The quality of gaming AI and bots

Machine or man?

The gaming culture is one of ebb and flow, fads and trends, habits that come and go. Sometimes that is for the best, and sometimes it's for the worse, but either way, it makes gaming and game design a world in constant flux. To pull out that Bennett Foddy quote again, "It's like building on drying concrete." We all have those eras of gaming that we wish we could return to, times that we can point to and go "There! They had the right idea with that one." But time moves ever onward. That cannot prevent some wistful folk, like myself, for sparing a nostalgic thought about what was and what might be had certain trends played out differently, with that in mind, let's talk about AI.

No, I'm not talking about the traditionally accepted definition of AI (Which can be more accurately defined as 'super-intelligent AI') but rather the collection of algorithms and processes that make up the mind of a computer; it's 'Artifical Intelligence'. In gaming, we commonly use the term 'AI' to refer to the handling of bots and NPC's by the software, it's a catch-all term that encompasses their behaviour, reaction and believability. A game that would considered having 'good' AI, would be one wherein the NPC's make appropriate use of their tools, navigate their environment succinctly and pose an actual threat to the humans; whereas a 'bad' AI would be the type you see running into walls and standing around waiting to be shot.

In the early days of gaming, AI wasn't too much of concern for programmers as their games were a lot more simplistic in scope. Enemies didn't really need to be programmed with a wide range of possible actions and route planning algorithms, they just had to operate a simple patrol task with the player's one job to be to avoid them. It was in this vein that famous video game bosses such as Super Mario Bros' Bowser, resorted to little more than jumping up and down and shooting fireballs every now and then. The only real challenge on the player's part is jumping over the Koopa king and hitting the axe-switch to plunge him into lava. Difficulty ramped up as patterns became more unpredictable and/or erratic, which is why many a player still has nightmares about the Hammer Bros from Super Mario Bros 3 and the Gorgon heads from Castlevania.

Games gradually evolved throughout the years, however, and so too did people's perception about what made good enemies in video games. In my opinion, the real watershed moment was when 3D world's became a thing with the advent of the Nintendo 64. Suddenly, AI would need to navigate a whole 3D environment and it became difficult for Developers to get away with simple patterns for the enemy AI. Now they had to code in path-finding and write in extra rules to determine line-of-sight and determine when to use certain abilities. The old guard method of planning would be to have enemies attack the moment they rendered on the screen or whenever the player got too close, now games consoles had become so powerful that this was unfeasible, enemies could be rendered from far away and players could navigate in 3 dimensions, requiring the system to evolve.

This really started to take route in the early 2000's when Developers began to expand the sorts of games that they could make. On of the biggest games of the time that boasted about it's AI's capabilities would have to be, possibly the first game I ever played, Metal Gear Solid. That was a game which ushered in a whole new genre of play, stealth, and with it a whole new set of requirements when it came to coding enemy AI. Patrolling guards had to follow their routes, sure, but they had to be able to react to their situations in a way that felt dynamic and realistic. Should they become alert, they needed to comb the area in search; if someone held them up with a gun, they needed to freeze in fear of their life. This revolutionized the way that people viewed AI and laid the ground works for where it would evolve next.

From this point onwards it became something of a point of pride for developers to boast about the cool new AI that their games had to offer and boast about how clever it was. Battlefield 1942, for example, had one of it's key selling points rest on the strength of it's bots and their ability to mimic real life opponents. (Isn't that weird? A purely online game that teases the offline components.) This trend caught on too, with future online games like TimeSplitters putting considerable effort into ensuring that their offline play was just as exciting as their online play. During this time it was actually feasible for an offline gamer, like I once was, to buy the newest multiplayer centric game under the knowledge that I wouldn't be left out.

One might have thought that this influx of innovation would be never-ending considering the huge jump forwards in software tech in the years since, however that has not been the case. It seems as standard AI procedures (AI good enough to hold their own against a human) became less of a novelty and more of the norm, there grew less of an incentive to strive for improvement in this general area. Games stopped boasting about how smart their AI was and some multiplayer titles started forgetting about AI Bots altogether. (COD has never had AI bots in their multiplayer as far as I know.) I guess that creating the perfect online opponent was too close to literally cloning gamer brain patterns for Devs to continue down that road. (Although, some of the best advancements in the development of general AI have been made in Video game settings. Maybe these game companies are selling themselves short.)

In the modern age, the only time you'll hear a big fuss made about the quality of AI is when something truly spectacular has been achieved. Who remembers the reveal gameplay demo for 'The Last of Us' when we saw Ellie dynamically react to a situation when the player was in trouble? It was an incredibly impressive showcase and one that should have, in a perfect world, sparked interest in bot development for the future. But it didn't. The same was true for the impressive AI systems behind the Xenomorph from 'Alien: Isolation'. With a reputation for being the 'perfect organism', Creative Assembly knew that they had to do something more imaginative with their Alien beyond giving it a patrol schedule, and so they designed two AI 'storytellers' to manage it's behaviours. One storyteller would give the Alien's AI clues as to where the player was, simulating the 'it's always nearby' paranoia from horror movies, whilst the other would send false clues to distract the alien, ensuring it wasn't always on the player and making it's movements difficult to predict. Despite the creation of this ingenious system, 'Alien: Isolation' was not the spark to revive the AI trend.

So is the concept of great AI complexity dead in the world of gaming? Not quite. Some games have started to look into bringing bots back into multiplayer games, like Battlefront 2, and advanced AI scripting is slowly becoming more of a talking point thanks to pioneers like 'The Last of Us part II'. But perhaps what we really need is a huge leap forward in the technology to really fan the flames of creativity once again in the minds of creators and push the boundaries of what can be possible. I've seen AI demos in simulated environments that go so far as to start simulating the action/reaction motion of human emotions, effectively creating artificial wants and needs; the least we can do in gaming is create an AI that chooses to take cover once and a while.