My hope and dreams.

Algea

Insider
I am all for that as long as this won't be case for whole villages running from simgle man wielding a sword. I would like to see them run for their friends and then swarm you and kick your butt.
I was thinking about different kinds of personalities randomly chosen and applied by that awesome AI the devs keep hinting on about. So yes, not whole villages, but some of the villagers while others will take their rusty sickles and scythes to fend off the attacker (i.e. you). And city dwellers will run to city guards. Or try to hide behind those barrels in that alley.
 

TheScythian

Insider
Not very realistic though, just look at most hostage situations. People don't try to charge the guy with the weapon.
Sometimes they do if their backs are against the wall, for instance the passengers of flight 93 rushed their hijackers (although I do see your point, it took them a while to organise they didn't immediately fight back).
 
Well medieval villager was used as cheap fighting force ... they had to have some guts. It is whole village against one man with a sword. I dont wan't them all to run.

Yes that seems right.
Well yeah they were, the thing is that if its wartime there won't be fighters there and if its not than all the fighters were killed in the last war. Not to mention they weren't used as cheap fighting because they were particularly good at it or brave.
 

Algea

Insider
I think there is no need to underestimate someone who has nothing to lose either way, right? So while not everyone will be fatally reckless I'm sure there will be some who will be ready to die if it means they can take their attacker with them. In short I'm all up to diversity in patterns of NPCs' behavior.
 
That's because the weapon is a shotgun rather than a sword. If you can run, you can organise a quick swarm against such a threat. Against a gun, you've got no chance.
 

Wakko

Insider
Feels like a village would try to defend themselves if they feel they actually could stand a chance. If a lone man with a sword arrives to raid and pillage, chances are he would just be overwhelmed by mob of angry peasants. Just because they dont know any fancy techniques doesnt mean they are incapable of putting up a fight. Just seems logical to me to try to defend whats yours, if possible. On the other hand if a band of bandits would attack, the odds for the villagers to win would probably be too small for them to have courage to put up a decent fight. Theres not only physical warfare, there's also psychical warfare, mindgames.
 
  • Like
Reactions: Rob
This calls for a Shichinin no samurai quest lol.

But yeah, creating diversity amongst NPC's and enemies is sooo important. I've often played games where you just butcher enemies and I'd think "Why aren't they running away from me?" Enemy AI is too often set on cannon fodder. I'd like to see different approaches to this kind of AI.

The first time I realised this was probably when I played Medieval Total War. Where troops would flee the battlefield, generals dying was huge and such...
 

SergeDavid

Insider
I would also love to see something along the lines of an npc that breaks into peoples houses and redecorates the place. It would be interesting to see npcs that show a wide variety of upbringings such as those who know nothing of hardship (nobles) and how that upbringing changes their view point of the world and what they value. Maybe they prefer big talk over actions in regard to getting them to do something.
 

Rob

Moderator
I was thinking about different kinds of personalities randomly chosen and applied by that awesome AI the devs keep hinting on about. So yes, not whole villages, but some of the villagers while others will take their rusty sickles and scythes to fend off the attacker (i.e. you). And city dwellers will run to city guards. Or try to hide behind those barrels in that alley.
Yes, I agree that having ground-breaking advanced AI would be fantastic. I can't wait to find out what AI the devs are using!

I think that, in practice, such AI would be possible by constructing huge neural networks. The same network could apply for every single NPC. It is important to have a single framework that is both extensive and extensible, allowing all behaviours for an NPC to be automatically generated procedurally (whether during development, or in-game on-the-fly, allowing more varied game experiences).

In practice, such a network would involve an observation->response, or action->reaction cycle. For each observation (e.g. see a man with a sword), there would be a potentially-large number of potential responses (e.g. run away; attack; ignore; sit down and cry, etc.). The response would be chosen stochastically, with predefined state transition probabilities. Each transition could optionally have a pre-defined base level probability (e.g. relatively few people would sit down and cry, hopefully).

Importantly, different NPCs could automatically have vastly different "personalities" by assigning them different state transition probabilities. This assignment (or, rather, relative adjustment) would be done stochastically upon NPC generation. Of course, certain transition probabilities would be dependent on more general traits (such as bravery, cowardice, stupidity, gluttony) and thus would be highly correlated. This correlation would have to be taken into account, otherwise you'd get some interestingly bipolar NPCs! Bring on the crazies! :mad::(:confused:

Also, it would be important for state transition probabilities to be allowed to be dynamic in-game. This would allow NPCs to intelligently learn from their experiences, changing their future actions based on their past behaviour. Also, it would allow limitations on certain transitions (e.g. legs got chopped off -> can't run away!).

