On the subject of AI and "It's not hard, just do it"

+
Found this on the Outriders Reddit (yes, Reddit, I know, but there is some good stuff there) and it makes a lot of sense. It echoes what devs have said to me and others.

Development is always a trade off. No, this is not an excuse 2077 ah, "interesting" AI. It does illustrate that it is always a challenge, and an expensive one in both time and money, even for big studios.


From user "neatchee" on this thread.

neatchee:

1 day ago

So, just to point it out since I'm in the industry:
tl;dr: making boss fights harder by making them 'smarter' is costly, and if you want that then you need to be okay with slimming back somewhere else in the game to cover the costs.
There is a (not very good) reason why devs don't do the things you recommended sometimes...it's easier and cheaper to add health than literally anything else.
I know it's easy to think of AAA game studios as having a giant vault of money, but in truth dev studios are often run very close to maximum budget, funded purely on capital they received from their publisher, and they have a time-limit. If the game doesn't release before the money runs out, that's it, the company is in big time trouble, loses a lot of money (due to obligations in their publisher contract, usually), and will never see a penny of any "performance-based" payout they might have had in their deal. Best they can do is ask for more money from the publisher. They're the ones who foot the bills upfront.
As for the specific recommendations you made:
  • Smarter AI - We'd all love smarter AI. But it is surprisingly difficult to walk the tightrope between "AI does dumb shit that is exploitable" and "AI is so good they're like aimbots and wreck your shit". If you want more complex/better AI, you have to pay for it.
  • More enemies - Typically fights are designed with near the maximum amount of enemies on screen that the engine can handle. So you either wind up respawning the existing combatants more frequently - which is equally boring, having to deal with trash mobs too often - or making the adds themselves spongier which isn't great. Even if your PC or console is a beast, the game needs to run at a stable framerate on the lowest supported hardware. There are very real technical limits here.
  • More elites - This one is viable but risky without significant playtesting and AI behavior tweaking. The big issue here is that adding more elites doesn't just add more damage targets and damage sources; it dramatically changes the encounter mechanics. Multiple big enemies that can deal significant damage create additional difficulty by reducing available angles of attack, increasing the threat of getting flanked unexpectedly or otherwise being attacked by an unseen foe, etc. I'm sure you know that feeling of "wtf, I can't keep track of this many targets at once, this is stupid, no matter where I try to go there's someone there to kick my ass."
It's not possible to "just add something, it's not hard" because I promise you there is no game studio in existence that releases a game with work hours to spare. Literally every minute of available dev time is accounted for already. Game dev backlogs - work that we want to do but haven't started yet - are always big; there is always more work to do, and not enough time to do it.
So usually the best way to think about these problems is like this (because it very closely reflects how these decisions are actually made in a process called "triage"):
  1. If the new idea costs more design, dev, or test time than the current plan, what other idea/feature/bugfix are we going to cut to fund this new one?
  2. If we can find something to cut because this is more important, can we reallocate the resources without changing staffing? Cutting the number of weapon models down doesn't give AI programmers more time to make better AI.
  3. What's the RoI? Does this change align with other ongoing work, or is it a one-off? Will players experience this one bad thing and then move on, or will it keep being a problem throughout their time playing the game?
None of this is to say that this fight is anywhere near reasonable! It's not. It's broken af. But my guess is that either this was a late bug with some scalar value (enemy health, move speed, recast time, damage, etc) OR it was a known problem and the fix got cut during the triage process.
If you made it this far, I applaud you. Thank you for coming to my TED talk.
 
I do some mapmaking for another game in my free time and while I can cross off money as a actual budget on my part, I do have experience in the sheer amount of time, testing, debugging etc that is involved in such a thing.
If I had to convert my time spend on something as 'trivial' as that into working hours for an appropriate rate, then yeah.


It was a nice TED talk... I guess..;)
 
Is this why they spent time and resources implementing note-perfect guitar animations?

Or is that something that isn't hard and could just be done?
 
Is this why they spent time and resources implementing note-perfect guitar animations?

Or is that something that isn't hard and could just be done?

Are you talking about Kerry's side-quest? Because yeah, obviously they're going to prioritise animations for all the important quests (main quests plus important-character-side-quests).
 
I have no idea what fight (or game) the person is talking about, but I can appreciate someone who clearly knows what they're talking about and presents it in such a clear, informative, and neutral way. What they're saying also makes perfect and logical sense from even just a layman's (my) perspective.
 
Cyberpunk 2077 starts with a lot of corridor sequences where the player is being shown the city but can't really interact with it. We can't just *go out there* until we exit Megabuilding 10 later on. And what do we see? Cars constantly ramming into road blocks next to the place where Jackie's eating and pedestrians acting as if all of them all of a sudden forgot what they were supposed to do right now. And that situation- crowds and traffic- is in that state for the rest of the game. All the time. I guess the trade-off here was "either we make the city bielievable, or we make... what are we doing again?".
 
personally it dont think money/time is an argument for the current status of ai/physics in this game... its just not lacking a bit... its lacking totally. : / they built up an awesome city to look at - but than again they reduce this worlds worth with unfinished physics and bernd-the-bread-ai... imagen how the city/world would feel (and not just look like) if physics and ai would be on same level... that would be the next-level they aimed and advertised this game for. - for now the city just feels like a fassade to me - like the saying "not everything what shines is actually gold".
 
