AI or not AI, that is the question!

[andrey12345 v2.0]

The problem that worries many players is where this notorious AI is hiding and what is he doing in the game. Judging by the numerous posts, players believe that AI is everything that the player does not do: path finding, what projectile to shoot, kick the enemy or throw a grenade or shoot at him, lie or run, turn front of the hull to this AT gun or to that guy with a grenade launcher... write a "script with the FSM" at the level of a building elevator - yes, this is definitely AI!
But this all obviously has nothing to relate with AI. But due to the fact that in most games there is no AI at all, but from a marketing perspective it looks like a profitable feature, they come up with this term calling it whatever comes to hand. We have pathfinding a little more complex than walking in a straight line - let's call it - advanced AI! We have an intelligent choice from two options which projectile to shoot at the enemy - let's call it advanced tactical AI!
Developers can be understood, AI is an extremely unpleasant thing in practice, but for the players it sounds attractive.

But our goal is different, to figure out what this AI is.
Here another difficulty awaits us, since outside of games this is also a difficult marketing issue: a washing machine with an element of fuzzy logic has also become AI since some time. And if somewhere the "neural network" phrase appears then this is generally 100% AI. Marketing is hitting the AI horse with all the might elsewhere.

But let's look at it from a common sense point of view, removing all the marketing nonsense.
If we enter a battle a kitchen and see a dangerous enemy who shoots at us touch a hot frying pan with our hand... aim the weapon at him / hide for cover we pull our hand back ... Can we say that it was a manifestation of intelligence?
Obviously not, this is a conditioned reflex, it will be exactly the same in worms and other animals not burdened with intelligence. Why do we call it AI in games? Or maybe even an advanced tactical AI? The question is, of course, rhetorical.

We want to go down in the elevator from the 5th floor to the first, the elevator goes up and does not stop, and when it goes down it will stop. And if the weight of all those who entered exceeds the norm, it will not go further. Could it be AI? Hardly anyone in their right mind would say that. But in games, the same FSM is always presented as a super cool tactical AI. Heck knows why.

There are a lot of such examples, and it never occurs to anyone to call them intelligence/AI, except for games.

A natural question arises: where is this AI?
It's very simple, AI is always making decisions in conditions of uncertainty without human participation and without strict algorithm.

Let's imagine a game where one human player makes game scenarios for another human. And the second one plays according to these scenarios.
We call this singleplayer story-driven games (or newfangled nonsense - asynchronous multiplayer!). Now let's go further - scenarios are made at the request of the second player and taking into account some initial conditions. And the last step is to replace the first human player with a computer!
Well, here it is, our game AI, with a clear and specific task.

[Headshotkill]

İlk olarak andrey12345 v2.0 tarafından gönderildi:

İlk olarak Sandor tarafından gönderildi:

Yeah, I'm not sure if I have free will.

No one can be sure of this, because this concept does not have a clear robust definition.
But nevertheless intelligence (at least natural) definitely exists, the proof is at least this topic and its participants.
This means that we can define this concept quite clearly, and then transfer it to the virtual world (in the game), and there it will remain no less clear. The main thing is not to give it unnecessary unimportant properties like "free will".

It is done like this:
Let's take a game where two human players play against each other. What we can say for sure is that both sides have intellect, it helps to perform some actions in the game in pursuit of certain goals.
For example, if the games are in the RTS / Wargame genre, then these actions are orders to units, depending on the situation and tasks.

Now let's replace one of the players with a computer. If the game continues to function as before (i.e. orders are given to units, some goals are pursued), then we can say by induction that the intelligence still remains on both sides. But in the second it became artificial, not natural.
That's all.

Or in short a turing test.
In order to proof this game has good AI we would need a multiplayer mode and compare that to singleplayer. Or better have the player not know if he's playing against a player or a computer, if he can't notice the difference you have an excellent AI.

[andrey12345 v2.0]

İlk olarak Headshotkill tarafından gönderildi:

Or in short a turing test.

No, this is not a Turing test, we do not need a computer to be indistinguishable from a human (there is no this goal), it is enough for us that it performs the functions that a human performs.

A simple example: a vacuum cleaner robot is radically different from a human with vacuum cleaner in hands, but it is enough for us that robot autonomously cleans dust throughout the apartment using AI for navigation. There is absolutely no need for him to be indistinguishable from a human.
This is generally a strange delusion that came from the last century, that AI should be similar to a human, as if the 8 billion people who already exist are not enough for us and we urgently need to make more artificial people. This is obviously stupid, like all these rules of robotics Asimov style and other strange notions.

İlk olarak Headshotkill tarafından gönderildi:

In order to proof this game has good AI we would need a multiplayer mode and compare that to singleplayer. Or better have the player not know if he's playing against a player or a computer, if he can't notice the difference you have an excellent AI.

We don't need to prove that there is good AI somewhere (what does "good AI" mean?!?), we need to find out if there is AI (any) in the game or not.

[MarkShot9]

What I want to know is while Fritz is on his knees with just 15 seconds left saying, "Ich bin getroffen!". Is he suffering? Is his pixelated life flashing before his eyes? Is he mumbling "mutie"? Does he fear the void?

It doesn't seem so. The game is completely unrealistic. It's very hard for me to take combat seriously in such a shallow representation. Yes, graphics true, but where is the emotive AI?

