Accounting for the *Interestingly* Tuned AI in Brink


(Coolaguy) #1

Accounting for the Interestingly Tuned AI Parameters in Brink

In this post, I propose a potential explanation for the ‘interesting’ design decisions made by Splash Damage in regards to the implementation of the Artificial Intelligence of the ally and opponent computer-controlled players (i.e. bots) in Brink…

…where I use the term ‘interesting’ as a euphemism for ill-conceived and poorly executed, as judged by the “broader gaming community”

…where the “broader gaming community” represents the majority of non-super-hardcore video game players who hail from FPS franchises such as Halo, Call of Duty, etc. including the members of the press who review video games, but necessarily excluding (solely by virtue of this definition) elite gamers who themselves include but are not limited to the testers and developers at Splash Damage.

i.e.
broader gaming community = a + b - c - d - e
a. First Person Shooter fans familiar with the FPS mechanics in franchises such as Halo, CoD, etc.
b. Members of the Press required to review new commercial properties by virtue of their profession
c. Elite (“Super-Hardcore”) FPS Gamers (excluding testers and developers at Splash Damage)
d. Developers at Splash Damage
e. Testers at Splash Damage who worked on the development of Brink

Defining What People Mean When They Criticize Brink’s AI as “Poor”:

Richard Ham, Creative Director of Brink, gave an excellent interview with Machinima’s Hutch, where Richard explained Splash Damage’s rationale in defining and setting the parameters that govern the way that AI (ally and opponent bots) behave in Brink.
[Part 1 (starts at 36:04); Part 2]

Before we get to that, I’ll attempt to summarize what people seem to consider sub-par about the AI in Brink. The following list is not comprehensive.

  1. Ally AI players will stand around “doing nothing”.
  2. Ally AI players do not seem to support the player in completion of primary objectives when the game timer has more than two minutes remaining on the clock.
  3. Ally AI players labour excessively over the achievement of trivial secondary objectives [i.e. Command Post capturing]
  4. Ally AI players make a predictable and concentrated effort to complete the primary mission objective when roughly less than two minutes are remaining on the objective game clock.
  5. Ally AI medics will go to heroic lengths to traverse the entire map, completely ignoring objectives, their teammates, and enemies with the sole purpose of tossing a downed human-controlled player a revive syringe.
  6. Enemy AI players have poor situational awareness but have uncanny response time and accuracy as soon as they are fired upon by a human-controlled player.
  7. AI players sometimes bug out and get stuck on objects.

With the exception of item 7, which represents a bug that I have not been privy to, every “behaviour” in the above list is governed by a set of parameters that Splash Damage defines in a “configuration text file” which can be tweaked, updated, and modified at any time according to Splash Damage’s discretion and does not, supposedly, require the release of additional patches.

If you don’t buy this explanation up front, and insist that the ‘stupid’ AI behaviours are the result of incompetent programming, then consider the following:

The enemy AI has no problem completing primary objectives and otherwise overcoming the combined efforts of you and your AI allies. Furthermore, the enemy AI ramps up in difficulty when additional human-controlled players are added to a team.

When one starts to understand that the AI behaviours that initially seem manifest of poor programming are, in fact, the culmination of defined parameters and deliberate game design choices, one can appreciate why the manner in which the AI has been tuned is ‘interesting’ as opposed to “terrible”, “pathetic”, “broken”, “inexcusable”, etc.

Richard Ham states that many of these choices were implemented with the intention of providing players with the experience of “being a hero” on their team, as opposed to having the game seemingly play itself on auto-pilot while the AI complete primary objectives for the player.

Although this is a noble intention, the implementation does not achieve this lofty design goal in instances where the skill of the player is unable to compensate for the disadvantage that the human player is subject to by virtue of the nerfed ally AI.

That is, Brink is quite clearly a team-focused objective-based first person shooter that relies on cooperation and coordination with your teammates.

