Putting the 'role' back in role-playing games since 2002.
Donate to Codex
Good Old Games
  • Welcome to rpgcodex.net, a site dedicated to discussing computer based role-playing games in a free and open fashion. We're less strict than other forums, but please refer to the rules.

    "This message is awaiting moderator approval": All new users must pass through our moderation queue before they will be able to post normally. Until your account has "passed" your posts will only be visible to yourself (and moderators) until they are approved. Give us a week to get around to approving / deleting / ignoring your mundane opinion on crap before hassling us about it. Once you have passed the moderation period (think of it as a test), you will be able to post normally, just like all the other retards.

AI in future rpgs?

Makabb

Arcane
Shitposter Bethestard
Joined
Sep 19, 2014
Messages
11,753
One thing I think the future RPGs will have in common as AI will get better is that you will be able to have actual conversations, there will be different 'character AI' and instead of having 3-4 answers, you will have a conversation like on chat bot AI.

Basicaly it will be what 80's RPGs did, but instead of predefined answers to your input, the AI will have dynamic answers.
 

Bester

⚰️☠️⚱️
Patron
Vatnik
Joined
Sep 28, 2014
Messages
11,121
Location
USSR
Do you have coherent chat bots now? Cause you don't.

That future is distant. I'd say 20 years at least.
 

Jimmious

Arcane
Patron
Joined
May 18, 2015
Messages
5,132
Steve gets a Kidney but I don't even get a tag.
Laymen vastly overestimate how advanced AI currently is, I blame this on the (albeit very cool) stuff produced via machine learning.

What I'd like to see from AI and machine learning is an AI that adapts to the player's most commonly used combat tactics.
That's actually not far from happening. But the gaming industry is not interested into implementing such stuff, because making games too hard will drive customers away (and of course it costs a lot money)
I mean Google's AI and Musk's AI beat top level profs in LoL and DotA consistently nowadays and one of the reasons is that they can predict the most probable moves of human players
 
Last edited:
Joined
Jan 14, 2018
Messages
50,754
Codex Year of the Donut
Laymen vastly overestimate how advanced AI currently is, I blame this on the (albeit very cool) stuff produced via machine learning.

What I'd like to see from AI and machine learning is an AI that adapts to the player's most commonly used combat tactics.
That's actually not far from happening. But the gaming industry is not interested into implementing such stuff, because making games too hard will drive customers away (and of course it costs a lot money)
I mean Google's AI and Musk's AI beat top level profs in LoL and DotA consistently nowadays and one of the reasons is that they can predict the most probable moves of human players
It's also expensive and difficult. The people implementing those AIs are researchers at the cutting edge of their field.
It is well known to anyone in programming that the gaming industry pays absolute shit. Nobody capable of that talent is going to work for bottom dollar.
 

eXalted

Arcane
Joined
Dec 16, 2014
Messages
1,213
shitposter.png
 

Modron

Arcane
Joined
May 5, 2012
Messages
10,056
Yeah 10 more years on that radiant AI and bethesda will produced a good dialog system
:happytrollboy:
 

jewboy

Arbiter
Joined
Mar 13, 2012
Messages
657
Location
Oumuamua
Very unlikely to happen in your lifetime, Makabb. Best thing for AI optimists to do is talk to someone who actually works in the field of artificial intelligence research. Ask them about the problems and why the solutions are so intractable and so distant. It could quite easily be 200+ years before we see game characters who can pass a Turing test. Maybe longer. Progress in the past half century has been decent actually but mostly in areas peripheral to the main goal: speech and pattern recognition, speech synthesis, robotics, and some other narrow areas that used to be considered AI. We have made practically no progress at all toward even the very dim and basic consciousness of a snake or a grasshopper.

