Why F.E.A.R.’s AI is still the best in first-person shooters

The shadows on the wall tell me they’re coming. Two of them, both with assault rifles swinging idly at their hips. If I’m quick enough, I’m sure I can take them both out in one go. I peek out of cover as they round the corner, and let my stake gun sing, pinning the first enemy to the wall with 10mm steel projectiles. But at the sound of gunfire the other one legs it back the way he came, hunkers down in cover, and yells for reinforcements down his radio.

This five-second episode tells you a lot about the attention to detail in F.E.A.R., a 12-year-old game with AI that puts many modern-day shooters to shame. Its army of clone soldiers feel smarter than any enemy I’ve faced in an FPS since, and remain razor-sharp to this day.

The clones will shout “covering fire” as they spray bullets at your position at the same time as their squad mates move to safer spots. They’ll hide in dark corners to try and get the drop on you, and fall back to more solid cover if you’re getting the better of them. In the clip below, the enemy on the left peeks out of cover (at the start of the gif) to get a look at my location. In most other games, he’d poke his head back out at the same spot, so I’m ready and waiting to splatter his brains on the wall. But instead, unseen to me, he crouch walks behind cover and emerges somewhere else, making me re-adjust my aim.

Grenades are a big part of the AI’s arsenal, thrown both to cause damage and to flush you out of cover. In the clip below, I’ve taken refuge behind a pillar, but an incoming grenade from the right (the enemies actually shouted “flush him out” before they threw it), forces me to the left and into their line of fire.

This is all smart stuff, but it’s the kind of thing most FPSs should be able to get right. However, if there’s one thing that sets FEAR’s AI apart from the crowd, it’s movement. Enemies are constantly shifting from cover to cover, retreating around corners, changing spots to get a new angle on you. They’re looping back on themselves and diving through windows, crashing through doors to open up new parts of the battlefield, and they’re never in one spot long enough for you to get comfortable. It means that crouching behind cover can be disorientating, because when you lean out again the enemies won’t be in the place they were when you were last firing.

And this movement isn’t just random, either. Enemies are aggressive, and will actively hunt you down in packs. More than once I’ve died because an enemy simply ran over to my cover position and shot me mid-reload, before I even knew they were there.

What’s most remarkable is the simplicity of the system behind the AI, as revealed by Jeff Orkin, senior engineer at Monolith productions, way back in a 2006 GDC presentation. The team had just one AI programmer, and so designed a ‘planning’-based system that lets the AI soldiers think for themselves.

Instead of telling an AI exactly how to behave in every situation, F.E.A.R gives the soldiers a set of goals and a set of possible actions, and then lets them figure it out for themselves. Essentially, the AI reacts dynamically to what’s going on in the environment – for example, if they’re in danger they will look to retreat, but only if they can identify a safe path to follow. If not, they will hunker down, and perhaps blind fire from cover to try and slow your progress.

Where it gets really impressive is where AI soldiers work together. They can form impromptu squads based on proximity to each other, and then act as a unit to kill you, each aware of the other’s actions. In practice, this means that while they normally start together in any gunfight – coming through the same door, for example – they quickly fan out, and before you know it you’re twisting your head this way and that, falling back to try and keep them all in your field of vision. It almost feels like they’re being controlled by single unseen player on high whose one objective is to see you dead.

None of it is dictated. There’s no direct command to flank, for example – but complex behaviours emerge from the soldiers analysing and reacting to their environments, which are built to showcase the AI at its best.

In the clip below I’m in a strong position, firing my shotgun down a narrow corridor, with all my enemies in front of me. At least I think they are, until one enemy flanks into a small room on my right, chipping away my health and forcing me to retreat to where I can engage him one on one.

The cherry on top is communication. As the soldiers fight they provide a narrative for the chaos in the form of orders and updates barked down their radios. “He’s flanking us”, one will call out, as his squad reacts to my movement. “I’ve got nowhere to go!” another will call, aware that he’s trapped but unable to identify another safe spot.

As Orkin explained, “Having A.I. speak to each other allows us to cue the player in to the fact that the coordination is intentional. Of course the reality is that it’s all smoke and mirrors, and really all decisions about what to say are made after the fact, once the squad behaviour has decided what the AI are going to do.” Smoke and mirrors it may be, but it really makes you feel like you’re facing an intelligent force.

In his presentation, he observed that the AI reminded players of the first Half Life game, which came out seven years earlier. “It seems that we haven’t made much progress in seven years. There has to be more we can do with game AI.”

Unfortunately, what F.E.A.R demonstrates is how little AI has progressed since 2006. There’s no one aspect to it that’s unique – no one individual behaviour I can pick out and say that I haven’t seen in other games since. But it nails the simple things, and brings them together into something that’s greater than the sum of its parts. 12 years on, it’s still the pinnacle of shooter AI.


  1. Snowskeeper says:

    Pretty much everything this article makes me feel can be summed up with “Aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaah.”

    • jonfitt says:

      Spoilers for an 11 year old game. But I remember one early pant-shitting moment when you climb a ladder and just as you reach the top the baddie peek-a-boos over the top, your character recoils, and when you get back up it’s gone.

      • Daemoroth says:

        I will never, EVER, forget that scene! I know I still had a wired mouse at the time since it took my coffee with it.

