2
$\begingroup$

I'm about to write a non-player character (NPC). I wonder how much the AI should know about the game's world. So, my question isn't about the amount of training data the AI has to collect. I'm interested in how much the AI is allowed to know about what's going on in the game's world. For example, can (shall) it have knowledge about the build queue of the player?

To provide more details: while a human plays a game against another human, not all information of what the opponent is doing is available (e.g. the queue of the units your opponent is building). This could give you an advantage (so that you can prepare for a rush, when he's building many cheap units). Theoretically, an NPC could access and make use of that knowledge and, in addition, spare resources for scouting/spying/exploring.

But is this the way of constructing an NPC AI? Or should this data also be restricted? I have never done anything like this before.

I don't know where to ask else wise or what more information I could provide. So, if something in my question is unclear or unfit, please let me know what exactly.

$\endgroup$
1
  • $\begingroup$ I am not an expert on the topic, but my intuition says that you could have different versions of the AI with different levels of difficulty, which the human player can set (or could be aware of). However, I think you should wait for the answer of someone that has some experience in the field. $\endgroup$
    – nbro
    Sep 14 '20 at 10:26
1
$\begingroup$

Art of NPC creation

I'm assuming this is a standard game, not a game theory application. These AIs tend to be far simpler in theory than actual artificially intelligent agents which are used to solve real world problems. The challenge of building games is that there are few right answers. A game NPC opponent could know practically nothing of the player and be guided by random world exploration activating certain characteristics (attack, strategy) whenever a human is in range or on the other end could know basically everything and have some limiting factors built into the code which reduces their challenge level to the player. This is the art aspect of gaming design. NPCs with a high level of complexity could learn from their interactions and build a "pattern" with some internal reward scoring system for success (similar to reinforced learning). This would represent a very challenging game NPC, especially if the strategy changes upon failure. Think about who will be playing your games, the audience that you would tend to draw, and hone the NPC challengers to them. Hardcore gamers deserve a challenge, and hobby gamers deserve a chance to win. Where that balance between rewarding game experience and challenging game experience lies depends on the 'feel' which you are cultivating as the game designer. You mentioned resources and exploration as elements of the game. For exploration types of games, you may rely on environmental interactions to offset player interactions. A distracted NPC (gathering resources) is less likely to heavily attack the human player unless this player enters the perimeters of the NPC base or scouting area. This may work to keep the gameplay balanced. Keep in mind that some limitations for your NPCs must be absolute, or the game will experience unexpected behavior (bugs). A completely unrestricted NPC is a liability. If the NPC goes too far from the player or gets stuck, it completely ruins the feel of your game.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.