There was one small jump forward in neural networks which combined with the ever increasing power of GPUs has led to a little bit of progress in very basic machine learning, but it's mostly supervised learning. Even in some of the peripheral areas like robotics progress has been a lot slower than I hoped for. Where are the home robots for instance? Boston Dynamics has created some truly amazing robots like something straight out of a science fiction film, but they've all just been prototypes. Only recently has there been some chatter about actually getting something to market. I remember when AIBO was popular and people hoped or even expected progess in home robotics would proceed at a brisk pace from there, but instead Sony just stopped making them.

In terms of harnessing neural nets with the newer 'deep learning' algorithms and the powerful GPU tech we have available today (if you can afford it) to create an improvement on the 3 to 5 sentence choice 'conversations' we have with game characters now that may be somewhat possible but only in a very limited way. I am thinking of something like Tay the Hitler loving chatbot combined with some natural language programming and a neural net trained to make the right dialogue responses by example. It would be a pretty ambitious project though. Hard to imagine any game developers even giving it a try. It would likely result in a lot of ungrammatical but hopefully understandable sentences which ideally would be entirely unpredictable and different every time you talked to them. Probably wouldn't work for giving out quests, but it could make for some interesting conversations with people on the street or in taverns. Or maybe you could use it for one special character like a dragon or an orc who speaks English (or whatever) but not that well. Or you could invent a language like Ultima Underworld did with Lizardman and just use that. I don't think we will see real progress until synthetic biology researchers get more hints from the programming in DNA. We need to be able to read the instruction sets responsible real brains before we can build them ourselves..
 
Last edited:

DJOGamer PT

Arcane
Joined
Apr 8, 2015
Messages
7,517
Location
Lusitânia
One thing I think the future RPGs will have in common as AI will get better is that you will be able to have actual conversations, there will be different 'character AI' and instead of having 3-4 answers, you will have a conversation like on chat bot AI.

While the idea is cool, an NPC with dialogue written by the devs is far better than a procederal generated dialogue by an AI. Because the devs unlike the AI are the ones with the defined vision and concept of their games.

That's actually not far from happening. But the gaming industry is not interested into implementing such stuff, because making games too hard will drive customers away (and of course it costs a lot money)

While AI is one of the things that do need to improving on today's games, an AI that doesn't have a defined behavior which the player can study and plan around and further more one that can effectively prevent the player from executing any action, it's pretty much a shit AI to implement in any game.

Laymen vastly overestimate how advanced AI currently is, I blame this on the (albeit very cool) stuff produced via machine learning.

What I'd like to see from AI and machine learning is an AI that adapts to the player's most commonly used combat tactics.

Play Phanntom Pain with the Hardcore Mod.
 

JarlFrank

I like Thief THIS much
Patron
Joined
Jan 4, 2007
Messages
33,153
Location
KA.DINGIR.RA.KI
Steve gets a Kidney but I don't even get a tag.
That's actually not far from happening. But the gaming industry is not interested into implementing such stuff, because making games too hard will drive customers away (and of course it costs a lot money)

While AI is one of the things that do need to improving on today's games, an AI that doesn't have a defined behavior which the player can study and plan around and further more one that can effectively prevent the player from executing any action, it's pretty much a shit AI to implement in any game.

Ideally, I'd want an AI with a personality that has a preference for certain tactics but can also learn to predict player moves with time.

In RPGs, it's harder to determine when and how an AI learns as compared to a strategy game.

Let's take a Total War style strategy game as an example. Armies are led by generals with certain general traits. Cautious, aggressive, etc. When you encounter a nation you haven't fought before, its generals will react with their default AI profile. But the more often you fight against that nation, the more the generals of that nation will learn about your preferred tactics and unit compositions and will adapt to counter you. So if you keep trying to do the Alexander the Great move of pinning the enemy's main line with phalanx and then flanking with heavy cavalry, the enemy will prepare moves to counter this particular tactic because he expects it from you.

But if you encounter a different nation that never faced you before, their AI generals won't specifically counter that tactic because they don't yet know what you're likely to do.

Something like that could work in RPGs, too. Maybe make it faction based.