        • jonfitt says:

          Haha. Oh dear. I had kept myself free of spoilers, so I didn’t know it was going to be that kind of game! I had just seen screenshots of soldiers leaping through windows.

          I felt like they didn’t overuse the jump scares which stopped them from becoming boring (although I would have jumped at every single one, I would have then said “oh yes very clever”).

      • CAMN says:

        Oh god… seeing Alma up there was made me let go of my mouse and say more than a couple bad words.

      • Legion1183 says:

        There was that ladder scene and then the other one that freaked me the hell out when you’re climbing down a ladder and as your character turns around the guy (I forget his name) is slowly walking towards you and then disintegrates as he passes through you.

        • Elisianthus says:

          The greatest part is it’s the exact same ladder for both of them. You Panic, drop away from spoopy ghost girl and turn right into current main bad guy three feet from your face.

          • Cataleast says:

            Pretty much emptied both my shotgun and bladder when Fettel came at me after getting off the ladder. That and a headcrab leaping at me in a dark ventilation shaft will remain the two most bowel-evacuating moments in gaming history for me.

          • jonfitt says:

            Oh man! Headcrabs in the dark ventilation shafts! Forever after I would pull out the crowbar and prepare to flail at the darkness like trying to swat an invisible fly.

      • Shaileen says:

        Yeah, this game was so awesome with it’s scares. I remember one moment when you were walking down some corridor and a body was thrown through the glass on the left side. I literally almost fell out of my seat and emptied the entire mag into the ceiling. I think that was the strongest physical reaction I have ever had to anything games or movies. Never again have I experienced that fear until Hannibal (the series with Mads Mikkelsen) came around.

  2. MooseMuffin says:

    My rose-colored memories assure me there has never been smarter fps AI than Half-Life’s commandos and female assassins.

    • Heliocentric says:

      A wild Laser Tripmine appears.
      Laser Tripmine uses Thoughtful Placement.
      Thoughtful Placement was every effective.
      Female Assassin has fainted.

    • Plosive Scotsman says:

      do yourself a favour: keep those tinted shades on and don’t replay it

    • JackMultiple says:

      Slightly off-topic, but… isn’t F.E.A.R one of the first games that had several humorous references to the movie “Office Space”? In particular, I believe if you snoop around in one of the upper floors of an office building, you’ll find “TPS Reports” lying about, and Milton’s red stapler on one of the desks. Always loved those refs to one of my fave movies at the time.

      This may have been the level where the “invisible ninjas” started falling from the ceiling. I hated that nasty trick.

  3. ikazrima says:

    Without the slow-mo I’m always dying in matter of seconds. T.T

    • Cryio says:

      The game is so much more visceral when not played with slow motion.

      • haradaya says:

        Yes. I find it’s better to use slow-mo sparingly. And screw only making precise shots, let the bullets fly and kick up dust everywhere.

  4. Ejia says:

    Everyone’s been gushing about this so much maybe I will go back and play it, even though I am a giant wuss and hate the scary bits.

    • FurryLippedSquid says:

      I found that if you play it Hell-for-leather it becomes a lot less scary. Particularly if you have the scope to shout loudly without annoying anyone at home.

  5. Darth Gangrel says:

    As a matter of higly interesting fact, I just finished F.E.A.R.’s campaign for the first time yesterday. Gonna play the expansions next and then we’ll see about the less admired sequels.

    Even though it’s a game from 2005, I was amazed by how much the AI moved around and talked to each other, even though I realized that was also an easy way to communicate to the player what was going on. If I lit up a dark area and the enemies saw it, they would yell “Flashlight!”. I also saw them dive over low walls to get to cover.

    • Heliocentric says:

      I pushed though the expansions and they are a little tortured, Fear 2 has some genuinely brilliant moments, but madly, playing as creepy ghostman Paxton Fettle in Fear 3 was my serious highlight.

    • MattM says:

      The first time I played through Extraction Point, I didn’t really like it. There isn’t much new gameplay and the plot feels like slog to get to a conclusion you can pretty much predict 10 minutes in.
      The second time around it click with me and I loved it. The characters are highly trained spec ops but they are completely out of their element and are scrambling to even come up with a plan. They are do something because they can’t do nothing, but the sense of inevitable failure really keeps the emotional level high.
      I really wish FEAR 2 had continued the story of Extraction Point instead of ignoring it.

      • scatterbrainless says:

        Yeah, despite FEAR’s Ringu creepy little girl, the right movie analogue is actually Aliens for the most part.

    • bill says:

      Weird. I just finished FEAR’s campaign for the first time a few days ago.

      I can’t face the expansion though as the game as a whole is incredibly dragged out and the environment is so repetitive and dull.

      I agree with the article on the combat though.

    • dskzero says:

      Both expansions are pretty cool in my book, but you might end up burned out in the end, since it’s more of the same (the second one does amp up the horror, but that’s it).

      As for the sequels… well…

    • Cryio says:

      The first expansion, Extraction Point, is IMO even scarier and has better gameplay/setpieces than the core game. Perseus Mandate on the other hand is kind of a drag and not nearly as exciting.