Cyberpunk 2077 starts with a lot of corridor sequences where the player is being shown the city but can't really interact with it. We can't just *go out there* until we exit Megabuilding 10 later on. And what do we see? Cars constantly ramming into road blocks next to the place where Jackie's eating and pedestrians acting as if all of them all of a sudden forgot what they were supposed to do right now. And that situation- crowds and traffic- is in that state for the rest of the game. All the time. I guess the trade-off here was "either we make the city bielievable, or we make... what are we doing again?".

I get the feeling that they were back to their old practices of just painstakingly handcrafting those corridor sequences so they look great instead of building a functioning open world game system that actually functions. This is not the first CDPR has done something like this. They have a tendency of making demos that have little to do with the actual game, just so they could sell the product. Anyone remember the wall-climbing and riding of the 48-minute demo? Guess why we can't do that? Because they faked the demo to appear like they were actually playing.

Similarly we can only drink beverages in the cutscenes in 1st person , because those are animated by hand instead of using the game engine. Cant in open world cause no such system exists.

The gap between those uninteractable sequences and the open world part is massive.

ps. Im not going to say crafting a good AI is easy, but when you have 7+ years and 20-50 million budget, there are very few excuses allowed. Other game companies give the impression of doing better stuff in half the time.
 
Last edited:
Similarly we can only drink beverages in the cutscenes in 1st person , because those are animated by hand instead of using the game engine.
I'm confused by this point, as the animation files exist in the game and the "cutscenes" are done "in engine". Unless the meaning has changed, it is the difference between a pre-rendered scene or FMV and the scene playing out using the in-game assets. Since the animations are in the code, they could be used if CDPR wanted to.
 
It will come down to a cost/time analysis because game AI is not one of CDPR's core competencies. They have no in house experience creating behaviour simulations and you can tell from the previous games they have released. So if they want to start bringing their game AI up, they will need to build that knowledge internally (takes a lot of time and money) or hire externally (a lot less time and a lot more money).

Game AI is a core competency of Rockstar and they have been building chaos simulations for a long time so they have a lot of people and know how to do this sort of thing really well. At this point they probably have the whole thing pipelined and workflowed out the hilt and they probably have their own tools and training so new people can quickly understand how their game AI systems work.

If you haven't done game AI like this before, you have to build all of that and it takes a long time and a lot of money and it still might not work out the way you hope it wil.

CDPR's core competency is more towards single player narrative experiences so things like the writing, character design, environment design are on another level.
Post automatically merged:

I'm confused by this point, as the animation files exist in the game and the "cutscenes" are done "in engine". Unless the meaning has changed, it is the difference between a pre-rendered scene or FMV and the scene playing out using the in-game assets. Since the animations are in the code, they could be used if CDPR wanted to.
They are in the game files and there are people in the modding community working on anims but its very hard because of the enormous number of bones that we do not understand the functions of. There are facial anims in the game that control hundreds of bones (!!). This is the player face cyberware 04 decal mesh:



In the facialsetup files theres a bunch of floats that control bone transformations. I assume this is all black box JALI stuff but either way, we have to be careful screwing around with it because we don't know how it works.

You can model swap basically everything in the game except for performance captured sequences (and I'm not sure why that is but could be a bug?). All the face rigging translates well, even to different body/head meshes and morphtargets. Heres some screenshots of V swapped to various npcs. It works very well: Imgur: The magic of the Internet

 
Last edited:
I can't remember the name of it, but there's an old game for PS1 where you build the AI for a robot and send it off to battle.
It was really eye-opening to see how differently the bots performed compared to what I thought when entering the conditions and reactions into the bot's programming.

I would expect that in a way, it's the same when programming enemy AI for any other game. What makes perfect sense when mapping it out and looking at it has completely different results when put into practice due to unknown or unanticipated variables.
 
Are you talking about Kerry's side-quest? Because yeah, obviously they're going to prioritise animations for all the important quests (main quests plus important-character-side-quests).
So quests with combat aren't important? Got it.
 
So quests with combat aren't important? Got it.

I'll wipe the dust off the part where I said, "all the important quests (main quests plus important-character-side-quests)."
Last I checked, most of those have combat in them.

The point was that animations in all of those quests are going to be high up the list of things to polish since they form the main story. The story is the star of the dish. Police AI isn't.
 
I'll wipe the dust off the part where I said, "all the important quests (main quests plus important-character-side-quests)."
Last I checked, most of those have combat in them.