And AIs don't need to be perfect and perform the best possible move at the best possible time all the time. They can - and should - have flaws so it doesn't become frustrating. All I want AIs to do is to know the strengths and weaknesses of their units, as well as the strengths and weaknesses of the player's units, and make use of basic tactical maneuvers such as flanking and encirclement.

An encounter against 4 goblins and 5 orcs is going to be much more exciting when the goblins go for a skirmishing maneuver against your flanks with throwing weapons, while the orcish warriors try to pin you in place with melee attacks, rather than if they all just bumrush you.
 

jewboy

Arbiter
Joined
Mar 13, 2012
Messages
657
Location
Oumuamua
Proceedurally generated responses are not the only option besides canned but well written responses. You could try to train a neural net to make responses that are more or less what the developer is looking for but with built in unpredictability. You would never really know what it was going to say exactly, but you might be able to train it to say like 10 different sorts of things depending on what the player says. Maybe you would still have to have the player use canned responses so that the neural net can be trained with them. That would be a lot easier. At least you would never know exactly what the game character was going to say no matter how many times you've played the game. Just imagine if every time you talked to Firkraag he went on about all kinds of crazy stuff and it was never the same thing twice. Haha. I would love that. Even if he started ranting about Hitler's pajamas. This would be really really hard to program though. It would be a major accomplishment and a big AI project.
 

Jimmious

Arcane
Patron
Joined
May 18, 2015
Messages
5,132
Steve gets a Kidney but I don't even get a tag.
Yeah the hardest part is to implement an AI that is fun to play with. That means a combination of the following:

  • Can play the game well and provide a challenge
  • Can make mistakes occasionally
  • Said mistakes must feel natural and not awkward (eg enemies stop in the middle of combat randomly)
  • Must fit the "concept" of the game. If it's not a highly competitive game, if an AI just plays the optimal way to win, it can lead to it playing with all the tricks in the book which will break the game. That's not what you want. Imagine playing BG and the enemy parties would use all the tricks the player can use, eg pre-buff to eternity, throw clouds from the edge of the fog of war etc. Wouldn't be a lot of fun for the vast majority of players
 

Drowed

Arcane
Joined
Dec 28, 2011
Messages
1,679
Location
Core City
Although this is something that sounds incredible in theory, I imagine that in practice this will hardly happen. And I'm not getting into questions here about the advancement of technology, but simply for design purposes.

A point that everyone needs to note is that the "AI" inside the games has hardly advanced since the end of the 80s, the behavior of NPCs and enemies still follow very basic scripts that are little more than a flowchart. If X, then Y. That's it. And that's not because there are no more efficient or intelligent ways to deal with that problem, but because one of the things that game developers are most concerned with when making a game is precisely its predictability. How will the game world react when the player does X? Or if the player does Y? What if he does Z while charging J during event W? The more variables you put into a situation, the harder it is for designers to analyse how the game will react, and therefore the harder it is for them to create an experience the way they want it to be.

It's not only a matter of bugs, but also of being able to determine the difficulty of certain challenges, to balance the game, to determine solutions. And ironically, the average person tends to think that an AI is smarter when it's doing things that are actually stupid. For example, games where you have AIs capable of "listening", they'll check out a noise you've created, but in the most stupid way possible - basically, they usually go straight to the place, and stop there for a while, making a point of talking to themselves about things like "it must have been my impression". That's not smart. But it creates the impression for most players that the NPC has intelligence, and more importantly, it creates a mechanic that players can use to interact with the world.

These mechanics are quite obvious, predictable and evident just so that the player is able to accurately predict how enemies will react. If the enemy is really smart, he will probably react in a difficult way to predict, and often the player will think that the NPCs are actually cheating. I can't remember which interview I read this, but there was some FPS game that decided to implement a more advanced AI, where the enemies really acted in a group, surrounded the player and looked for more intelligent positions to attack and escape. The result was that most of the players hated the game, and accused the enemies of "coming out of nowhere". The result was that the developers simply tossed all the AI in the trash and went back to the standard behavior and... Surprise! Now the players liked the game.

If all these problems appear only in the way enemies act, imagine then when we talk about conversations?

