclear on that. What I meant to say was that we'll never see remarkable overhaul-level improvements in AI unless we get a "copy" once or twice. Over time, the AI absolutely as improved. I still remember it trying to charge in Medieval 2...

Oh you are absolutely right about that, but even then, if like you say we get a "copy" or alternatively no TW game at all for 8 years, the AI still might not be as good as people want, AI is bound be the technology of today.

[OZZO]

All I can say, it has to be a balance. It is a strategic game. If the AI is poor, even a beautifully crafted game will not be successful.
I have minor concerns about AI to be honest when I see how people in the streams win on Legendary difficulty every time. I really hope that with a bag of gold developers already got from presale the AI team is already engaged for further improvements.

[Omegabigb]

Originally posted by OZZO:

All I can say, it has to be a balance. It is a strategic game. If the AI is poor, even a beautifully crafted game will not be successful.
I have minor concerns about AI to be honest when I see how people in the streams win on Legendary difficulty every time. I really hope that with a bag of gold developers already got from presale the AI team is already engaged for further improvements.

The thing about AI's is that they are predictable, so if you know how to beat them on legendary, you will always be able to beat them on legendary, and this cannot really be "fixed" completely since it's just in the nature of AI technology.

[arthur jean]

there could be a group system for multiple stacks in campaign for a compromise of bad AI, bc then at least the AI stacks of large factions wouldn't be beaten one by one repeatedly. AI of Nobunaga's ambition: iron triangle looks better bc once AI's factions become big, they tend to invade with all their troops altogether. if players are overwhelmed by numbers of AI in a single battle so it's challenging to defend, they might less whine about ai I guess

[Omegabigb]

Originally posted by arthur jean:

there could be a group system for multiple stacks in campaign for a compromise of bad AI, bc then at least the AI stacks of large factions wouldn't be beaten one by one repeatedly. AI of Nobunaga's ambition: iron triangle looks better bc once AI's factions become big, they tend to invade with all their troops altogether. if players are overwhelmed by numbers of AI in a single battle so it's challenging to defend, they might less whine about ai I guess

I mean this happens TW too, the AI can have bigger/more stacks than you, but the main complaints are not about the campaign AI but rather the battle AI.

[Wenatchee Willie]

Originally posted by Sheriff of Nothingham:

Originally posted by Aeternull:

You forgot to mention two things:
-AI will never be better than a human (unless it cheats)
-Sometimes it's really un-fun to playing against a very smart AI. Despite what many believe, people play games to win.

Very true. Play against a machine that's literally programmed to outmaneuver its opponent and does so perfectly well, and you'll lose every time. Computers compute faster than humans.

Not if you use pause to give the human brain time to think. Which is why I use pause.

[viranto]

In a game like chess, the ai can be very good. Often too smart. I would never play against a smart ai, then i would lose always. A computer can think better than a human brain.
But i wish me a ai, with different templates. Why are there not like 50 templates, which the ai can choose from? she don't have to think. The Ai needs only templates. Like Where are the cavalry of the player, where are the archers, are they in forest, on mountains, etc? And when one of the template match with this what the player does, the ai can act to it. The ai should only check every 10 sec in the battle, what the player does and choose from 1 template. Always.
All of these templates can save from a human in. And the ai must never think. Only choose the correct thing and check it regularly. When there a lot of templates, it would feel like a thinking ai.

[arthur jean]

Originally posted by Omegabigb:

Originally posted by arthur jean:

there could be a group system for multiple stacks in campaign for a compromise of bad AI, bc then at least the AI stacks of large factions wouldn't be beaten one by one repeatedly. AI of Nobunaga's ambition: iron triangle looks better bc once AI's factions become big, they tend to invade with all their troops altogether. if players are overwhelmed by numbers of AI in a single battle so it's challenging to defend, they might less whine about ai I guess

I mean this happens TW too, the AI can have bigger/more stacks than you, but the main complaints are not about the campaign AI but rather the battle AI.

this should happen more often e.g. 3 stacks of AI vs 1 stack of player like mongol invasion in M2. AI Cao Cao always leading full 3 stacks would be more threatening and challenging to defeat. OP said AI improvement inevitably has a limitation for those reasons

[zgrssd]

One large part of "Artificial Stupidy" comes down to one fact of gaming AI:
It has the planning abiltiy of a falling slice of Pizza.
And this side of AGI that will not change anytime soon.*
It is not a mater of skill. We literally do not know the math for it. And without knowing the math, we have no hope of teaching a computer how to do it.

All those Economic Cheats, Rule Cheats and unit spawns? Those are meerely there to compensate for that issue.
They are "can not plan" offsets.

*There are a few games where we amanged to simulate planning ehaed:
- TicTacToe is so simple, a basic weighting can emulate "planning".
- With chess we have literal centuries worth of played games, to tree search across towards a winning board
- Itterated Destilation and Amplification can get us a "simulates planning" Network.
But this side of purely turnbased 2 player games, all bets are off.

[jaersavo]

I have the feeling that they don't even try tp make the ai appear smarter. There are many "cheap" ways to do it, without actually that much work. One would be to add variation. Playing against the ai you are never suprised by what they do. Give them a template to randomly choose from.
battle has:
chance to be normal
all cav left/right flank
set ambush on map
Take strongpoint and stay defensive

Honestly, this is not much work and would give the player tactics to react to. At the moment you only react to strategy and battle development/where are what reinforcments needed

[Omegabigb]

Originally posted by viranto:

But i wish me a ai, with different templates. Why are there not like 50 templates, which the ai can choose from? she don't have to think. The Ai needs only templates. Like Where are the cavalry of the player, where are the archers, are they in forest, on mountains, etc? And when one of the template match with this what the player does, the ai can act to it. The ai should only check every 10 sec in the battle, what the player does and choose from 1 template. Always.
All of these templates can save from a human in. And the ai must never think. Only choose the correct thing and check it regularly. When there a lot of templates, it would feel like a thinking ai.

This is an over simplification the AI needs to be able to do way more than just choose templates.
it also needs to be able to take terrain into account, and needs to be able to judge timing, when is it neccesary to charge etc.

also those templates in your scenario only work if the AI can see all your units, you'd still be able to trick the AI by hiding some of your units.

I agree that having more tactical variety is a good thing, but it's not the only part of making a good AI by a long shot.