When we think about the future of MMOs, we usually focus on the backstory and gameplay -- or more recently, how single-shard technology might change the nature of the genre. Less discussed are the innovations in artificial intelligence which will utterly alter the game worlds in which we’ll soon play.
We’ll see some of the first glimpses of this shift with Seed. While emphasizing that the player community and its diverse groups will shape the game more, Klang CCO Ívar Emilsson hints at some of the ways that AI will influence Seed.
“[W]e‘re creating a virtual world where players should recognize their characters as intelligent, emotional beings,” as Emilsson put it to me. “The simulation is continuous, and even when the player is offline their characters should go about their day-to-day lives, they should respond to threats, make friends, be human. We aren’t trying to figure out the best solution to a problem (the goal of machine learning), but rather simulating human behavior. The choices our characters take don’t need to be perfect, in fact, when they’re too perfect they feel unnatural and robotic.”
However, he added, Klang is using machine learning to test gameplay: “By emulating players (rather than characters) we can see how they interact with the simulation and try to ‘win’, allowing us to catch issues and make more informed decisions.” Furthermore, advisor Adam Prospero states, “At a certain point you can only pile so many polys into the screen. As Minecraft proved, consumers care less about how high-res the world looks versus how believably real it feels. And now we finally have the tech to truly kick believability up a notch.”
NPCs Which Learn by Observing Players
Now imagine NPCs imbued with the latest artificial intelligence technology, able to “learn” by recording a given player behavior, then repeating and perfecting it through trial and error. In an MMO context, imagine an NPC that can learn by observing (and repeating) the behavior of player avatars.
We are already seeing a glimmer of this technology from Deep Motion, a Silicon Valley-based company creating AI for game environments. They recently unveiled a demo in which an NPC agent “learns” to dribble a basketball through many cycles of trial and error and acquired locomotion patterns.
"Our methods are dynamic, we use linear methods for simple movements and deep reinforcement learning for complex motion skills,” Deep Motion's Libin Liu told me earlier this year.
The process of becoming a proficient dribbler, he added, “varies from an hour to several hours on a mid-tier server”.
Consider how much faster an NPC could learn complex physical activity like this, if it imitated the actions of MMO players around them. Record how they explore a town, for example, then imitate that pattern as an NPC. Record and imitate enough player behavior, and NPCs would no longer interact with the world on a set, predictable path. On the surface, NPCs like these would be difficult to distinguish from actual players.
Sentient AI Emerging from Within MMOs
We’ve talked about ways in which AI will enhance MMOs, but for many technologists, the real goal is exactly the other way around: To enhance AIs with MMOs. This already makes practical sense in many cases. For instance, if you want to QA the artificial intelligence program of a self-driving car, wouldn’t you rather test it out by having it drive a virtual car around in GTA Online -- as opposed to driving an actual car on actual streets full of actual pedestrians?
Some futurists go even farther than that, suggesting that we can use virtual worlds to create an artificial intelligence that’s complex enough to become “conscious”.
The exact nature of human consciousness is still not exactly known, but many experts have moved away from the traditional “the brain is consciousness” perspective.
“You are not your brain”, as acclaimed philosopher of the mind Alva Noe puts it, arguing that consciousness is what emerges when we interact with the world around us through our entire bodies.
Approached this way, it’s easier to see how sentient AI might emerge from within MMOs. Create a simple AI agent with a rudimentary flight or fight response system, an ability to record and pass down its “memory” to successive copies, and release it into the virtual world. As it interacts with the environment -- MOBs, NPCs, players -- it responds to these encounters, and survives, or dies. Through trial and error, its offspring begin to “learn” the best way to survive. After thousands of generations, and millions of interactions, its knowledge of the world becomes so complex, it effectively becomes aware.
“It is one thing to defend the position that we will not be able to figure out how to breathe consciousness into modern AIs,” as Second Life creator Philip Rosedale put it recently. “But [it] is bolder to argue against the likelihood that such things will evolve on their own inside digital worlds simulated by increasing powerful computers.”
After spending so much time simulating whole alternate existences, with their own ecosystems of artificial flora and fauna, it seems inevitable that artificial consciousness will soon follow.
"It'll be possible for constructs that we build in a simulated space to actually think,” as Rosedale put it to me in 2008. “It's only a decade away." So maybe it’s past time to ponder when AI will become self-aware -- and wonder if some of them already are. - Wagner James Au All opinions are the writer’s own.