You will never be completely sure what kind of interaction the player will have with certain characters. Each player can have a completely different conversation, which sounds great on paper, but it's a nightmare from a design point of view. Suddenly, you no longer have any idea if an X player will discover the Y information in a certain place. Because the conversation he had (or didn't have) is just not going to be something "guaranteed" anymore. That way, it's practically impossible to create a coherent story. And worse, you won't even be sure how the AI of certain NPCs could interact with other NPCs - what kind of situation could occur? You will simply create a situation where it is virtually impossible to create a story in the traditional sense, with beginning, middle and end.

I believe that in the future, it is likely that an AI will indeed be implemented to help game creators, but it will not be something present in all RPGs and/or games. Possibly, some open world RPGs, roguelikes, or other open-ended games may try to use things like that. But most traditional RPGs will keep the dialogs pre-determined, perhaps using AIs to create generic NPC dialogs in a procedural way to simplify the work of producers, but not much more than that.
 

Shadenuat

Arcane
Joined
Dec 9, 2011
Messages
11,969
Location
Russia
RPGs don't need learning AI or any super AI. All they need is deep enough character development, reasonable scripts for enemies (like enemies prioritizing debuffed/paralyzed characters and such) and good enemy variety (some enemy spellcasters preferring summoning, others cc, others damage/kiting).

Regardless if enemy AI will make dumb mistakes or not it will still make for a fun game. Moment player finds a way to break one enemy pattern you can just throw new enemies with new tricks or mix them in a unfair combo (beholder + lich + vampire + high level wizard).

An enemy can easily even be a one trick pony (like in blobbers; or fukken blinding wasps in wiz8) and its only script can be attack_nearest, but if its deadly enough you will still have fun fighting them.

An encounter against 4 goblins and 5 orcs is going to be much more exciting when the goblins go for a skirmishing maneuver against your flanks with throwing weapons, while the orcish warriors try to pin you in place with melee attacks
You don't need AI for this though. Simply putting some ranged enemies on the flank behind cover and melee enemies in front does half the job. If you give said goblins poison arrows which drain dex/movement or something you are now a Grandmaster of Enemy Design. They might as well have auto_attack on them at that point. For example in Kingmaker there is a fight when kobolds ambush you - pack of ranged ones + caster are positioned behind the river (which is difficult terrain and slows you), and melee ones rush you (and are Hasted by caster). They're all probably dumb as fuck and have attack_nearest.dat but it is an already a massive upgrade over same melees and ranged ones just standing in one single blob in random area because designer just had a checklist of encounters for location to do.

It just takes some effort that's all. I'd say AI is highly overrated by players; while encounter design especially using rpg systems to their fullest is very underrated.
 
Last edited:

newtmonkey

Arcane
Joined
Aug 22, 2013
Messages
1,726
Location
Goblin Lair
AI, as in the type of AI being talked up by hipster CEOs, is never going to be brought to games to make a game tactically more interesting, I can guarantee that.

AI will NEVER BE USED for anything fun. The only reason anyone spends the time/effort to introduce AI, the kind of AI being discussed here, is to save $$$. Pathfinder 2 isn't gonna be hooked into a neural net to give you smarter orcs.

Games don't even need this technology. The reason games don't have better AI is 1) it's not worth spending the effort/time/money because no one will appreciate it and 2) a harder game is more likely to be given a lower score by reviewers who just want to get every game over with ASAP so they can post cat videos and/or how much they hates gamers on their social media accounts.

If we are talking about using AI to generate intelligent dialog, good luck with that. No one is gonna spend the $$$ on using AI for that, when you could have an intern or whoever sit down for a week and write all the random dialog you'd ever want (and it would be better, unless the intern was retarded). Or just use a system like Arena/Daggerfall. The stumbling block here is voice acting; the same thing that resulted in RPGs with heavy dialog going off-course after Planescape or even the Elder Scrolls and Ultima series.
 