As such, in the absence of solid game-sense and significant FPS skill (not to mention the cooperation and coordination necessitated by Brink’s game design), a human-controlled player is at a loss to compensate for the disadvantage that they are placed at by virtue of Brink’s AI design. The result is more of a frustrating experience for players than a rewarding one, at least for a significant component of Brink’s player-base. Compensating for deliberately inadequate ally AI just isn’t fun for this demographic.

Furthermore, these design choices are obscured from the player and come across as the result of a far more endemic problem among video games: poorly designed and programmed AI*.

* - As opposed to Brink’s robust but apparently toned down ally AI

Accounting for the Interestingly Tuned AI Parameters in Brink

So, how can we account for the disparity between the intention of Splash Damage’s developers and the actual gameplay experience in the release version of Brink?

Well, some believe that we lack the programming language to describe a perfect world in Brink… however, I have a different explanation.

I propose that members of Splash Damage and its testers didn’t properly accommodate the skill variance among gamers, and that they tuned and balanced the game for the limited player-base that they had at their disposal.

The player base that I am referring to consists of the developers of Splash Damage and the testers who worked on Brink. This player base is a skewed representation of the gaming population, and reflects a pool of gamers who possess, on average, markedly higher levels of both game sense and skill than the broader gaming community.

In conjunction with their own biases of familiarity (i.e. they know the Brink gameplay mechanics inside and out) and according lack of an objective frame of reference, they couldn’t properly anticipate or account for the relative lack of game sense and skill of that the remainder of gamers would have in regards to Brink. Brink is a complicated game, after all. Likewise, Brink has a number of elements that have attracted a relatively wide variety of gamer segments who initially found Brink compelling for one reason or another and who ended up playing (or being mandated to commercially review) the game despite not having backgrounds as super-hardcore FPS players (i.e. Splash Damage’s videogame developer lineage).

How Have AI Design Decisions Affected Brink’s Reception Among Gamers?

The result is that Brink has been met with harsh criticisms of it’s AI.

However, as Richard Ham has stated, these AI choices have apparently been more of a reflection of game design rather than the limitations of the AI programming itself.

What’s Next For the AI in Brink

Now that this wide variance in game sense and skill among players has been recognized, hopefully Splash Damage will make the appropriate adjustments to more-successfully cater to a wider audience… which was supposedly their intention from the outset.

Richard Ham has stated as much.

As an obvious starting point, allied teammate AI, in one implementation, could be beefed up on the easier campaign gameplay settings. That is, easier game modes should allow AI controlled allies to place an emphasis on completing primary objectives provided that they receive an appropriate level of human-controlled player support or involvement.

I trust that making these improvements to Brink’s AI is far easier said than done, but the current implementation of AI in Brink seems to be inadequate in the eyes of a wide variety of players.

Whether or not this is considered ‘game-breaking’ in the eyes of said players depends on where their gaming preferences lie, be they multiplayer, co-op, or single-player (in conjunction with their aforementioned FPS skill and game-sense).


If you made it this far, thanks for your time and consideration.


(Nail) #2

tep, the bots have been playing for years, they know the game very well


(Coolaguy) #3

[QUOTE=Nail;307872]tep, the bots have been playing for years, they know the game very well[/QUOTE]I know what this is a reference to, but it doesn’t account for why your teammate bots don’t “know the game” as well as the enemy bots. That’s the crux of what I’m saying.


Incidently, I’ve played a few games with Jolly as an AI member of my team. When I buff Jolly’s gun, rarely does Jolly ever reciprocate such kindnesses. I would have thought that when I help my AI teammates (especially those named Jolly) that’d they’d be super-excited to help me in return. Sadly, this is not the case.


(SinDonor) #4

Excellent analysis. But, even as a noob playing on Easy, solo, I’ve had a fun time in Brink and the friendly AI has been fine for me. I get buffed and revived all the time, I have AI allies run to the main objective area with me. And I have help completing main objectives. While I am attempting tha main obj, some of my ally bots are capping com-posts, building barriers and MG nests, manning MG nests, etc etc.

I don’t get understand the criticism. I really don’t. It’s like I am playing a different game compared to all the haters. I’d like to know which games have better bots.


