Pan'Theon: Rise' of th'e Fal'Len - #1 Thread in MMO
- Thread starter gogojira_sl
- Start date
tyen
EQ in a browser wait time: ____
- 4,638
- 5,164
if gods strategized their trash mobs based on a guild's positioning? Shit dawg. Kill the minis to make the trash easier because you cut off lines of communication between their knowledge sharing.
All kinds of shit you can do with neural network machine learning now.
All kinds of shit you can do with neural network machine learning now.
DickTrickle
Definitely NOT Furor Planedefiler
- 12,938
- 14,841
There's about a 1% chance Pantheon will use any of those more interesting aspects of Unity in their work. Maybe I'll be wrong, though. Pretty much all trash mobs in any mmo ever have been super dumb so it would be nice to see even a tiny change.
- 3
- 518
- 210
1% is probably generous. Easy lower hanging fruit in progressive AI design improvements with their admitedly low resources. They are low staffed already whytf would they make something complicated, unpredictable, exploitable and incredibly expensive and difficult to test.
- 2
Dandai
<WoW Guild Officer>
- 5,907
- 4,483
You don’t actually want smart AI in games, though. You want predictable, consistent patterns that you can read and overcome. Imagine a dark souls boss that learned you preferred slow moving, heavy hitting weapons so it only used fast attacks? You pick up on this and switch to a faster weapon, so it starts using a shield.
This sort of system would make the game infinitely worse, not better.
This sort of system would make the game infinitely worse, not better.
- 10
- 1
DickTrickle
Definitely NOT Furor Planedefiler
- 12,938
- 14,841
They don't have to be unpredictable, but they can have a set of behaviors that are not the same every single time, or that change based on certain circumstances that can be evaluated or discovered. Some variance is welcome, imo. The phantom StoryBricks thing, if it had been real, would have been great. They would be actors with agency in the world but they don't suddenly become Deep Blue level chess masters after being idiots the day before.
I'm pretty sure "no one, ever" was talking about making them smarter than humans.
- 1
- 518
- 210
You want AI that is *smart* in a way that saves you (the developer) having to hard code content and enable reuse of content in novel ways without recoding and retesting. In other words more automation of basic (a fixed level of competative mob *inteligence* to guarantee base level play experience) rather than an increase in inteligence that affects the difficulty of challenge to the player. The AI needs to provide unpredictable richness of variety of play @ a natural feeling & just challenging enough to be fun (to player) difficulty without predictability.
AI is a bit of a misnomer in a sense for what is needed in MMOs to improve mobs, it is better automated, simple fantasy context, behaviour modeling than inteligence
AI is a bit of a misnomer in a sense for what is needed in MMOs to improve mobs, it is better automated, simple fantasy context, behaviour modeling than inteligence
- 2
- 518
- 210
But keep in mind whatever tools they have for the AI/content automation have to scale (with lots of mobs, and thousands of players) and be affordable to run on cloud. It has to provide predictable level service to all mobs reliably. Imagine the game server having a brain fart, literally, because players figure out how to max out available server resources for AI, like some old EQ exploits exploiting network trafick bugs, to *dumb* down encounters to be exploitable.
They have ONE guy on a tiny dev team who is working on this I believe. So... them needing to keep things limited to simple and *progressive* improvments is a 100% safe bet (no ML or anything like *real* AI) on this first try.
They have ONE guy on a tiny dev team who is working on this I believe. So... them needing to keep things limited to simple and *progressive* improvments is a 100% safe bet (no ML or anything like *real* AI) on this first try.
- 1
tyen
EQ in a browser wait time: ____
- 4,638
- 5,164
itd be a lot easier to tune encounters with pre-trained AI that utilizes strats that everyone uses to slay a dragon. Can get good estimates on how hard it will be to take down, learning curve, etc.
Maybe instead of some nerd coming up with Cheal rotations, AI could come up with a strat we never thought of. Then you force the users to figure out the winning strat to get the dragon loots.
Maybe instead of some nerd coming up with Cheal rotations, AI could come up with a strat we never thought of. Then you force the users to figure out the winning strat to get the dragon loots.
- 1
- 1
Nirgon
YOU HAVE NO POWER HERE
- 12,738
- 19,636
yamikazo
Trakanon Raider
- 1,361
- 546
Would be great.
Can you imagine Avatar of War? Oh you want to pull him to a safe spot? No, he's learned not to follow that one slimy monk to your giant horde of friends; he now hides behind 800 giants.
Avatar of War has also learned to kill clerics first. Also after he wipes your raid he corpse camps without despawning. Checkmate, motherfuckers.
Smart AI does not work in holy trinity games.
- 1
- 1
Denaut
Trump's Staff
- 2,739
- 1,279
You absolutely, 100%, do NOT want smart AI NPCs in an MMO.
NPCs are mini-puzzles that exist for the player to solve. You create variation by having more puzzles, and sometimes by combining them into more interesting meta-puzzles. There might be some use to AI "learning" but it would be as a tool to solve some of the more tedious aspects of content creation (like animating walk/run cycles) or for testing using bots. But, by and large, AI design is too specific and goal oriented to benefit much from just having good scripting tools.
It is like general procedural generation, it doesn't make content creation easier, just different.
NPCs are mini-puzzles that exist for the player to solve. You create variation by having more puzzles, and sometimes by combining them into more interesting meta-puzzles. There might be some use to AI "learning" but it would be as a tool to solve some of the more tedious aspects of content creation (like animating walk/run cycles) or for testing using bots. But, by and large, AI design is too specific and goal oriented to benefit much from just having good scripting tools.
It is like general procedural generation, it doesn't make content creation easier, just different.
- 2
Daidraco
Golden Baronet of the Realm
- 9,217
- 9,326
Im sure the MMO atmosphere could benefit from some AI. Interactive NPC's in towns and such should have some very clear cut goals for the AI to follow. Always wanting to keep food on the table, regardless if it was eaten or taken by a player - the NPC may go out and barter for some, go out to the orchard for some apples, or if the NPC saw the player take it - whip his ass/get the guard. But to put smart AI on just every single monster NPC out there? No thanks times a million. Little puzzles on how each monster works in order to defeat them is fine. Super evident and fun in Witcher 3 when you had to use basic strategies and such for each type of monster. If Smart AI was to ever be implemented upon an NPC monster, I would sincerely hope that its a rare scenario and not a monster we have to deal with all the time in monotonous situations.
- 1
Ukerric
Bearded Ape
- 7,926
- 9,576
As one dev (whose name escapes me at this time) famously said, you don't need AI. Just script the boss to go for the healer first and ignore everything else would be enough to turn every encounter into a nightmare.
- 2