This would also allow various types of NPCs, e.g. humans, wolves, birds, ogres, etc. to be treated as part of the same framework. Impossible choices (e.g. humans can't fly away; wolves can't set fire to things) can simply be treated as zero-probability options. Anyway, I've rambled too much...
 

Algea

Insider
This would allow NPCs to intelligently learn from their experiences, changing their future actions based on their past behaviour.
It sounds awesome but I'm not really sure it's possible with current level of technology (I'm not tech savvy though, and I certainly didn't follow current AI trends). I think Bethesda Softworks promised something along the lines with their *revolutionary* Radiant-AI, but I'm not sure if they delivered what they had promised.

Is this kind of AI even possible? It'd blow my mind if it is, but it all sounds so futuristic that I can't believe it. Is it hard to implement?
 

Rob

Moderator
It sounds awesome but I'm not really sure it's possible with current level of technology (I'm not tech savvy though, and I certainly didn't follow current AI trends). I think Bethesda Softworks promised something along the lines with their *revolutionary* Radiant-AI, but I'm not sure if they delivered what they had promised.

Is this kind of AI even possible? It'd blow my mind if it is, but it all sounds so futuristic that I can't believe it. Is it hard to implement?
Is it possible to realistically mimic human intelligence? No, probably not.

The main problem with AI is that it usually involves interpretation of complex data, e.g.:
  1. Visual. It's incredibly hard to identify and process the relevance of individual objects from a given 2D scene. It's incredible that us humans can do this, and challenging to make robots do it.
  2. Audial. Interpretation of sound can be difficult, especially when sound-location is important. E.g. how do you tell a robot the difference between speech and environmental noises?! Of course, this is not independent of visual information, which further complicates the problem.
  3. Speech. Complex language patterns make it incredibly hard to determine the meaning of phrases, assuming you've even managed to interpret the words correctly. Especially when complex emotions / hidden meanings are present.
Even if there was an algorithm to do this well, the amount of processing required must be staggering.

Thankfully, in games this is much, much easier. All data is already stored in a meaningful interpretable format, and the variety of potential behaviours and experienced is relatively more limited than what can be experienced in the real world, no matter how much realistic detail you aim for.

It is possible to allow NPCs to learn from their decisions, if we're talking about objectives rather than detailed thought processes. Take the following example with the very simple decision tree:

Enemy approaching:
  • 50% chance of fighting;
  • 50% chance of running away.
Here, an NPC is equally likely to fight as to run away (although in practice it wouldn't be 50% - it would be randomly chosen for that NPC as discussed above). Consider the following examples:
  • NPC runs away, and survives a deadly encounter. He's glad to be alive, so decides that running away is a good idea. Add, say, 10% to his future probability of running away from an encounter, turning this probability into 55%.
  • NPC runs away, and survives, but then is ridiculed in a bar for being a coward. He then decides that next time he'll fight, and that decision probability is altered accordingly.
  • NPC fights and wins. He gains confidence, so the probability of fighting in future is increased.
  • NPC fights and loses, but lives. He loses confidence, so the probability of fighting in future is reduced.
So, basically, different encounters can alter the likelihood of taking certain actions in future.

It would take a lot of work to build a relatively comprehensive database of observations/actions and consequences/reactions, but it would certainly be possible. There would be two classes of observations and consequences, which can be dealt with in the same framework: physical (e.g. "see enemy"->"fight") or hidden (e.g. "ridiculed for being a coward"->"increase probability of fighting").

Transition probabilities can be stochastic, so little maintenance would be needed there. And the framework inherently encompasses "learning", as discussed above.

Hope that helps clarify!
 

Algea

Insider
Rob, have you ever played Bethsoft's TES series starting Oblivion? They did promise to make NPCs to react to situations accordingly and to learn from experience but I still fail to see if it's true. I mean for me what they have done is just a big pile of triggers and while it may seem like true life-emulation it's staggeringly dull after some time. I think the engine they use (it's Gamebryo if I'm not mistaken) isn't suited for this kind of AI despite all their promises.