(obliviondoll) #5

Playing solo on PS3 (where I am, it’s still the only option I have), I’ve had a lot of time with the bots.

Easy and Normal, the friendly AI is pretty much only going to even pretend to acknowledge that the primary objectives exist once you get the timer to 2:00.

On Hard, they’re a LOT more helpful, but the enemy AI is significantly more aggressive in combat, and is almost always going to beat the friendly bots when there’s equal numbers on each side.

Also, with any difficulty but Hard, friendly Medic AI will often run from 50m away to try and heal you - which would be great if they tried to take cover or defend themselves when enemies showed up. On Hard, they’ll usually do this, but on Easy and Medium, not so much.

Apparently, the PC version HAS received an update which included AI tweaking, and the XBox and PS3 will see similar updates as well - but having not received this myself yet, I can’t say much for it at this stage. PC players who have mentioned it on the BethSoft forums have said it’s made a huge difference, though.


(Maawdawg) #6

I completely agree with the OP regarding the chasm in friendly and enemy AI in this game. Whether actively intended or simply underthought it severely needs to be adjusted. I truly can’t understand how through testing an polish this AI choice (or the “lobby system”) were thought as good ideas for building a multiplayer community for a new IP.

After playing against the Hard AI for the past 4 days it shows more and more how poor/terribly imbalanced the friendly AI is. The enemy AI is tuned to unrealistic response time and has weapon damage that far exceeds anything human player weapons are allowed. Many times another human and I will have a 2 on 1 against a bot with his back turned but as soon as we start shooting we are downed then he proceeds to stomp finish us. Our two weapons with a jump are not good enough to overcome his nil reaction time to turn 180 instantly (which is literally impossible on controller for a human), ridiculously overpowered gun, and improbable accuracy. Then while waiting to spawn and have it happen again you watch your own bots relentlessly seek command posts and defend secondary objectives completely ignoring what they should be doing. The imbalance, whether purposefully imbalanced or just poorly programmed is really unacceptable.

Right now on 360 I can’t play full human games, which is what I wanted from Brink for in the first place, because there is too much lag. Then Campaign hard and some of the Challenges are basically luck based now with the AI as shipped.

There is a difference between good, challenging enemy AI and giving the enemy AI unfair/unbalanced advantages that just can’t be overcome. Right now this game has the latter. This is a game I was waiting a long long time for, played at PAX with no discernable problems, recommended to tons of people, but now I don’t even want to play (and probably won’t) until the absolute cheapness of the enemy AI is addressed or the ally AI is tuned above moronic and given equal ground with the enemies. I am fine with challenging games. I play L4D2 expert and while it is hard I never come away from a play session feeling cheated or that the game was cheap. I have yet to come off a play session of Brink with anything less than that feeling after playing hard Campaign or doing Challenges.

Here are some of the big buzz lines that were used very often before launch.

"You can win the campain without firing a shot." - Not with this AI you can’t
“There are no one hit kills” - play the enemy AI on hard, you are constantly dropped instantly and often right out to the spectate mode without the downed bar even popping up. Sure, there are no one hit kills for humans, great, that doesn’t help when the computer controlled characters have ARs and miniguns so powerful that 2 bullets drop you after you put a half a clip in their back with the same exact gun and only chipped away around half their health.
“You should always know how you died” - yes I know how, got killed through a window while sprinting down a hallway, while I had full health and 5 health pips, from two bullets that came from 40 yards away, by an AI with a SMG. Awesome.

The absolute imbalance in AI makes the amount of cheap unavoidable deaths in a match ridiculous and completely frustrating. I could and would gladly avoid this issue if the game was playable with all humans but I can’t due to other deficiencies with the online system.

Right now my choices are:

  • slideshow with all humans
  • respawn zerg against unrealistically tuned Hard AI and hope to get lucky
  • try challenges with worthless friendly AI
  • play the game on easy and get little to no satisfaction from it.

Not what I was hoping for at all and none of which I feel are worth my gaming time at this point. I expected full 8 on 8 matches, which I can’t have, and “human level AI” which may be true but due to miserable tuning choices it just seems cheap as an enemy and useless an ally.