:)

[pfmm]

"Good AI" can have a meaningful metric in Graviteam Tactics, and that is how much a player is convinced it will provide a challenging contest (given the forces at hand), and also that the player is up against something other than a bot (with all the poor connotations that the word bot can mean).

For me, Graviteam Tactics' AI infrequently challenges and is still a bot.

[MarkShot9]

How does one recognize a "bot"?

The OPFOR takes the high ground. Is that a human or a bot?

The OPFOR fails to take the high ground. But takes the woods below the high ground which is a bald ridge. Is that a human or a bot?

The OPFOR makes a costly and ill conceived frontal assault. Is that a human or a bot?

---

I think we see the "botness" of many games mainly due to the lack of variability; not that humans don't become predictable. With just a few executions it is hard to spot a bot.

---

Finally, in the real world, I think lack of variability is less likely to be a "bot". Why? If your combat tactics work, the enemy dies. The opportunity to realize invariant behavior is reduced by mortality.

[pfmm]

Bots are easy to spot due to their predictability. That's why they are called bots, they do one thing.

[andrey12345 v2.0]

İlk olarak pfmm tarafından gönderildi:

For me, Graviteam Tactics' AI infrequently challenges and is still a bot.

AI in tactical mode does not have such a goal - to make a challenge.
In order for the AI to make a challenge, two things are needed:
1) Strict win/lose rules
2) The balance of troops between opponents, which is strictly maintained
For obvious reasons, both points are unacceptable in a wargame that tries to reproduce the historical course of events.

And AI is a bot anyway because it is AI.

[Zephyr]

İlk olarak andrey12345 v2.0 tarafından gönderildi:

İlk olarak pfmm tarafından gönderildi:

For me, Graviteam Tactics' AI infrequently challenges and is still a bot.

AI in tactical mode does not have such a goal - to make a challenge.
In order for the AI to make a challenge, two things are needed:
1) Strict win/lose rules
2) The balance of troops between opponents, which is strictly maintained
For obvious reasons, both points are unacceptable in a wargame that tries to reproduce the historical course of events.

And AI is a bot anyway because it is AI.

Your point 1) and 2) are true, of course, but in the game it is still possible to apply them. Just the conditions need to be right or alternatively an evaluation can be done how well assets are used under the specific given conditions (at least roughly).

1) The AI has often enough the goal "take this square", even if it does not want to win.

2) Often forces are very unequal, but then the test for "challenging/good" AI could simply measure how many casualties the more powerful side suffers vs. the less powerful side. This is basically already done in the game.

And as my favourite part 3) To really compare AI performance (and by that how challenging it could be) with the player it would be essential that all test-players play strictly with the same set of game rules the AI uses.

[andrey12345 v2.0]

İlk olarak Zephyr tarafından gönderildi:

but in the game it is still possible to apply them. Just the conditions need to be right or alternatively an evaluation can be done how well assets are used under the specific given conditions (at least roughly).

This always leads to player questions like "I killed everyone, why did I lose?"

[Zephyr]

İlk olarak andrey12345 v2.0 tarafından gönderildi:

İlk olarak Zephyr tarafından gönderildi:

but in the game it is still possible to apply them. Just the conditions need to be right or alternatively an evaluation can be done how well assets are used under the specific given conditions (at least roughly).

This always leads to player questions like "I killed everyone, why did I lose?"

That is not displayd openly anymore, so no problem :). But everyone can compare how much personel, vehicles and guns are lost and then make a ratio from that using own/enemy losses and power ratio. Killing everyone but losing as many men when the actively used forces (that is another important parameter) ratio is 5:1 just is not especially great.

I wrote this just for the hypothetical case that somebody wants to recruit 1000 players to play the same battle and then collect all relevant data to make statistics which might be used as tentative indicator to measure the quality of the Mius AI :).

[andrey12345 v2.0]

İlk olarak Zephyr tarafından gönderildi:

That is not displayd openly anymore, so no problem :).

This is drawback again! - it is not clear who won and who lost.

[Zephyr]

İlk olarak andrey12345 v2.0 tarafından gönderildi:

İlk olarak Zephyr tarafından gönderildi:

That is not displayd openly anymore, so no problem :).

This is drawback again! - it is not clear who won and who lost.

You can probably see it that way. It is not as straightforward as before, but it is still quite possible to see which side was better (call it win/lose if you want).

[Pandelume]

IMHO, having a "winning" (for what winning means) AI is not so much important as having a credible AI. By credible I mean using by the book tactics. Using supporting fire, keeping a reserve, using covered approaches to the form up point, using combined arms, having flank protection...

[Headshotkill]

İlk olarak Pandelume tarafından gönderildi:

IMHO, having a "winning" (for what winning means) AI is not so much important as having a credible AI. By credible I mean using by the book tactics. Using supporting fire, keeping a reserve, using covered approaches to the form up point, using combined arms, having flank protection...

I feel as if this is the main point the players want to adress, the AI doesn't have to perfectly play the game as an RTS. But it could function as a slightly more competent commander back in the day than it does now.

[KingMuzza]

OMG I got to the end of the thread.
Thank you for a very interesting Fathers' Day morning read everyone above.

I'd love to throw my 2 cents in the ring, and I'm going to post this in 2 parts because the discussion seems to me to be in 2 parts. ...