    • Cryio says:

      FEAR is slightly scary, Extraction Point is the scariest in the franchise, Perseus is meh. FEAR 2 isn’t scary at all and FEAR 3 is anything but horror.

      F2 too me is like the most consolized FPS of all time. While F1 feels like a true PC FPS, F2 is just standard, generic fare console shooter. In controls, feeling and general visual design. The story DLC is very meh as well.

      F3 is … another game entirely with the name FEAR in the title.

    • haldolium says:

      None of the two expansions for F.E.A.R. are from Monolith.

      Both Extraction Point as well as Perseus Mandate were made by TimeGate studios and lack very very notable everything that made F.E.A.R. great (or let everyone overlook those dull levels :)

  6. sgt. grumbles says:

    Great article. As an old person, I wish gameplay, including AI, would advance as quickly as graphical bling. New dazzles are lovely, but if it feels like I’m just playing yet another generic shooter, I grow bored.

    We need further advances in AI, destructible environments, and so on.

    • Snowskeeper says:

      The frustrating thing about this is that it seems like AI isn’t just stagnating; it seems like people are actively ignoring what other games are doing, out of fear that they’ll be accused of just copying them. For some reason, people seem to believe that taking something that another game did well makes you an unoriginal hack.

      • Heliocentric says:

        Every lead developer (and Japanese Devs are particularly guilty here) I hear say “I haven’t played game X” (where game X is the market leader for excellence) like it’s a freaking badge of honour, I think “Why the hell not?” why are you trying to see How Company of Heroes does RTS barks, why don’t you want to see how SWAT 4 handles non lethal takedowns in an FPS why isn’t your 3rd person climbing as good as Grow Home?

        • jonfitt says:

          Can you imagine a scientist saying “Oh I deliberately haven’t read the paper by X. So I’m working from a starting point that has been completely surpassed by the work of X and iterating on an already outmoded model.” !!!

        • scatterbrainless says:

          It’s this whole creativity myth, like the myth of the genius in the 19th century, that real creation just magically comes out of your being, completely independent of context. Kind of like how IP polices ideas, as though they were the property of a person, instead of acknowledging that ideas grow in relation to each other. It’s like screaming “I don’t need to stand on the shoulders of giants, LOOK HOW TALL I AM, you callin’ me short?”

          • Michael Fogg says:

            I agree in general, but intelectual property laws are quite specific in that they concern expressions of specific ideas and not ideas themselves. You can use another’s idea all you want without a license, as long you turn it into your own expression.

          • keefybabe says:

            It’s true. It’s also true that once you stop consuming the media you’re creating, you kind of start to suck at it. I’ve seen it with music, games, what have you. It’s nearly impossible to truly grok something unless you’re knee deep in it. Otherwise you just get bogged down in systems and rules.

          • jonfitt says:

            That’s why I’m really excited to see some nu-games-journalists making a career shift into making games. For example, our Alec, our Jim, and Tom Francis (not of this parish).

            These are people who started writing really good analysis about games above and beyond 8/10 graphics, 6/10 sound. I think with their early careers spent in games journalism, they will be able to stay plugged in for a lot longer than people who go straight into a games company.

            I think it’s one of the promising trends of recent years (that hasn’t been written about afaik probably because of the inside nature of the situation).

            ( I recommend the Crate and Crowbar as a weekly podcast that goes deep into games, and Tom is one of the a regular hosts. )

          • stringerdell says:

            The new Zelda is a great example of how getting out of this way of thinking can lead to amazing result

          • Snowskeeper says:

            Weapon durability, though.

        • stuaxo says:

          Making games is just really intensive and long hours. Somebody that likes writing code for graphics or AI might not even be that into playing games, or if they were into games – working on it all day, can take the joy out of playing them.

          So yep, all sorts of reasons gamedevs might not play that many games.

          • jonfitt says:

            I agree it’s hard work, and not everyone in a team needs to do it. But those in control positions *need* to be aware of prior art and trends. You can’t just rely on (historically) publishers to say “people like team shooters, make a team shooter”. You need to be plugged in in a way that only a technical person or artist can be to say “I saw this Siggraph talk, and I think I know what’s coming”.

      • Czrly says:

        Why would you waste money on AI developers (exceedingly expensive resources) when your “whack-a-mole” AI is already passable for a cover-shooter targeting the lowest-common-denominator and you know you can basically print money by churning out the same AAA shite, year after year?

        I wonder if this is a console thing. I’m pretty certain that the cover-shooter concept is due to consoles (there’s honestly no way that cover-based game-play is superior when one has a mouse for aiming and a crouch key. Add in a go-prone key and the competition is even more laughable.) and I suspect the fact that we see so much of this “whack-a-mole” AI stuff is because it’s easy to get headshots with a controller if the pawns pop up in the same place all the time.

        • Snowskeeper says:

          It shouldn’t be significantly more expensive to, as the author noted, develop a series of organic (sorry) responses to simple situations, as opposed to a series of responses to complex situations.

      • Baines says:

        Some time after Half Life and FEAR, there seemed to be a spreading push back against AI development by developers.

        You started seeing FPS developers saying that AI development wasn’t worth the time or effort. “Cheating” was indistinguishable for the majority of players, required less effort, and gave greater control to developers.