What you describe sounds like future technologies to me, I wonder why developers haven't tried their hands at this yet. It'd be a revolution in RPG gaming (not like those boring kill another giant Skyrim quests). Or maybe there are some titles that have this kind of self-learning AI?

I actually love Skyrim and Oblivion, but I see their faults for what they are - faults. I wonder if it's because it's impossible to implement (too much time or manpower needed) or because their publisher doesn't want to repair something that is not broken in their opinion.

Anyway thanks for clarifying. When you write it it seems quite easy (well, at least not as hard as I expect it to be) to implement. And thinking about Madoc saying that they do everything to optimize the making of the game I think it's even possible to have this kind of AI in Sui Generis. If the engine allows.
 

Rob

Moderator
I think that this sort of approach isn't too complicated in theory, and would be very effective if done properly. However, what I describe is a very general abstract framework. In practice, it could get complicated as the list of potential actions increases. The effectiveness of such a scheme would really depend on the particular design and implementation.

I'm not sure exactly what Bethesda are attempting to do, but clearly it's not good enough to make us all fall to our knees! I too like Skyrim, but have noticed how repetitive the NPC behaviour is after X hours. Also, it always amuses me how enemies just stand there and take arrows if they can't properly locate/access you! Using the framework I describe above, such enemies could quickly learn that it's not a great idea to just stand there and take arrows until they die.

I'm sure that Madoc has already thought about all of this, although it would be interesting to know whether NPC AI is already done, or on the to-do list!
 

Tony

Insider
Rob, have you ever played Bethsoft's TES series starting Oblivion?
I am not Rob, obviously, but I played Oblivion and I know what you are referring to. In Oblivion Bethesda neutered the Radiant AI from their original implementation of it. They said the AI would screw up their quest system (which is why I'm glad Bare Mettle isn't going to have a generic quest system - it just wouldn't work with even semi-intelligent AI). Here's a quote from one of the devs that worked on Oblivion:

"In some cases, we the developers have had to consciously tone down the types of behavior they carry out. Again, why? Because sometimes, the AI is so goddamned smart and determined it screws up our quests! Seriously, sometimes it's gotten so weird it's like dealing with a holodeck that's gone sentient. Imagine playing The Sims, and your Sims have a penchant for murder and theft. So a lot of the time this stuff is funny, and amazing, and emergent, and it's awesome when it happens. Other times, it's so unexpected, it breaks stuff. Designers need a certain amount of control over the scenarios they create, and things can go haywire when NPCs have a mind of their own.

Funny example: In one Dark Brotherhood quest, you can meet up with this shady merchant who sells skooma. During testing, the NPC would be dead when the player got to him. Why? NPCs from the local skooma den were trying to get their fix, didn't have any skooma, and were killing the merchant to get it!"

The entire gameplay experience in Oblivion felt neutered to me :mad: I had such high expectations for that game but it was a huge let-down.
 

Tony

Insider
Let me quote myself from another forum:
I enjoyed some things about Morrowind but others drove me nuts. Like how their cities were just an exact grid. I was thinking "wtf kind of medieval city had a layout like this?". So when they hyped up Oblivion, saying how they added advanced AI and intelligent traps and all this other bullshit that never got implemented, I fell for it. Never again will I buy another Bethesda game. The devs are amateurs and have some of the worst design concepts I've seen in cRPGs.
 
I enjoyed some things about Morrowind but others drove me nuts. Like how their cities were just an exact grid. I was thinking "wtf kind of medieval city had a layout like this?". So when they hyped up Oblivion, saying how they added advanced AI and intelligent traps and all this other bullshit that never got implemented, I fell for it. Never again will I buy another Bethesda game. The devs are amateurs and have some of the worst design concepts I've seen in cRPGs.
I enjoyed Morrowind immensely. I think it was actually the first RPG I'd ever played, so it remains ideal in my mind. Oblivion was certainly a step diagonally backwards. Somethings had improved, but many seemed worse! Skyrim is a step in the right direction for sure. It's much more
like Morrowind than Oblivion, and can be improved even further with mods. I suggest checking it out. *shrug*
 
Top

Home|Games|Media|Store|Account|Forums|Contact




© Copyright 2019 Bare Mettle Entertainment Ltd. All rights reserved.