"People playing the game don't care if the game is using a huge database of scripted decisions, is making directed searches of a decision tree, or is building an accurate knowledge base of it's surroundings and making inferred choices based on logical rules...." -GameDev.net "AI Game Engine Programming" Here's what get me on this subject. In the excerpt stated above, the writer implied that, no matter how "intelligent" an NPC is when it comes to artificial (scripted) logic, reasoning, etc., what really matters to players is whether that grenade will hit them if a hair atop their head is exposed. So, what I understand here is that, it's unimportant whether an enemy AI with a grenade launcher "thinks" about his advantage or not and reacts accordingly. But, what matters here, I assume, is the percentage of how many shots are accurate. Basically, the precision of the AI's targeting is what displays it's intelligence.... and not how the weapon is used. The reason I bring this up is due to the fact that, in any game, I wonder which scenario would be more interesting to experience: A. The player retreats to a nifty spot behind a container, and thereby gains an advantage over the BFG-wielding enemy. Typical duck-and-fire battle ensues until enemy is eliminated. B. The enemy surveys the situation and decides, not to fire directly at the player (Since he/she is well hidden behind an obstacle), but to aim slightly to the right- that the impact against the wall would thrust the player into the open momentarily. Here's a classic scenario that challenges the statement made above concerning what's really important when it comes to implementing "believable" AI. However, more information on the subject of AI could be found at: Http://downloads.gamedev.net/pdf/AIGameEngineProgCh1.pdf I simply decided to post on one particular statement, but any other feedback, opinions, extra knowledge, etc., would make for an interesting discussion.

I think you are reading too much into that sentence. I think all the author means is that the player doesn't 'see' the techniques or methods applied in designing the AI, they only see the result.

If game 1s gameplay/result is better than game 2, it doesn't matter if game 2 has more complex/advanced/modern/correct AI. It failed to provide a better experience.

Consider the long-standing First Rule Of Game AI: It doesn't have to be intelligent, it only has to appear intelligent. That is, it doesn't matter what is under the hood of your game engine driving your NPCs behaviour, provided that to the player, within the context of the game, that behaviour appears as if directed by an intelligent agent. So, the player doesn't care if it's a Script driving the transitions of a Finite State Machine or a dynamic decision tree adjusting its classification based on the latest perceptions of its avatar. If it does what you want it to, when you want it to, then that is good enough (usually).

I think that is what is being suggested here... but then, the quote is out of context, because we don't have the other information (unless we choose to read that chapter of the book). Personally - and remember, this is just my personal opinion - I didn't like this book.

Cheers,

Timkin

A good example of this might be the special forces AI in the original Halflife. Those guys were really very simple.

They had a maximum number of them that would come out to shoot the player at a time. They would hide when their clip was empty so somebody else could shoot in their place while they reloaded. And that's more or less it. Still they appeared to act pretty intelligent (at least, to me).

Edit: Ok, I'm actually not 100% sure about this being correct, but I read this somewhere, can't quite remember and don't really feel like checking right now (It's 04:00 over here).

He sure is talking about not matter the technique you use but the results you get...

In my opnion this is true in the way that you must program something that you fell confortable with it... If you don't know how to use a decision tree, better go with a rule based or something else...

And from my little experience... machine learning algorithm are a lot more dificult to tune and make a good gameplay...

In my college some guys implemented a Street Fighter like game and the NPCs were implemented using Reinforcement Learning... and after no more than six, seven fights the NPCs were unbeatable (don't know if it is write).... If time they implemented a way that the NPCs forget about some knowledge if the perceived that he was too good for the human to play... the game is pretty good now, but it was a lot of work

Timkin,

I was wondering why you didn't like the book. I was thinking of getting it...and just wondering why you said that.

Thanks.
Lederhosen