        You also started to see a growing fear and/or excuse that intelligent AI would become unbeatable, killing any fun. This was a silly belief, and a misunderstanding that failed to realize that “intelligent AI” wasn’t synonymous with “unbeatable.” (It also failed to understand that coders didn’t actually have to give AI lightning reflexes, perfect execution, or other inhuman advantages.)

        There was also the belief that players would reject more capable enemies. (Which was easy to argue as true, even in the face of successful high difficulty games.)

        • Panther_Modern says:

          The trend of underinvestment in AI is bizarre to me given the massive success of the Halo franchise and how much of what made the single player experience seem groundbreaking at the time was a consequence of the AI

          • Baines says:

            Halo is used to support both sides of the AI argument. Halo does nice stuff with actor logic states (pro-AI), but it also does a fair amount of smoke and mirrors illusion (supports both sides), as well as relies on a lot of set-piece scripting (often an anti-AI argument).

            Halo might not be using the most egregious cheats, but it does use a number of things to make it look smarter than it is. For an example illusion, Bungie understood that players tend to think that enemies that die quick are less intelligent, so they simply gave enemies more health. For more of a cheat, in order to discourage players from run-and-gun, the AI simply becomes more accurate the longer the player remains out of cover.

            Others might point out that Bungie actually simplified its actor logic system a bit, because the original implementation turned out to make it too difficult for players to suss out NPC behaviors.

          • Otterley says:

            @Baines: That’s some interesting background info :) I know I enjoyed the result at the time.

            As far as the development of AI is concerned, I don’t really mind which approach or mixture of approaches is used, it’s the result that I’m interested in as a player. At least they should be getting significantly better at the smoke & mirrors after so much time. It’s an area that feels more stagnant than many other aspects of game development.

        • benzoate says:

          It’s almost as if developers were ignoring their duty, or calling if you may, when it came to ai design and instead chose to throw wave after endless wave of dumb enemies at you.

          I haven’t the faintest idea where they got such ideas from.


    • LewdPenguin says:

      If this article is to propose FEAR as a missed signpost to where AI could/should have headed, may I propose Red Faction as it’s companion for environmental destruction?
      When the first game arrived it certainly wasn’t a great shooter but I felt surely it was a sign that at last games had caught up to my imagination of how things should work, and would now steadily adopt comparable world-wrecking systems across the industry, at least so far as applicable ie in the shootymanface genre. Yet here we are many years later, the series itself has seemingly died out (probably in part due to having relied on its one spectacular trick a bit much over actually being overall good games), and it remains a massive outlier in the body of videogames as a whole rather than a sign of glorious level-levelling carnage to come. Of course not every game with guns and/or explosions needs to let you trash the whole level so thoroughly, but that ‘simple’ things like being able to shoot through a flimsy door or (usually heavily signposted) bit of wall is still a bit of novelty ‘oooh that actually works’ factor shows how far we haven’t come as a whole.

      As a side note for “things in videogames that don’t ever seem to get better” I’d also like to mention sound, and specifically sound effects, because I don’t care how much money you spent having [famous orchestra] perform your grand, sweeping original soundtrack by [famous composer], if there’s only one sound for footsteps (played quieter when crouched ofc), one sound for every single door the player ever interacts with and so on, and absolutely zero ambient/incidental sound effects then your game has crap sound design. I notice this especially badly in [activity] simulator type games where a developer spends a great deal of effort modelling a bunch of vehicles to somewhere between reasonable and incredible detail, then the whole lot sound like coffee makers/each other. There are a few more exceptions here where companies really put effort into great audio to match their fancy pretties on screen, but there’s also still a few too many Train Simulator ####’s around that throw a few generic sounds onto everything and call it quits.

      • aethereal says:

        On the topic of destruction, I would also bring up Silent Storm, released in 2003. It was a turn-based WW2 strategy game, (similar in principle to the new XCOM), but with wonderful destruction mechanics. You could blow holes in buildings to make your own entrance, with a big enough machine gun you could carve holes in floors (making baddies plummet to their doom), even little things like seeing a toilet shatter when someone falls on it.

        To date, I have not seen a game allow utilization of environment destruction so strategically. The Battlefield series to an extent, but there the destruction is always oddly limited to only specific parts of buildings and in specific ways, somewhat limiting the creativity you can have.

        • nearly says:

          I don’t really know how it holds up on the technical side of things, but it’s both exciting and a little disappointing that Rainbow 6 Siege is a recent highlight for destructable environments. That they made it so central is great, but that it’s also still fairly limited is disappointing.

          I think I also read that one of the newest maps supposedly adds bullet caliber based destruction, which is interesting because it implies that this is something that can be iterated on within a given game’s lifespan if the work is done and people know what they’re doing. I guess maybe different Battlefield expansions/map packs have had varying levels of destruction, but it’s nice to imagine games in general just upping their game through updates in an increasingly stagnant (mechanically) industry.

  7. haldolium says:

    Yes, the F.E.A.R. AI is still one of the best out there, or at least one of the systems you can observe in all its prettyness.

    If anyone is actually interested in the GDC documentation, here is the GDC PDF:
    link to alumni.media.mit.edu

    • aldo_14 says:

      It’s fairly astonishing to consider that FEAR’s AI approach is deriving from decades old automated planning work (the GDC paper, for example, mentions HTN planning – which dates back to the 70s as a concept).

