Whenever people talk about "good AI", they invariably come back to the same three examples

- I'm talking about the replica soldiers in the original FEAR, the various aliens in the

Halo series, and the marines in the first Half Life game.

And, yes, these games do have some really clever behaviours, and Halo 2 and FEAR actually

pioneered some new AI technologies that are still being used in games today.

But one other thing these games have in common is that they all feature particularly aggressive

enemies, who actively try to hunt down and kill the player, and also have a lot of health

points so they can stick around for longer.

And this is something that makes these enemies feel a lot smarter.

But don't take my word for it - this is something that Bungie knew, all the way back in the

first Halo game. During development, it set up a playtest with two versions of the game,

with the exact same enemy AI on both - but on one machine, the aliens didn't do much

damage and died quickly, and on the other they did more damage and had more hit points.

The number of players who thought the enemies were "very intelligent" jumped from 8 percent

to 43 percent, when playing against more resilient enemies.

But, the thing is, aggressive enemies just don't work for every type of game. This is

something that id software found out during the development of Doom 2016.

It started with enemies that would chase you down, but this made players act defensively

- "We’d end up in these situations where you would instantly start backing up," said

director Marty Stratton. Instead, the studio told the enemies to hold their ground a bit

more, and let the player be the one who's pushing forward.

So, just like everything else in game design, AI must fit the game's intended experience.

Which means while aggressive AI fits the Xenomorph in Alien: Isolation - it would be wildly out

of place in Batman: Arkham Asylum. No one gets the drop on Batman, after all.

And this all means there must be more to "good AI" than just "enemies who can kill the player".

And that's exactly what I'm going to talk about in this video - with everything from

good, general practices - to examples of truly ambitious and ingenious AI.

Good AI lets the player cheat. Just, not in ways that the player will actually notice

- like how you can put a bucket on a shopkeeper's head in Skyrim and rob them blind.

So, hopefully in more subtle ways, like how in the Uncharted games, when you pop out of cover,

enemies will start with a zero percent chance to hit Drake - giving you a chance to take

a few shots. In the Far Cry games, only a few enemies will be allowed to shoot at you

at once to improve your odds of winning the fight. And in the Arkham games, enemies are

told not to turn around during the predator sections so Batman can sneak up behind his prey.

These are things that the player should never really notice - but you would definitely feel

their absence if they weren't there. It's all about making the game feel more fair,

even if most games are actually biased wildly in the player's favour.

Good AI tells you what it's thinking. This is most often done through short vocal clips

- known as barks - where patrolling guards say things like... "Sounds like someone's over there..."

and... "Must be nothing".

But this can also be expressed through animation and body language, or with more gamey-elements,

like vision cones, light and noise sensors, and those ghostly images that show where the

enemy last saw you.

I will talk about how much of this stuff you want to surface to the player, and how much

you want to keep a bit fuzzy, when I talk about stealth games in a future video.

A different way to achieve this is to give AI characters distinct personalities, like the

coloured ghosts in Pac-Man, or the different leaders in Civilization's single-player modes,

who all have their own unique quirks.

Developers have actually found that this stuff makes AI characters seem smarter to players.

Because, if an AI has complex decision making and perception skills, like being able to

notice that a door along their patrol path has been opened, the player may never know

that the guard is capable of such a thought if they don't open their mouth and say...

"Did I leave that door open?"

But this is also critical feedback that the player can use to understand what the AI is

doing, or about to do - and can plan accordingly. Which brings me onto this next one...

Good AI is predictable. Which, sounds odd but hear me out.

In 2004, Halo tech lead Chris Butcher said, "The goal is not to create something that

is unpredictable. What you want is an artificial intelligence that is consistent so that the

player can do things and expect the AI will react in a certain way".

This allows you to play a game with intentionality - which, Far Cry 2 designer Clint Hocking

defined as "the ability for the player to devise his own meaningful goals through his

understanding of the game dynamics".

To break that down a bit - when you play a game you start to build up an understanding

of how different things work - like, you shoot a red barrel. It explodes. And now, going

forward, you know that every time you see a red barrel you can shoot it to create an

explosion - and can use this to your advantage.

But this applies to AI behaviour, too. If guards always return dropped guns to crates,

and turning off a generator will always make an enemy come check it out, you can then use

this information to create plans, diversions, and traps.

Without predictable behaviour, the player can't create satisfying plans. Halo man Butcher

gave the example of sneaking up behind a Grunt, and the Grunt running away - "it would be

bad if they only ran away half of the time, because then the player can build a plan that

will only work half of the time".

Instead, Bungie went for predictable actions but unpredictable consequences. "The Grunt

will always run away," said Butcher, "but you don't necessarily know where he'll run

away to".

So predicability certainly doesn't mean easy. Take a game like Spelunky where every enemy

acts in an almost completely scripted fashion - which would make them effortless to avoid

or kill, until the enemies start appearing in big groups, begin to navigate different

environments, and start interacting with other characters. Now it's Spelunky.

Good AI can interact with the game's systems.

This is like how an enemy in Breath of the Wild doesn't just walk up to Link and start

wailing on him - but will run off and pick up dropped weapons, set their wooden clubs

alight, kick away bombs, and even throw their fellow monsters at you.

Oooh, that's gotta hurt!

Again, this has the added benefit of making an AI character seem smart. An enemy in Bioshock

who runs to a health dispenser midway through a fight looks like he is aware of his surroundings,

is interested in self preservation, and feels like he has similar abilities to the player.

But it also means you can screw them over by putting a trap on the health dispenser.

Hehehe!