(Coolaguy) #7

[QUOTE=obliviondoll;307947]Apparently, the PC version HAS received an update which included AI tweaking, and the XBox and PS3 will see similar updates as well - but having not received this myself yet, I can’t say much for it at this stage. PC players who have mentioned it on the BethSoft forums have said it’s made a huge difference, though.[/QUOTE]Yeah, I expect that Splash Damage will eventually tweak the AI to the community’s satisfaction (even if it is never “great”, it will get to the point where it is “good enough”).

Sadly, though, the critical reception is in, and reviewers everywhere have shat all over Brink’s AI… Interestingly, when the AI is more appropriately tweaked, I think it could actually be considered one of Brink’s strengths. This is just another missed opportunity of Brink’s launch.

Perhaps Sales and Critical Reception aren’t directly correlated, and maybe marketers and publishers understand this… so maybe that’s partly why so many games get launched incomplete and patched later…

However, Brink’s single player ‘campaign’ lives and dies based upon the caliber of the AI. Considering what has been said about the ‘lack of content’ in that regard, I can’t imagine anyone who’s read a review and who cares about single player actually wanting to pick this title up.

Also of note is that with the current implementation of the console multiplayer, there will be roughly 4 AI team members on your squad at any given time. Clearly, even for chiefly multiplayer gamers, AI will still be a significant concern for gamers at the very least until a netcode resolution is implemented.


EDIT: ^This.


(Artheval_Pe) #8

The Parkour, Turret Defense and the challenge with the Bot don’t involve friendly AI.


(Supbitches) #9

Bad ai is bad


(Vlad) #10

Whilst I agree with some points, I strongly disagree about it being a case of skill.

I genuinely believe that you could be one of the most skilled people out there and the lone wolf styled AI, (you know, the way it tells you not to play through all the loading screens and on the manual if you “want to succeed in Brink”), could still easily screw you over.

Friendly AI runs about neglecting the primary objective thus leaving you alone for the most part if you’re actually trying to do the primary objectives yet the enemy AI swarms you, kills you in just a few bullets regardless of what size and abilities you possess and then camps at the primary objective, then you alone have no chance to get anywhere close to it.


(TruGamer97) #11

The ai on your team is ridicules i always feel like im doing all the work


(IVIAK) #12

completely agree, to hell with this game


(METRiC) #13

[QUOTE=Coolaguy;307863]Accounting for the Interestingly Tuned AI Parameters in Brink

i.e.

e. Testers at Splash Damage who worked on the development of Brink

You mean, it was tested?

Someone please fire those guys


(TruGamer97) #14

I still enjoy this game but a bit better ai is all im asking till i can go online on my xbox


(frizlefry) #15

Playing single player with these bots is torturous. All I wanted to do was get the Steam achievements as efficiently as possible but it was basically impossible even on Easy. My team is standing around doing nothing while the enemy is gunning me down during a hack over and over. I can kill all the enemy bots but they just come back or res on the spot and I am dead again and again. The ally bots do nothing and die like flies to the enemy bots. They will not shoot an engie removing a hack box with 95% percent complete on it when the enemy is doing it right in front of them.
The single player content is one of the most excruciating awful experiences I have ever had the misfortune to pay for. I do not believe the game was tested or that any significant amount of effort was put into the AI programming for the bots.


(Coolaguy) #16

[QUOTE=Vlad;308171]Whilst I agree with some points, I strongly disagree about it being a case of skill.

I genuinely believe that you could be one of the most skilled people out there and the lone wolf styled AI, (you know, the way it tells you not to play through all the loading screens and on the manual if you “want to succeed in Brink”), could still easily screw you over.

Friendly AI runs about neglecting the primary objective thus leaving you alone for the most part if you’re actually trying to do the primary objectives yet the enemy AI swarms you, kills you in just a few bullets regardless of what size and abilities you possess and then camps at the primary objective, then you alone have no chance to get anywhere close to it.[/QUOTE]While I’ve played my fair share of games of Brink that play out exactly as you describe, I am gonna stick to my guns on the topic of there being a skill disparity between the creators and the general gaming public. (Which is different than saying that this is the only possible explanation for why there is a friendly-enemy AI competency gap.)

