Next-gen could make it easier for AI programmers to create more complex games
The crowd NPCs in Final Fantasy 15 are dumb. They sit and talk, they share food together, and they find any perch in the environment to rest, but they are brainless. Literally.
They’re controlled by parasites in the environment. These particular brain worms are known as smart objects - objects placed within the levels that act as magnetic congregation points for these drones, drawing them to a table, for example, where different animations play out depending on how many other NPCs join them.
This method of controlling AI has been around since the ‘80s, but newer, more efficient methods will soon be making their way into games.
“The next evolution of AI is going to be crowd AI,” Square Enix lead AI researcher Youichiro Miyake tells me, via translator Ben Judd. “Before, it was smart objects forcing different actions on the characters, where next it’s going to be territories that you’ll designate as AI that will force different actions on the characters.
“In the ‘80s, the classic version would be this cable that would dictate actions on the players or the NPCs, but now it would be an area that would do that. It’s more about trying to be efficient with your processing power.
“The classic style of AI would set up multiple smart objects, a character would go in and it would know all the smart objects - then you could have really detailed, interesting AI, but it comes at a heavy processing cost. With crowd AI, you dictate an area, and then without having to use a lot of processing power, a character goes into that area and has different instructions or dictations that affect them. It’s not about being larger, it’s about being more efficient.”
In Final Fantasy 15, each character had its own AI, then there are smart objects, and finally there’s a meta AI - an all-seeing god that changes the rules depending on the context. One example is that, if you’re fighting an enemy with your companions and you decide to run away, the meta AI will tell your companions to disengage. It breaks the usual rule of combat being a priority to make the game bend to suit the player’s whims. Miyake says future generation platform architects will know that developers are using these techniques, so future hardware could potentially make the process easier.
“But still, as a programmer, you’ll still need to have smart AI programming and there will be certain limitations based on chipset and other factors,” Miyake explains.
“Another piece of the future of AI is obviously going to be machine learning coming in. You’re going to see more and more AI that reacts to a player’s playstyle, to their personality. Are they a good gamer, a bad gamer, do they tend to go into combat, or are they exploratory? We’re going to analyse those patterns and then the character AI will largely have a lot of automation, analysing these patterns and dynamically adjusting to that playstyle.
“However, a lot of the evolution of AI is going to come in the meta AI level, which will be controlled by the programmers who are basically the gods of the game. So them defining the overall rulesets of the game, but within it there’s pocket-sized machine learning-based AI - that’s probably the new thing we’re going to see with AI evolution.”