When you hear about “AI” in the context of video games, your first instinct is to probably think about how enemies behave. This isn’t exactly wrong, after all that’s how you interact with games that aren’t about playing with other people. AI dictates enemy behavior, and for a long time that dictation has been about linear paths. AI says enemies do X, Y, or Z based on either a predetermined path or reactions to the player. But that’s all about to change, and it isn’t just about how NPCs behave. From AI that is actually learning, to using programming to “remaster” classics, using AI is altering what we know of video games drastically.
Let’s start with the enemy stuff. NPC behavior is evolving beyond predetermined paths, albeit very gradually. But that gradual movement is even happening in AAA spaces, where normally one on the outside might expect more safe allocation of resources. Electronic Arts, for example, has a division called SEED that is devoted entirely to technological innovation. SEED has been working with development studio DICE to teach an AI program how to play Battlefield 1, the company’s World War 1 shooter from a couple years ago.
This AI uses a combination of learning techniques to develop its “skills.” One is based on watching players play, getting a taste of how real people play Battlefield 1 and attempting to mimic that, then using that mimicry as a base. That isn’t good enough by itself, however, so SEED is also using a reward-style learning technique. This programming rewards the AI for taking various actions, which helps it learn to avoid things like running into walls, landing shots, avoiding damage, etc. This reinforcement is meant to supplement what the AI learns from players, to help it get even better at playing Battlefield 1 without being told exactly what to do.
Another big way to use AI with games that is emerging recently is a “remastering” technique. Using machine learning, neural network-based AI, homebrew developers have taken the all-time classic JRPG Final Fantasy VII and “remastered” its aging visuals. This is particularly used for the pre-rendered background art, which while timeless in design is not so timeless in resolution. This technique has allowed PC players to download a mod that boosts the resolution of the pre-rendered art without having access to the original assets.
But that’s not all. Similar techniques are even being applied to old Full Motion Video. Remember the original Resident Evil? Yeah, we’re talking about the 1996 PlayStation game, not the fancy-looking GameCube remake. That old introduction video is legendary, both for its cheesy tone as well as its striking (for the time) violence. Well, now you can hop on over to YouTube and watch that video, which has long been a fuzzy mess, running at 4K resolution and 60 frames per second. It isn’t perfect, but it’s the most amazing a mid-90s PlayStation FMV will ever look. And it’s all due to AI machine learning.
The applications for this technology have so much potential. How many classic games have gone untouched for years, due to dead publishers, a lack of mass market interest, or lost source code? Dozens, if not hundreds, is the answer. If pre-rendered backgrounds can be remastered, along with FMV footage, there are so many games that could benefit from that, especially those classic RPG titles.
At the same time, just imagine what games will be like when AI opponents are no longer predictable. Difficulty options could be completely dynamic, no longer based on increasing numbers to make enemies more bullet-spongey. On the other hand, you could also have AI companions that aren’t the bane of every player’s existence. The future is exciting for video games, old and new.