      • haldolium says:

        I think it’s not that surprising (though definitively very interesting), many methods that are still used today in technical areas derive from rather old systems and haven’t changed in their core definition for decades because it is just the best way to deal with things.

        • Josh W says:

          Also, despite some closer academic links on the side of procedural design (possibly because of the “leave it run” nature of the problems being dealt with), the development of enemy AI in games isn’t super organised, so older approaches might be expected to have more chances to percolate into games design.

  8. zulnam says:

    Yeah the AI in FEAR was a selling point from the very beginning, but this was back in the day when not all marketing was lies.

    I remember the advanced AI was also the reason why you would never face more than 6 enemies simultaneously (in the first game). That might’ve been marketing lies, since I imagine 10 smart bots in such small environments would bug out pretty fast.

    Still, great game! Shame about the sequels. I would’ve prefeered they follow FEAR into a new adventure but NOPE, same fucked up kid every time.

    • Hedgeclipper says:

      there would probably be ways round it – two squads of six from each side perhaps, or one on your level and another on a higher separate floor. My guess would be that more than six starts to get too confusing for the player and hides all the fun of the AI in the mess.

  9. caff says:

    Great article. Wish I’d persevered with this when it came out – strangely, it felt a bit generic shooter-ish to me, but perhaps I was just going on the initial early visuals and not paying attention to the AI.

    • Nauallis says:

      That would be because it is and was a very generic shooter. While the mechanics of the AI are impressive in retrospect, the entire plot is contrived trashy psychological horror, and the gunplay really isn’t anything to write home about.

      But hey, that’s the beauty of hindsight.

      • Ralsto says:

        Good lord, you sound like something that slid right out of John Walker’s ass.

        • Juan Carlo says:

          It’s true, though. The action was great, but the horror was a straight “The Ring” rip and already a cliche, even at the time. It’s still a fun game though, due to aforementioned action

        • Josh W says:

          That man has such a rich diet he poos gold accurate games commentary.

          Although actually I seem to remember the gun representation being extremely well put together, and the mix of tones from horror to action being very well balanced.

          FEAR helped establish that fast scenes, slow omnious ones, jump scares and tactical problems were not clashing moments, but able to reinforce one another far more powerfully and with much more shifts between them than they could achieve in other media, that lead to many other successful games (and also many other failures as games also absorbed it’s self absorption, the decay of the deadspace series being an example of this problem).

  10. CartonofMilk says:

    i seem to be one of the few people int he world that didnt get what was so great with F.E.A.R. I played and finished it then and frankly its one of those many very unmemorable game i’ve played in my lifetime i can barely remember anything about. An ok title to spend time playing while waiting for great games.

    • MisterFurious says:

      I remember the game crashing every time I got to the room full of blood and guts and I couldn’t finish it.

  11. ffordesoon says:

    I think the problem is that the default assumption in videogames is still that enemies – and especially the generic mooks which make up the rank and file of any game’s bad guy roster – are there to be killed. Why waste time on complex (or apparently complex) baddie AI when most players won’t notice it? Not to mention that the only thing players will notice – and instantly upload to YouTube for all to see – is when the complex AI screws up and does something idiotic (which it inevitably will), at which point the game will gain a reputation for “bad AI.”

    I mean, look at what Mass Effect Andromeda is going through right now – whatever you think of it, the game’s animations are mostly no less competent than the usual Bioware stuff. But the competent animations don’t make for entertaining YouTube videos or GIFs, because animation is one of those things players have been trained to not notice unless it fucks up. And animation is something players are supposed to see, even if it doesn’t register as especially good or bad; AI is meant to be invisible.

    All of which is a shame, because great AI can make good games awesome.

    • Baines says:

      the game’s animations are mostly no less competent than the usual Bioware stuff

      That’s a pretty low bar, though. If Bioware’s usual animation was better, then Bioware games wouldn’t be such ripe targets for animation complaints. The companies that routinely get raked over the coals for particular issues are generally companies that have done something to earn that response.

      As for ME:A, that game’s animation isn’t just being unfavorably compared to games by other companies (both new and a decade old), it has been unfavorably compared to previous Bioware titles. (Heck, even saying that a big budget game’s animation is of similar quality to its decade-old grandparent is arguably an unfavorable comparison.)

      • nearly says:

        I don’t know, I think people forget that the original Mass Effect (on Xbox 360) was streaming from disc and thus had some of the worst texture loading issues in recent gaming. There were some pretty significant delays between starting a cutscene or loading into play, and having the actual textures show up.

        I’m also at a bit of a loss when people say Bioware has always struggled with animations because their animations have generally been pretty unobtrusive or at least equivalent compared to something like Bethesda’s third-person perspective animations. When people say “bad animations,” I definitely don’t think “Bioware!” first and honestly can’t even really think of another game from them with notoriously bad animations. Bad hair in Dragon Age: Inquisition, sure, and Dragon Age 2 is hilarious with gender-swapped animations, but it sticks out as weird to me that people have latched onto bad animations as a Bioware thing. There also seems to be some serious rose-colored lensing.

