skacky on 16/4/2020 at 19:38
Hey taffers. I have a simple question for you that should hopefully produce complex and interesting answers. I'm posting this in ThiefGen but it doesn't necessarily apply to Thief fans only.
Thief is 22 years old. We've seen many stealth games since then; Metal Gear Solid, Splinter Cell, No One Lives Forever, Mark of the Ninja, Dishonored... the list goes on. Still, it's hard to argue the genre hasn't evolved a whole lot since its inception considering Thief's mechanics are usually said to be the best and most robust of them all, especially when it comes to sound tech.
So, in your opinion and with hindsight, what aspects of the stealth genre should/could be improved to create better/more fun experiences?
Starker on 16/4/2020 at 21:51
I think you could argue that The Dark Mod has at least some improvements on the original formula and that other games have at least taken it to different directions -- for example Hitman with its social stealth.
So that's also the answer, I think -- you could improve on the original formula by creating a better AI that's more fun to play with, better physics to aid game design with puzzles and whatnot, better world simulation for more reactivity to make more use of social stealth, and so on.
Meowdori on 17/4/2020 at 02:19
The field where most meaningful and crucial improvements could be made is definitely AI, which plays a central role in this type of games. First of all, NPCs should have a much more persistent memory and number of states to accomodate to whatever way the player affected them in. Say, in case you managed to alert the whole mansion, guards should realistically never return to fully relaxed state, or the time needed for this to occur should be very long, provided you'd be able to flee and stay undetected for the cooldown to happen.
Another kind of thing i pondered about before in Discord server, is the audio propagation and AI's responses to it. Dark Engine doesn't account for the influence of the external noise sources, which could affect the chances of AI hearing you. It's gonna be easier to understand looking at this diagram:
Inline Image:
https://i.imgur.com/zkjiBHA.pngAssuming the green pentagon is the player and remaining pentagons are enemies, the chances of the AIs getting alerted by the player's actions should be resultant of some kind of equation between the influence of all audio sources in the proximity, with my picture presenting what is probably the most naive method.
Last thing that comes to my mind at this moment is shadows casted by player which should be visible to the AI, making it possible to alert them that way.
Starker on 17/4/2020 at 04:05
Another thing to keep in mind is how the player is able to take all of this into account and react accordingly. Something like player loudness in contrast with the level of ambient sound at the player's current location might be more immediately understandable and translatable into gameplay actions.
skacky on 17/4/2020 at 10:55
Starker: yes of course, my phrasing isn't a jab at TDM or Hitman that are great stealth games too and do things differently. I thought about loud areas impacting AI hearing as well, and have diegetic indications such as signs in the world that say "wear hearing protection beyond this point", or something like that.
Midori: nice! Your three points are exactly what I think could be improved as well, especially a visible shadow and proper AI behavior on high alert. I would even go so far as to think having the player's silhouette against lit surfaces visible could be interesting, but I don't know exactly how this could be communicated in a simple way. For example when you are in a shadowed doorway but the wall behind you in the next room receives light, you should still be visible.
Nameless Voice on 17/4/2020 at 12:18
I think a really important thing to bear in mind is that gameplay has to be more important than pure realism.
Thief generally manages to strike a really good balance in this area.
Improving the AI to make them smarter and more realistic is only valuable if that increased intelligence makes them more fun to play against.
For example, if you are ghosting a guard and repeatedly run around behind his back at different times, each enough to illicit a comment but not enough for him to start searching, then that guard would realistically remember that he's heard suspicious sounds 26 times in the last half-hour and decide to go on a higher alert. That would be realistic, but it would make for awful gameplay, as you'd effectively have a limited number of times you could go past the guard before that path becomes permanently locked out to you.
Good stealth characters need a good balance of both artificial intelligence and artificial stupidity.
The same should generally apply to other mechanics, too. Does a proposed addition add more interesting gameplay, or just more frustration?
Another point is that the systems also need to be understandable by the player.
I was recently playing around with an AI perception system where everything was numeric, with floating point precision. A sound could be 35.53% suspicious, then another sound could be 37.25% suspicious, and combined they could be suspicious enough to make the AI reach high alert, but then there was an environmental background noise that blocked 27.53% of the sound, so it wasn't a high alert any more.
That whole system was so complicated that it would have been impossible for the player to easily grasp what was going on, especially with how variable it was based on slight changes of distance. (I was even having trouble coding the AI behaviours against it because it was so complicated.)
The lesson that I learned there was to keep the systems simple. Having a sound be "low alert" suspicious at a given distance is much easier to work with and to understand than having that sound be "12.53% suspicious" at that distance.
For the actual question, I think one thing that I don't see enough of is AI manipulation.
Being able to adjust the world so that the AI react in certain ways which will be beneficial to you.
At its most simple, an example would be turning on a generator so that a guard leaves his post to take a look at it and turn it back off, allowing you to either blackjack that guard while he's alone, or get past his post while he's not there.
Starker on 17/4/2020 at 17:25
I think I'd like to see some better simulation in general. Proper schedules for NPCs, fire burning things (Underworld Ascendant), ice freezing things (T2X), etc. Thief was already remarkable in that you could trigger/disarm traps with objects and push buttons by shooting arrows at them (compare and contrast with the conteporary Half-Life the rebooted Thief 2014, for example), but I think this could be taken much further.
skacky on 17/4/2020 at 19:02
NV: excellent points! Balancing fun vs realistic is definitely the biggest challenge when it comes to AI. You want your AI to be clever but not too clever, otherwise it becomes tedious to play since you have the feeling the computer is cheating. I prefer using the word believable versus realistic, since believable things don't have to be realistic, they just need to make sense in the context of the world the game is showing you and its internal rules.
