Simple question?

R

Risuli

Veteran XX
Question is, why no A.I.?

Currently I've been playing The Division with its rather impressive graphics, giving a very good sense of immersion within a large cityscape. You can shoot the tires and windows out of cars, etc. etc. etc.

This got me (once again) considering the fact that graphics have made advances in leaps and bounds over the last 20 years or so. Yet, this same cityscape is populated by a handful of "citizens" doing the same three or four actions on pretty much every block, followed by the gang NPC's that either charge you, or sporadically use cover but otherwise stand up and get shot. The bosses themselves pretty much just come straight at you with loads of armor and a more damaging than normal weapon. Rinse, repeat.

So, since graphics have advance so much in twenty years, why in the world hasn't anyone apparently invested any effort in trying to put forth a realistic A.I.? Something at least somewhat realistic, like fighting retreat, breaking and running if overpowered, or individual NPC's with somewhat of a broader range of action so they aren't just all the same?

Best game that I recalled at least trying something along those lines was STALKER, but that was ages ago.

So, is it just too tough or no one gives a sh*t?
 
Last edited:
cuz ai is hard to do
only a couple weeks ago an AI was able to beat the best go player in the world, and thats a board game

what could you possibly expect from shitlords like ubisoft
 
Its kind of like trying to simulate *real* physics.

for 20% effort, you can fudge things to make them feel good in a game, or you can dump time and money into being ultra realistic, which the player A. Will never experience 99% of and B. Might not even like the realism they do experience over what you got for 20% effort.
 
Risuli said:
Question is, why no A.I.?

Currently I've been playing The Division with its rather impressive graphics, giving a very good sense of immersion within a large cityscape. You can shoot the tires and windows out of cars, etc. etc. etc.

This got me (once again) considering the fact that graphics have made advances in leaps and bounds over the last 20 years or so. Yet, this same cityscape is populated by a handful of "citizens" doing the same three or four actions on pretty much every block, followed by the gang NPC's that either charge you, or sporadically use cover but otherwise stand up and get shot. The bosses themselves pretty much just come straight at you with loads of armor and a more damaging than normal weapon. Rinse, repeat.

So, since graphics have advance so much in twenty years, why in the world hasn't anyone apparently invested any effort in trying to put forth a realistic A.I.? Something at least somewhat realistic, like fighting retreat, breaking and running if overpowered, or individual NPC's with somewhat of a broader range of action so they aren't just all the same?

Best game that I recalled at least trying something along those lines was STALKER, but that was ages ago.

So, is it just too tough or no one gives a sh*t?
Click to expand...

Easy answer: profit.

Like it has been said before, AI is hard to design and takes a lot of time. Creating some good AI for a game like Warcraft I was not that easy but the possibilities were pretty limited by a simple game with a lot of restrictions. It still took years though to create those games. Imagine designing a realistic AI for a game in a complex world...

To be honest, I consider game designers to be lucky because the mass consumer market vastly embraced multiplayer which removed a huge part of game creation. The "AI" was actually replaced with "other players". All that was left was to create an engine, levels, music, concept art... not that it is not important but far less complicated to create... and easy to make money out of it.

If people didn't get that interested in multiplayer, the history of gaming would be a lot more different.
 
Last edited:
Intelligence is more complicated than our shallow approximation of physics we use to create graphics. Half the pretty stuff you are seeing in games are post-processing shaders, they don't reflect how visual phenomena is made in real life, but it's convincing because there's not that many variables to how we interpret it.

Constructing an approximation of intelligence is much different, since there are so many variables, and not a very strict understanding of the underlying subject. We find it much easier to detect the patterns when they are not 100% aligned with what we expect naturally

Also, I think diminishing returns. We like games because we can find these patterns of interaction, cuz it simplifies how we play and gives us a path to gamify the system. Advanced AI doesn't enhance that experience other than for immersive games
 
Last edited:
Why do you need AI? Just up the damage to the enemies. That's all gamers seem to care about anymore.

The Division actually has some of the better AI in recent memory (which is not saying much, it is very basic).
 
Phantred said:
Its kind of like trying to simulate *real* physics.

for 20% effort, you can fudge things to make them feel good in a game, or you can dump time and money into being ultra realistic, which the player A. Will never experience 99% of and B. Might not even like the realism they do experience over what you got for 20% effort.
Click to expand...

I don't think it has to be an attempt at a full on simulation of what a human player might do, but honestly is it really that difficult to program in some more variable "human" traits into NPC A.I.? I would suppose in a "real" firefight some opponents would cut and run, go for back up, or into a building to heal (realizing that healing isn't something that is "real"). Some NPC A.I. could randomly be "aggressive", others offensively conservative, others cowardly, or some other more "human" variables. NPC gangs that have been pushed out of a area could attempt to come back with large groups to retake areas, etc.

I realize that multiplayer has, at least in part, let developers off the hook. But this game is meant for players to go against both NPC's and other players. You'd think they could at least attempt some more variability.

So, I guess the thought so far is that there are no programmers who even attempt to advance the science of A.I. in these types of games?
 
It can't be that hard for AI to advance at all. Honestly I can't think of games that have pushed things forward in many years. Half Life 2 had decent AI. Fear did as well. I'm sure there are others, but I can't think of too many I've played.
 
Risuli said:
I don't think it has to be an attempt at a full on simulation of what a human player might do, but honestly is it really that difficult to program in some more variable "human" traits into NPC A.I.? I would suppose in a "real" firefight some opponents would cut and run, go for back up, or into a building to heal (realizing that healing isn't something that is "real"). Some NPC A.I. could randomly be "aggressive", others offensively conservative, others cowardly, or some other more "human" variables. NPC gangs that have been pushed out of a area could attempt to come back with large groups to retake areas, etc.

I realize that multiplayer has, at least in part, let developers off the hook. But this game is meant for players to go against both NPC's and other players. You'd think they could at least attempt some more variability.

So, I guess the thought so far is that there are no programmers who even attempt to advance the science of A.I. in these types of games?
Click to expand...
Here's a little secret. No dev team ever creates its deadlines around AI. You have your milestones based on art, UI, gameplay features etc, because these are jobs that you do once, and you can estimate how long they'll take. And while you're doing these jobs, some people do AI. But when the rest of the game is done, you can bet your bollocks that nobody's going to wait around for AI improvements before shipping the game.


I have a friend who was hired at a game dev company to do AI. 2 months into the job he got taken off AI and put on GUI instead, because the GUI team needed the extra manpower to hit their milestones.


Really good AI takes fucking years to develop, and most projects can't afford to wait that long.

In conclusion:

tasks.png
 
You must log in or register to reply here.
Back
Top