The general “gold standard” for game AI uses (My background is Unreal so) pathnodes. Little blips that indicate where an NPC should be and that get built into a graph. That graph represents the paths NPCs can take. You add special pathnodes to indicate an interactives and weight the paths to prioritize routes. So imagine if you had the heaviest weight pathnodes on the streets of a town, lower weight nodes for alleyways, and very low weight nodes for a path through a swamp in case of emergency.
Now realize that many textures and assets have those baked in these days. That road texture? That is a path with “road” weight.
What benefit would AI provide for that? None. Because you would just be computing this “live” rather than offline.
As for interactives ? Again, you can bake that into the asset itself. That apple tree has an interaction to check if there is an apple and, if you are hungry, eat it.
Again, why take that online if you don’t have to?
Then you have the idea of NPC schedules. In theory, you can give each NPC a job and they can find food and blah blah blah blah. We already did that. It was called The Elder Scrolls 4: Oblivion. At the best of times, it was indistinguishable from just giving an NPC a schedule corresponding to a work day. The rest of the time, it was complete stupidity as people would walk five blocks for an apple. and they mostly forgot/overwrote their actual jobs (outside of the ones who were hardcoded… which were most of them).
This is one of those “holy grail” ideas that people think of to sell the idea. But you invariably end up with the Star Citizen bar tender where mostly it just breaks and the interactions people expect are the same ones we have been dealing with for decades. Asking Sam Malone for some weed is funny but it isn’t anything you are going to do while playing a game (outside of youtube content). You roll up, buy a drink, ask for some info, and leave.
As for dialogue trees? Need I remind everyone of Darth Vader’s penchant for slurs? Because that is what you get.
Mostly this is like a LOT of applications for AI. People want to replace humans. They end up making something MUCH more fragile than what those humans have been making for decades and then need to allocate resources to update the prompts when a new model drops and so forth.
Versus just continuing to bake pathnodes into a road texture and setting a property on that vacuum cleaner asset and so forth.
I would argue it is more similar to the last few bubbles for this same kind of tech.
A decade or so the big deal was computer vision (think “recognize stuff through the power of AI”). Massive promises were made, a LOT of people specialized in glorified signal processing, and then it mostly faded away when it didn’t live up to the hype.
Except… a significant percentage of the smart home industry is all about paying money to have your video feeds analyzed for people. Same with image hosting services and so forth. Hell, youtube runs on it for DMCA purposes.
And we’re going to see the same thing with “AI” yet again. People will realize chatgpt isn’t going to suddenly manifest itself as Stana Katic and suck you off while telling you you are ten times the man Malcolm Reynolds ever was. But it really does do a lot of “admin work” and is one of the best tools out there for text based pre and post processing of human readable data.
Honestly, NPC behavior would be a really good application for a specifically trained model
It really wouldn’t.
The general “gold standard” for game AI uses (My background is Unreal so) pathnodes. Little blips that indicate where an NPC should be and that get built into a graph. That graph represents the paths NPCs can take. You add special pathnodes to indicate an interactives and weight the paths to prioritize routes. So imagine if you had the heaviest weight pathnodes on the streets of a town, lower weight nodes for alleyways, and very low weight nodes for a path through a swamp in case of emergency.
Now realize that many textures and assets have those baked in these days. That road texture? That is a path with “road” weight.
What benefit would AI provide for that? None. Because you would just be computing this “live” rather than offline.
As for interactives ? Again, you can bake that into the asset itself. That apple tree has an interaction to check if there is an apple and, if you are hungry, eat it.
Again, why take that online if you don’t have to?
Then you have the idea of NPC schedules. In theory, you can give each NPC a job and they can find food and blah blah blah blah. We already did that. It was called The Elder Scrolls 4: Oblivion. At the best of times, it was indistinguishable from just giving an NPC a schedule corresponding to a work day. The rest of the time, it was complete stupidity as people would walk five blocks for an apple. and they mostly forgot/overwrote their actual jobs (outside of the ones who were hardcoded… which were most of them).
This is one of those “holy grail” ideas that people think of to sell the idea. But you invariably end up with the Star Citizen bar tender where mostly it just breaks and the interactions people expect are the same ones we have been dealing with for decades. Asking Sam Malone for some weed is funny but it isn’t anything you are going to do while playing a game (outside of youtube content). You roll up, buy a drink, ask for some info, and leave.
As for dialogue trees? Need I remind everyone of Darth Vader’s penchant for slurs? Because that is what you get.
Mostly this is like a LOT of applications for AI. People want to replace humans. They end up making something MUCH more fragile than what those humans have been making for decades and then need to allocate resources to update the prompts when a new model drops and so forth.
Versus just continuing to bake pathnodes into a road texture and setting a property on that vacuum cleaner asset and so forth.
This is so similar to the dotcom boom where people overestimated the technology and how quickly we would achieve the promised goals.
Half of the shit promised in the 90’s only came about in the early 10’s after literal decades of building the infrastructure.
I don’t think it’ll completely go away but people will eventually figure out it’s no magic bullet and it’ll see significantly less investment.
I would argue it is more similar to the last few bubbles for this same kind of tech.
A decade or so the big deal was computer vision (think “recognize stuff through the power of AI”). Massive promises were made, a LOT of people specialized in glorified signal processing, and then it mostly faded away when it didn’t live up to the hype.
Except… a significant percentage of the smart home industry is all about paying money to have your video feeds analyzed for people. Same with image hosting services and so forth. Hell, youtube runs on it for DMCA purposes.
And we’re going to see the same thing with “AI” yet again. People will realize chatgpt isn’t going to suddenly manifest itself as Stana Katic and suck you off while telling you you are ten times the man Malcolm Reynolds ever was. But it really does do a lot of “admin work” and is one of the best tools out there for text based pre and post processing of human readable data.