  12. SIDD says:

    The AI in Insurgency is the reason why the co-op is still so bloody brilliant. It has gone through countless iterations and it is by now one of the most superior pieces of AI coding in any recent FPS.
    It is as far as I can recall the only AI to rival F.E.A.R.

    • TheSplund says:

      I did rather feel that the AI in Insurgency was rather good – not sure I made a concious comparison to F.E.A.R. though.

  13. dethtoll says:

    Having replayed it last year I can confirm that it is some truly brilliant AI. The game has aged somewhat — those graphics, my goodness! — but it is still a solid entry with an unconventional story and truly fantastic gameplay. The environments may be samey, but that’s because they’re intended to be playgrounds for the AI.

    • dufake says:

      True. Levels should not be art gallery. Level designers can make it beautiful but it shouldn’t destroy its function as a game. Bioshock has insane level details, but it doesn’t ruin the gameplay.

    • vocatus says:

      I’d like to throw in that the sound design is outstanding. I remember cranking the sound way up just to experience the visceral “punch” of the grenades going off, the automatic weapons fire, and the awesome pitch shifting when you go into bullet time.

  14. Kasjer says:

    I have never played F.E.A.R. :( Will pick it up from GOG I think when I’ll trim my “next to play”/”need to finish” games list a bit.

    As for this AI stuff, it is for sure impressive. The AI I was always impressed the most with was Half-Life soldier AI (not female asassin AI, though, these were simply dumb enemies with overporwered guns, pre-calculated waypoints and springs for legs) and Unreal Tournament’s bot AI when they were set to reasonably high level. After that, nothing really had “wow” effect on me.

    • Hedgeclipper says:

      Some playstyles work better than others – pull back to a choke point for each fight and it doesn’t seem as much fun, but it really shines if you take an aggressive move and shoot approach as the AI scrambles to keep up and move about to counter your attack. I also found it fun replaying some of the bigger firefights several times with different tactics just to see how they reacted.

    • Optimaximal says:

      Half-Life’s AI (Marine and Ninja) was almost entirely trigger-based. The combat arenas were heavily planned and based on where you were/what you did triggered various pre-scripted actions (i.e. hiding behind certain cover meant they threw grenades near you, heading for the high ground meant they’d reposition etc), with the obvious allowance for grenade avoidance and firefights.

      Almost purely smoke and mirrors in retrospect.

  15. Premium User Badge

    Otamono says:

    I heard in a german podcast that one of the journalists talked to the developers of fear, and they told him that they just put many different scriptsequences for the enemies into the game, so everybody would think that they have a great AI. But in fact this never was the case.

    • Snowskeeper says:


    • Petethegoat says:

      I’d assume he was referring to the animation sequences that the AI would play as it’s primary means of interacting with the world. Stuff like knocking over filing cabinets for cover, jumping through windows etc was basically an animation it could choose to play depending on the circumstances, as were things like basic movement and shooting.

      • algor says:

        Yeah, the environments are filled with markup that is necessary for these AI to seem intelligent. They can flip over objects to create better cover. An AI on fire will move to water and put themselves out if its available or fling themselves dramatically off a ledge if not. All of the environment movements like jumping through windows or over railings or crawling under toppled shelves is hand placed. Well- at least it is a systemic feature that is hand placed at each instance. Every position the AI can intelligently fire from, like around a corner or over cover or searching for players who can’t be found is also hand placed.
        Specific verbal callouts like “he’s behind the sofa” is also hand placed. The AI has no concept of what a sofa is.

    • haldolium says:

      You should read up on how computer games A.I. is constructed (or read the paper for FEAR).

      There is no “A.I.” as you would imagine it from badly exlained SciFi.

      Every A.I. in games needs, for the sake of processing power and other factors, to know the environment and know simplified interaction abilities upfront through pre-defined places within the world. They do not “see” the environment and “scan” it visually (see at this marvelous competition for DOOM how that would look: link to vizdoom.cs.put.edu.pl), that is not how it works in games.

      They also always have a pre-set amount of rules they can apply. Every 3D world A.I. works that way to my knowledge (this is for example how the UE4 deals with it: link to docs.unrealengine.com), the trick is within the choice of behavior and how to get there.

      A.I. navigation and behaviour isn’t magic, it’s an art form as any other component of games is too.

  16. rymm says:

    killzone 2 came out a few years later and pulled all the same tricks described here, i specifficly remember the ai being pretty good at flushing you out, and letting you know through barks that they were working together to do so. or at least giving the impression. it was a fun game, pretty much solely because of this. sadly 3 didnt seem to remember the improvements and i didnt even finish it. cant remember the last time i noticed ai in a posative way since then

  17. Sin Vega says:

    Never managed to get into this one, as I grey found the grey opening grey corridor grey levels so boring and the grey ‘spooooky’ stuff so old grey hat, and the grey early fights and grey weapns so underwhelming it was grey a struggle grey to care enough to grey grey grey.

    I do wish AI in games was better though. The vast bulk of the industry seems to have spent the last couple of decades obsessing aout tedious crap like framerates and textures, while your average AI opponent has barely improved at all.

    • bill says:

      The combat is indeed great, partly due to the AI and partly due to the feel of bullets zinging past your head.

      But the level design is ridiculously dull and repetitive. And in some ways it’s basically the same fight repeated 1000 times over 10 hours.