The point was that animations in all of those quests are going to be high up the list of things to polish since they form the main story. The story is the star of the dish. Police AI isn't.
Police AI? I thought we were talking about the combat AI in general, which is pretty yikes across the board.

Given that combat is foundational to the gameplay, I would think that the combat AI would be pretty important. But you're saying otherwise, so I'm just extrapolating from the premises you've put forward.
 
Police AI? I thought we were talking about the combat AI in general, which is pretty yikes across the board.

Given that combat is foundational to the gameplay, I would think that the combat AI would be pretty important. But you're saying otherwise, so I'm just extrapolating from the premises you've put forward.
Not worse than other open world games but OK, let's juat assume...

You still compare specialised coder to animator when compare guitar play animation and combat AI.
One does not drain the resources from the other. It even was in the OP, wasn't it?
It is like uf you insist that low res texture would make the game longer or something.
 
Combat AI ? You talk about enemys ?
Because i also replay Borderlands 3 (optimized for Serie S/X) now and seriously, i don't know if it better or not. Enemys are not smart at all, while fight is the heart of the game. (simply run towards you, stay stupidly uncovered,...)
(Not be confused, i absolutely love Borderlands 3 :D)
 
Police AI? I thought we were talking about the combat AI in general, which is pretty yikes across the board.

Given that combat is foundational to the gameplay, I would think that the combat AI would be pretty important. But you're saying otherwise, so I'm just extrapolating from the premises you've put forward.

Police AI was just an example since it keeps coming up. But, yes, "combat AI" isn't the star of the dish either. That being said, I didn't say it wasn't important AT ALL. Obviously they have put work into it. I just don't buy into this idea it has to be state-of-the-art otherwise it's shit. Personally I don't see any problems and think it's pretty fun. Maybe it's because I'm so engaged by the story and characters that there's no space in my mind to notice AI issues.
 
Police AI was just an example since it keeps coming up. But, yes, "combat AI" isn't the star of the dish either. That being said, I didn't say it wasn't important AT ALL. Obviously they have put work into it. I just don't buy into this idea it has to be state-of-the-art otherwise it's shit. Personally I don't see any problems and think it's pretty fun. Maybe it's because I'm so engaged by the story and characters that there's no space in my mind to notice AI issues.
Let me break this down.

The premise of this thread was "AI is hard, m'kay?" and that development time is so carefully allocated that, gosh, it just wouldn't have been possible to ship anything better than what we got.

My rebuttal is that development time and resources weren't actually allocated efficiently. This proves nothing by itself, granted. But, it definitely refutes the idea that CDPR just couldn't possibly have done any better because they couldn't have spent another cent on NPC behavior. Maybe it wouldn't have changed anything, but there is clear evidence that the allotment of resources on the development of this game was not done purely with an eye of what was the most important or the most efficient approach.

In fairness to the game, the issues with the combat AI probably stem from more than just the AI itself. Level design, combat mechanics, and enemy balance play a large role as well. But taken as a whole, the quality of enemy behavior feels questionable, especially when compared to other, older games with similar combat and in similar genres.
 
Let me break this down.

The premise of this thread was "AI is hard, m'kay?" and that development time is so carefully allocated that, gosh, it just wouldn't have been possible to ship anything better than what we got.

My rebuttal is that development time and resources weren't actually allocated efficiently. This proves nothing by itself, granted. But, it definitely refutes the idea that CDPR just couldn't possibly have done any better because they couldn't have spent another cent on NPC behavior. Maybe it wouldn't have changed anything, but there is clear evidence that the allotment of resources on the development of this game was not done purely with an eye of what was the most important or the most efficient approach.

In fairness to the game, the issues with the combat AI probably stem from more than just the AI itself. Level design, combat mechanics, and enemy balance play a large role as well. But taken as a whole, the quality of enemy behavior feels questionable, especially when compared to other, older games with similar combat and in similar genres.

But that's not the premise of the thread... The OP is not saying that CDPR managed things so perfectly that Cyberpunk couldn't have been better. Of course it could've been better. The point was to give everyone a more nuanced appreciation of the challenges devs face in managing these things. That's literally it. If you want to make a post explaining how CDPR could've allocated things more efficiently, then great, go ahead and write that post.
 
But that's not the premise of the thread... The OP is not saying that CDPR managed things so perfectly that Cyberpunk couldn't have been better. Of course it could've been better. The point was to give everyone a more nuanced appreciation of the challenges devs face in managing these things. That's literally it. If you want to make a post explaining how CDPR could've allocated things more efficiently, then great, go ahead and write that post.
Then I don't see the point, at all.

Yes, game development is hard. So are a lot of things. Many companies in any number of fields undertake difficult, intensive, and complex projects, and produce an end product that meets or exceed expectations.

When a major studio is struggling to produce AAA-level quality work, then it speaks to deeper issues with that studio -- not just the fact that operating at a high level in a particular industry is difficult or resource-intensive.
 
Top Bottom