By exposing an AI to the game's systems, we can provide loads of interesting ways to deal

with foes in a more roundabout fashion, like making an enemy fight for you in Prey, or

tricking an enemy into attacking a Cucco, so this happens...

By the way, I propose we call this "chicken-boning an enemy".

Good AI reacts to the player. This can be as simple as guards becoming more frightened

as you take out their buddies in the Batman games, or something as complex as Shadow of Mordor.

In that game, special Orc captains are randomly generated with names, abilities, and relationships

- and will then remember their interactions with the hero. If you run away from a battle,

for example, the Orc might reference this the next time they see you.

ZUMUG: Hey! Not letting you run this time! I's gonna finish it!

This is a great way to create memorable, and very personal stories for the player.

ORTHOG: Your death will bring me even more glory!

Tracking the player can also be used to adapt the way the AI works. This doesn't need to

be as clever as the Shadow Fighters in the new Killer Instinct, or the Drivatar system

in the Forza games, where Microsoft tracks the way you play and can create AI doppelgängers

to race or fight for you.

Really, it can be as simple as something like in Metal Gear Solid V, where enemies track

things like how often you perform headshots, take out bases without being seen, or infiltrate

during the dark - and then change to different behaviours like wearing helmets, laying traps,

or using night-vision goggles.

These are all things that the AI has been told how to do - it's not actually "learning"

- but it just won't do them until the player hits a certain threshold. And the intended

effect is stopping the player from using the same boring strategy for every base in Afghanistan.

A similar system is used in Alien Isolation where the Alien unlocks new abilities as the

game goes on, to make it look like the Xenomorph is learning from the player, and to keep the

game interesting as the hours tick by.

Adapting the AI to the player is also used to build a good mood, or drive the game's

pacing. The most famous implementation of this is certainly the AI director in Left 4

Dead. This clever system tracks the wellbeing of each player, based on their health and

run-ins with special infected, and if the team is cruising along, the intensity of the

zombie horde is increased - before the AI director eases off to give the team a chance to relax.

But this tech isn't as new as you might think and something similar was actually used in

Pac-Man. Designer Toru Iwatani said "I felt it would be too stressful for Pac Man to be

continually surrounded and hunted down. So I created the monsters' invasions to come

in waves".

So, in the game, ghosts swap between chasing the player and just wandering off into the four

corners of the maze.

Good AI has its own goals. Beyond, "kill the player", I mean. This game is Rainworld, by the way,

where other animals in the game are hunting for food and will end up in territorial scraps

with rivals. Sometimes, it's best to just let them get on with it.

That's also how STALKER: Shadow of Chernobyl works. Or is supposed to work. Bandits make

plans, and then wander about the wasteland alone or in groups to enact those plans. Meaning

you could come across a raging battle that only occurred because two rival factions happened

to accidentally run into one another.

The AI is a bit buggy though, and you will need to install mods if you want the AI to

be doing stuff outside of your immediate location. But STALKER's A-Life system sure was ambitious

and we certainly need more games like it.

Or maybe more games like Waking Mars - a captivating sci-fi indie game about increasing biodiversity

in the belly of the red planet. When you figure out how different plants and animals react

to one another, you can forge self-sufficient ecosystems where AI critters can live and

breed - even while you're exploring a completely different part of Mars.

Finally, good AI isn't just about enemies, and we need better friendlies. Because even in games with

great combat encounters, the good buys can be as dumb as dirt.

SCIENTIST: Hello!

Now, some developers just cheat, and make their AI companions invincible - like Elizabeth

in Bioshock Infinite who can't get hurt during combat. Probably a wise decision, given the

tumultuous history of escort missions in games.

But there are other uses for friendly characters than just defenceless girls who follow the

male hero around.

In The Last Guardian, you work with a giant beast called Trico who can take care of enemies.

But Trico is nervous around stained glass windows that you can smash. This means the

player and the AI must work together - something which can get a bit frustrating when Trico

has been specifically told to ignore the player's instructions about half the time, in an attempt

to make it seem more like an animal.

In Event[0], developer Ocelot Society was inspired by internet chatbots

to make a game where you can talk with an artificial intelligence, and work with it

to solve puzzles on a derelict space ship. I've got much more on that game in an earlier video.

And, then, in Final Fantasy XV, one of your road trip buddies, Prompto, is capable of

snapping Instagram shots during your adventure. At specific triggers, or just when Prompto

feels like it, he'll automatically take a picture, and later present you with an album

full of snaps to sift through.

GLADIOLUS: Damn, this is a really good shot.

PROMPTO: I made sure I got the Regalia's good side!

This has basically no gameplay value, but is a

lovely way of preserving your personal experience with the game, and adds tremendous amounts

to Prompto's character.

Now, none of this is to say that we don't want, simply, "better AI". Stupid enemy decisions

can pull us straight out of an experience, and players derive little satisfaction from

beating an obviously unintelligent opponent.

Okay, maybe it is quite satisfying.

But developers should always be trying to make further advancements in AI tech, and create

more nuanced behaviours for enemies.

But it's important to remember that AI isn't just a technical problem - it's also a design

problem, and every game should approach the subject in a slightly different way.

Sure, we do need more games that are about fighting a tactical squad of aggressive enemies - because

something's clearly not quite right when most modern shooters are still lagging behind a campy 13 year old

game - but it's important not to lose sight of the real goal.

In Uncharted 4, Naughty Dog experimented with complex AI behaviours, before settling on

enemies who were, according to designer Matthew Gallant, "spread out in a layout, looking

human and smart, and moving in ways that are mildly predictable so the player has some