      I never checked to see if it had an AvP-style botmatch, but that might be the best way to experience it.

  18. Cvnk says:

    So how do the sequels stack up? Pretty sure I only played the first one. If 2 and 3 can boast the same level of AI sophistication I might pick them up.

    • Vandelay says:

      Don’t know about 3, but 2 certainly doesn’t. In one of those moments that makes me realise how different my game priorities must be to other players, they decided to go with the tedious pop in and out of cover enemies that have been so in vogue for about the last decade. Was very disappointed, as the AI and the combat was FEAR. Removing that made it pretty dull (and replacing it with nonsensical mechs didn’t help.)

  19. bill says:

    I just finished it for the first time, and I totally agree about the AI. The firefights and AI are excellent and do a much better job than most FPS I’ve played.

    The downside of that is that there’s basically only one enemy type and 2-3 weapon types (plus grenades) that just get repeated over and over again for the whole game.

    Assault Rifle / Shotgun battles with mercs * 1000 = FEAR.

    I enjoyed the combat a lot, but it became a real struggle to force myself to finish the entire game.

  20. Ralsto says:

    nnnngh. stop using a shotgun to try to do long distance shots!!!!

  21. lalabox says:

    I feel like what actually really sets FEAR apart is how the level and weapon design facilitates all the ai. The levels may look like boring warehouses and office cubicles, but almost all combat arenas have some sort of looping design, allowing you to circle around and flank the enemies, and for them to do the same to you. You very rarely get bottlenecked, and it’s rare (although it’s still a viable tactic) for you to have to peak out and try to headshot stationary enemies for the fight.

    None of the weapons are the super accurate and super lethal headshot guns that you get in a lot of modern shooters. They are all kind of inaccurate, and make you have to move up on enemies constantly to actually be able to shoot them.

    The actual “goals orientated ai” system has been used in countless games since, but there have been very few games where everything has coalesced to make it work together quite so well (i’d argue that Wolfenstein: TNO has really really great arena design, although enemies feel a bit less aggressive and don’t show off quite the same range of tricks).

    • KenTWOu says:

      You took the words right out of my mouth. F.E.A.R. AI shines because everything else in the game let him shine. That’s why people should stop praising F.E.A.R. AI and start dissecting the game’s core gameplay loop instead. Besides F.E.A.R. AI is not the best one in first-person shooters, its time has come and gone a long time ago. The foreword to this article about F.E.A.R. AI is proving my point.

      The F.E.A.R. AI was a turning point in Game AI for its use of planning — but it’s not “still current” in any way. Many games have taken planning further along with significantly improved underlying systems for cover and pathfinding. Other games based on Halo’s behavior tree approach have also shown that planning is rarely a sensible solution for action/combat/shooters, and the industry has moved beyond those for a variety of reasons including design choices and production benefits. For games with amazing bots or enemies, see KILLZONE 2+, RAGE, SECTION 8: PREJUDICE, TOMB RAIDER, THE LAST OF US, etc.

  22. phailhaus says:

    Fantastic use of gifs! Really highlighted your points without needing me to play a bunch of videos.

  23. kerndaddy says:

    I’m SO off to re-download me some F.E.A.R..

  24. Optimaximal says:

    One reason the article doesn’t really touch on as to why it’s clear AI hasn’t progressed is that Monolith had to give the player a super-human advantage to weight the combat in their favour. The same can be said for Halo – the player character is significantly more capable that his foes, except in certain one-on-one situations (likely for game balance reasons) or where vehicles are involved.

    With the sea-move to modern military shooters lacking supernatural themes, it’s clear that more advanced AI would easily overwhelm the player (especially with the typical jam-vision health system, which relies on the player being able to stop and recover).

    The only counter for this would be to introduce the same AI systems for allied characters, but this would have the side effect of the battles playing themselves (*insert CODBLOPS first level joke*). I guess it would also take more CPU time away from the shiny graphics.

    • UncleLou says:

      Interesting point, and this might explain why games like ArmA couldn’t have as good AI as FEAR even if they wanted, but 95% of first and third person shooters aren’t military simulations.

      And if AI generally was better, the cannon fodder numbers of enemies you get in most shooters could simply be reduced.

      Why not have two òr three smart enemies instead of two dozen dumb ones? I always wanted a shooter where I could have a “Hollywood” shootout that lasts for minutes – everyone hiding, trying to sneak around and get behind the other one, listening for sounds, etc. FEAR was a little like that at least, with the AI circling around you.

    • Dreggsao says:

      You get supernatural abilities in most military shooters too though.

  25. TheSplund says:

    I found AI around this time rather exciting and had decided on doing my final Computer Science Degree work on game AI (a plan which fell through in aprt due to pressures of being a very late returner to education) – I really felt that AI was going somewhere and games like F.E.A.R. were part of that movement.

    • Kaeoschassis says:

      I thought the exact same thing. Then I thought the melee in Dark Messiah would change the future of videogames. More recently I can remember almost literally yelling at my monitor that the way the Nemesis system in Shadow of Mordor works had better catch on.

      It’s a little upsetting, really.

      • haradaya says:

        Right there with you both. Back then I thought when a game pushed the envelope, that was the new standard for future games.