Last edited:

Saerain

Augur
Patron
Joined
May 27, 2011
Messages
495
Do you have coherent chat bots now? Cause you don't.

That future is distant. I'd say 20 years at least.

Yes, we do? It is time-consuming, but they're more than coherent enough for something as narrow as a specific in-game NPC for over a decade now. Give me a premise, I'll go make one.
 

Vault Dweller

Commissar, Red Star Studio
Developer
Joined
Jan 7, 2003
Messages
28,035
Vault Dweller what's your take on gaming AI? I'm interested
Good AI will always beat a new player because the AI has been programmed to use the character/combat system with max efficiency.

AI: if regular_attack avg damage is less than enemy's DR and THC_power >= 60%, use power attack, otherwise...
New player: I click on enemies but do 0 damage, yet the enemy always deals damage, the game must be broken

^ this is probably the main reason why won't good AI anytime soon (and no, I'm not implying that AoD had good AI). What my experience taught me is that most players don't want challenging combat. They want to kick ass and feel good about it, which is why Hard is the new Normal now. Until the situation changes, the demand for good AI will remain non-existent and supply will reflect that.
 

Serious_Business

Best Poster on the Codex
Joined
Aug 21, 2007
Messages
3,911
Location
Frown Town
A point that everyone needs to note is that the "AI" inside the games has hardly advanced since the end of the 80s, the behavior of NPCs and enemies still follow very basic scripts that are little more than a flowchart. If X, then Y. That's it.

I do assume that in terms of "action games", this is not exactly what's happening. They have to insert an element of randomness that will make the patterns unpredictable ; this is enough to make the "AI" more complex in itself, I would say, to a certain degree. You can get micro-scenarios where there are more variable involved, more agents and sources of input. The advancement of AI has certainly not happened in terms of dialog trees, but elsewhere, in things like world generation and general gameplay which is not based on scripts but environment, object-based programming ; it's also happening in graphic generation, I would assume. Everything that is engine-based could be considered to implement a kind of "AI", it' doesn't concern only the direct interactions the player has with the game. But, saying all this, it's true that rpgs in themselves don't seem to be the terrain for AI advancement. The use of something like turn-based combat, for example, is very much more simple to handle for designers than actual real-time combat ; but the efforts that designers do doesn't always translate to advanced complexity for the players.
 

JarlFrank

I like Thief THIS much
Patron
Joined
Jan 4, 2007
Messages
33,153
Location
KA.DINGIR.RA.KI
Steve gets a Kidney but I don't even get a tag.
An encounter against 4 goblins and 5 orcs is going to be much more exciting when the goblins go for a skirmishing maneuver against your flanks with throwing weapons, while the orcish warriors try to pin you in place with melee attacks
You don't need AI for this though. Simply putting some ranged enemies on the flank behind cover and melee enemies in front does half the job.

What if I want them to maneuver on their own rather than just staying where I placed them?

Hand-placing the enemies in interesting positions is one thing. But if they're able to assume such positions of their own initiative, the gameplay is vastly improved.

Sure you can also do that with scripts and current AI capabilities easily. But still.
 

Brancaleone

Prophet
Joined
Apr 28, 2015
Messages
1,005
Location
Norcia
It's also expensive and difficult. The people implementing those AIs are researchers at the cutting edge of their field.
It is well known to anyone in programming that the gaming industry pays absolute shit. Nobody capable of that talent is going to work for bottom dollar.

I'd happily settle for made-for-0-dollars modder-level AI.
 
Joined
Jan 14, 2018
Messages
50,754
Codex Year of the Donut
It's also expensive and difficult. The people implementing those AIs are researchers at the cutting edge of their field.
It is well known to anyone in programming that the gaming industry pays absolute shit. Nobody capable of that talent is going to work for bottom dollar.

I'd happily settle for made-for-0-dollars modder-level AI.
Look into the author of the mods, many of them are being produced by researchers at universities under government grants($$$).
 

As an Amazon Associate, rpgcodex.net earns from qualifying purchases.
Back
Top Bottom