For instance, I’ve seen Machinima youtube dudes overcome the hurdle posed by the friendly AI on one of their first attempts at playing Brink. Machinima youtube casters are just better FPS slayers than the vast majority of the public and they can compensate for the deficient contributions of their AI allies.

Also, I’ve developed enough crowd control and slaying efficiency that I can push a wave of enemy bots back just long enough that I can get the objective hacked / planted or whatever. It is frustrating as hell, though.

Tips:

  • Grenade Launcher attachment is your friend.
  • The Gerrund with COPA for headshots is key for pushing back a wave of enemy AI just long enough to start the objective yourself.
  • Some objectives like Engineer repairs require you just to get ~20% done each time before you get killed while the friendly AI does nothing.
  • Hack objectives are best handled by pushing the enemy back just long enough to get the first part done (while concealing yourself). At about 50%, an enemy will come to remove all your progress. Instead of letting that happen, fire a grenade from the grenade launcher at the enemy engineer AI and then incapacitate him followed by killing him immediately before a medic comes. Resume the hack and your allies will probably, at the very least, distract the enemy AI from killing you while you finish the hack from behind cover.

My belief is, though, that the friendly AI should be far more helpful on lower difficulties than it currently is.

As it stands, even if friendly AI players are directing themselves to the primary mission objective, you can’t even count on friendly AI to kill enemy AI players with any level of parity. Furthermore, they don’t actually complete the primary mission objective themselves…

Thus, the burden of slaying enemy AI lies squarely on your shoulders. Slaying proficiency is a function of FPS skill.

The only way that I can reasonably account for such a significant difficulty hurdle is if Brink’s creators (i.e. Splash Damage) were oblivious that such a difficulty hurdle exists in the first place. In turn, the only way that I can conceive that anyone would be oblivious to such a difficulty hurdle is if they were so proficient that they couldn’t detect it in their own play experiences… or if they tuned the difficulty to give their own skill-sets a workout.

Whatever the case may be, my account for why it is one way or another is just a theory. However, if my theory has any merit, then the basis for my theory gives an indication of how to solve the problem.

i.e.
where [SkillThreshold] represents the skill of players in the lowest 10* percentile (or lower i.e. least skilled) of FPS gamers

* - some arbitrary proportion of players, in order to be as inclusive of the FPS gaming audience as possible

(Brink’s difficulty in single-player)
= the numerical percentile of players actually able to win greater than 50% of the time in Brink on the lowest difficulty setting

FOR (Brink’s difficulty in single-player) > SkillThreshold
IF
(friendly AI) < (enemy AI)
THEN
INCREASE (friendly AI)

In other words, increase the AI of allied bots until the difficulty is no longer prohibitive to lesser-skilled players.

This is the philosophy of modern game design, whereas former (old school) game design tended to tailor difficulty curves to the most hardcore of the hardcore audience.

The basis for this change is a change in the video game industry as reflected by a change in the revenue model for video games and changes in the video game market demographic.

(i.e. Let’s say that a Developer wants to make more money off of their games? Okay, then they either target a game better to the desired audience or, as markets expand and a newer and less-hardcore market segment arises, you make the games more inclusive to a wider audience. This is because games don’t sell based on how much revenue they accrue in coins in arcade machines anymore… A videogame is an entertainment package that is paid for up front and in full, and creators don’t gain revenue based upon having pay-to-play mechanics… at least in most current embodiments of console videogames [DLC is a debatable example of a pay-to-play mechanic, potentially encouraging higher difficulty; Subscription is clearly pay-to-play].)


(Jess Alon) #17

Honestly the friendly AI has made me a better player. I have to basically make a hole for them to get through to the objective and develop brilliant tactics to finish a map instead of letting the AI carry it for me if I’m sucking. They are helpful but not too helpful. And then when I’m working with a decent team full of humans we steamroll over anything.