  26. Themadcow says:

    Chaos, the 33 year old Spectrum game by Julian Gollop has better AI than most modern shooters.

    • PanFaceSpoonFeet says:

      My knee jerk was to say that chaos’ AI requirement was way more straightforward but when you break it down.. you have a point. AND it was devious.. quite an achievement.

  27. UncleLou says:

    Great article, and it is pretty much by biggest disappointment in gaming that FEAR (and Half-life 1, to a lesser extent) were *not* a sign of things to come.

    Every shooter that’s not at least in the same ballpark as FEAR when it comes to AI gets deducted two points. Unfortunately, that’s pretty much all of them.

    • dufake says:

      It’s simply good AI will destroy causal playerbase. Unless there’s a crowdfund project I doubt we will see better AI than FEAR. You also need good level designers to make it work.

      • UncleLou says:

        FEAR wasn’t more difficult than shooters with dumber AI was, though, was it? It just felt more believable and real. And you can balance difficulty with all kinds of things (number of enemies, hitpoints, their accuracy), you don’t need to make them act stupid.

        Re: level design, I remember there was something (probably on Gamasutra) on how the level design in FEAR is one of the most important aspects why the ai worked so well.

  28. Spakkenkhrist says:

    It’s also coupled with what I consider to be some of the best gunplay in an FPS.

    I seem to remember Far Cry 1 & 2, and Crysis had pretty good AI.

    Great article, definitely the kind of thing I come to RPS for.

    • Dreggsao says:

      Only thing I remember about the AI in Crisis was that when you sniped into a base everyone in this base started walking towards your position. They all did so on the same route making them look like ducklings. Felt like a step-down from the AI in Far Cry that would have had flanked you (with half of the enemy’s AI bugging out halfway though.)

  29. Kaeoschassis says:

    I think I mentioned this on this very site recently, but it’s worth repeating.

    The best thing about FEAR’s (stellar) AI is that you can absolutely effing terrify it.

    I mean no, obviously, you can’t ACTUALLY scare it. But on those rare occasions when you’re completely curbstomping it, it KNOWS, and you can hear the panic in the enemy’s desperate cries of “cover me!” and “he wiped out my whole squad!”.

    The first time a squad commander yelled for his teammates to move up and the response was an immediate and emphatic “HELL NO” was probably, for me, the most iconic moment in FEAR.

  30. uzernaem says:

    I know this is a PC gaming site but what about Killzone 2 AI? At the time it was considered equal if not superior to F.E.A.R.’s AI.

  31. kud13 says:

    I need to play F.E.A.R. It’s been sitting on my GOG shelf long enough.

    My gold standard for AI remains S.T.A.L.K.E.R. it did a lot of similar things (flanking, grenades, barks), although none of the window-reading stuff.

    But those intense early firefights in Garbage, where it’s you against 2 or 3 bandits, and all you have is a sawed-off, and if you don’t kill them, you’ll die while you’re re-loading… yeah, that was real challenge.

  32. Muzman says:

    I tend to think Call of Duty was the end of the line for truly innovative FPS design and behaviours (at least for ten years).
    That game’s success just showed that if you put all the effort into scale and big set pieces, no one really cares all that much about complex AI and things like that. Which would save a bomb in innovation and testing (who really wants to spend time troubleshooting complex AI and its interaction with level design if you don’t have to?)

    After that it was just script script script. And the audience loved it. Developers probably did too (senior management anyway)

  33. seffard says:

    FEAR AI at the time was a real breakthrough, but I gotta say about The Division’s. Though the general ai isnt even worth mentioning, the Survival Hunters were really challenging and player-like to the point of people confusing them with real players when they first saw them in a pvp match. I consider them a progress.

  34. MOOncalF says:

    My most enduring memory of Fear (while I liked it) is the towering annoyance I felt every time I heard AI radio chatter before I reached the enemy, if creepy ghost girl can ambush me, why can’t the supposedly trained psychic soldiers?!

  35. headless97 says:

    I’ve been playing through FEAR again, which makes this a very timely article. My latest personal challenge is to play the game with no slo-mo, no healthpacks (besides picking up boosters, which restore health), and no weapons other than the pistol. FEAR ends up playing like a stealth game. Using the environment is critical in order to conserve ammo and positioning becomes an art.

  36. lukemiller150 says:

    Hey, It’s really amazing, i read your blog that was really awesome and about this game what did i say? superb ! You can also try it https://getasmurf.com

  37. Edski says:

    I started playing through FEAR for the nth time last week after purchasing a swanky new g-sync monitor. It’s good. The AI is very solid. Next stop, Far Cry.

  38. BaronKreight says:

    I played this game a long time ago and I still remember my impressions of it – the AI seemed very very good.

  39. manny says:

    Wouldn’t commercial squad combat a.i be monitored and prevented from becoming too advanced, since it’s military technology? After all, wargame boardgames were used in real life wars and highly classified.
    But regarding fears planner a.i, perhaps someone can kickstart a addon for unity or unreal game engine. This would enable devs to get access to more advanced a.i for their fps without much fuss.

  40. birowsky says:

    I stopped playing some time after this masterpiece. If I decide to get back at it, what are the three most amazing / intense feelings FPS games i should start with?