Starker: yeah, NPC schedules fall within the believable spectrum to me. It would be interesting for example to have a guard working night shift check out a few rooms every now and then along their patrol route, but not always, and have them be relayed by another guard so they can exchange their shifts after maybe 30 minutes.
vfig on 17/4/2020 at 19:08
I have thoughts, but they're all a jumble, so I'm just going to respond to some of the suggestions specifically, and try to put together some more general thoughts later:
Having loud ambient noise mask player noise is something Thief ought to do already: it's obvious for the player, because their experience of sound in the real world works this way; and the audio implies it already: the player can't hear their own footsteps over all the noise. But the game's audio system doesn't work this way. So this is an area where Thief's sound design is lacking. (Footnote: LGS evidently experimented with this idea for Thief 2: the "Loud Room" property—unused in the shipped game except for one room where it's applied but turned off—dulls sounds that travel into or out of it; but not sounds within the room. It looks like a relatively quick attempt to build a noise-masking feature using the existing sound propagation system without the schedule-wrecking complexity of changing the AI reactions to sound cues.)
Leaving aside the question of whether being spotted because of your shadow or silhouette is fun or not, making it effective is a problem of player awareness: if the player is not aware that they're visible, they're going to feel like the game is unfair when they get seen. Thief already has an uphill struggle getting players to understand their visibility in light and darkness, even with the light gem communicating this. Most players still operate on the idea that if the ground is dark then they will be hidden there—even if the surrounding light and shadow shows that only their feet and legs would be shadowed. In addition, much of Thief's success in stealth is that its design allows players to observe enemies while remaining themself unobserved; if standing in an open doorway or perched on a high ledge made you more visible (as is likely the real world), this has knock-on consequences for the player's ability to learn about the level and play purposefully instead of merely reactively.
Thief's AI is much more complicated than most people understand, I think. Take Meowdori's suggestion: an AI that has seen Garrett (reached level 3 alert) will already never go back to their fully relaxed state" (level 0 alert)! They only go back down to level 1. It's true the game doesn't do much with this; the default AI senses and behaviour at level 1 and level 0 are similar enough that only supreme ghosters tend to worry about it; although I think most players have probably had the confusing experience that sometimes in full darkness an AI can walk right into them without seeing them, and sometimes they will discover you when they walk into you; the former happens at level 0, and the latter at level 1. And the only feedback the game gives to communicate this different alertness level is that an AI at level 0 will whistle and hum to themselves; at level 1, they never do, only mutter and grumble. So this is all quite non-obvious to the player already! Which gets right back to that question of feedback and player understanding.
Similarly the question of numerical vs symbolic levels of 'suspiciousness' and the like that Nameless Voice brings up. Thief's AI awareness, both visual and aural, is a numerical simulation, with stimuli causing changes to awareness based on how intense and how frequent they are. The AI responses to their awareness however (their alertness) happens when their awareness reaches a particular threshold. And this is complicated by an AI's senses (and thus how quickly their awareness changes) also depending on their current alertness level. As NV says, numerical systems like this end up being difficult to design with (as any FM author who's tried to adjust AI parameters could tell you) and even more difficult to communicate back to the player. Thief even does a bad job of communicating AI alertness levels! A guard can be still at high alertness and patrolling seemingly unconcernedly, only to notice you creeping up behind them and swing around, as if they had eyes in the back of their head (again, something most players have probably encountered and been confused by). Thief's AI mostly communicates if they are in combat, or searching for you, or just walking around; these are largely correlated with alertness levels, but not identical to them.
I really don't think there's much to be gained for the stealth genre by making more complex AI simulations than Thief's already is.
Edit: a brief comment on burning and freezing mechanics: usually these are just fancy lock-and-key systems or fancy magic-spells-to-kill-opponents. I think the only game that has done fire well is Far Cry 2: and there, fire is a hazard. You can set things on fire intentionally, but it's as likely to get out of hand and hurt you as to help you. And you usually wouldn't try stealthing through an area that's on fire, because it's way too dangerous and unpredictable. (Far Cry 2's stealth system is probably worth another shout out for a detailed system that's badly communicated to the player, and so ineffective and confusing for most players—but highly engaging and effective if you do learn how it works.)
skacky on 17/4/2020 at 19:19
I understand the problems that are brought up with shadow and silhouette. Doesn't the AI in TDS notice Garrett's shadow? In any case, I think this could be mitigated somewhat by levels using spotlights effectively thus making the player's shadow quite long and obvious, and have NPCs comment on it if they see it "whose shadow is this?". Of course this is all theory, and maybe it sucks in practice but I'd still like to see this tried to see how well it works.
You're also correct about the AI not communicating the difference between Alerts 0 and 1 well enough, which is something that should be touched upon in my opinion. I think one thing that can communicate this better is NPC stance, they are relaxed on Alert 0 and look visibly more cautious on Alert 1, and so on, and could also have routines where they stop to check their surroundings while they just go on their merry way on Alert 0.
