Worlds first ever RPG with 'dungeon master ai' and 'story engine' in the works

agamenn · November 2019

Developers want to create a human like AI dungeon master in the game that will react to what you do in the game, they want to create a real 'do what you want to do, go where you want to go' type game based on tabletop D&D type adventures.

kitarad · November 2019

Tagging this must watch it later.

Vrika · November 2019

I didn't watch the video because it's too long, but this won't work. AI generated content isn't close enough to handcrafted content yet.

They may be able to make another roguelike game that ends up good, but that's the extend of what todays AI tech can make.

agamenn · November 2019

Vrika said:

I didn't watch the video because it's too long, but this won't work. AI generated content isn't close enough to handcrafted content yet.

They may be able to make another roguelike game that ends up good, but that's the extend of what todays AI tech can make.

They have a very good programmer (Julian Lafey) who has been working on the engine for this for years now.

Roguelike comparison is good, but this will be a rogulike with story and persistence.

AlBQuirky · November 2019

Marked for "watch later", too. An hour and a half of Q&A intrigues me

It will be interesting to see how these folks do. The biggest thing I missed from tabletop D&D to video games was the reaction the DM had to our actions. Video games are just canned stories with few, if any branches. Too few of them "evolve" much, if at all and I always feel "forced" into following the storylines, no matter what choices I make.

blamo2000 · November 2019

A minute long trailer with gameplay would be much better than a movie length video of people talking to sell this (in my opinion).

This sounds interesting, and as long as it ticks all my boxes and doesn't do anything egregious (like have like "decks" build around a card games or any nonsense like that), I will check it out.

But like Vrika above said - I think scripting and a human putting effort into thinking out what or how a person or group could respond gives much better results in an rpg than AI. But, without devs pursuing the possibility of AI and what it can do, or pushing towards what it hopefully will do in the future, we will never get there.

Sovrath · November 2019

Vrika said:

I didn't watch the video because it's too long, but this won't work. AI generated content isn't close enough to handcrafted content yet.

They may be able to make another roguelike game that ends up good, but that's the extend of what todays AI tech can make.

hmmm maybe.

It really depends how much granularity you are are looking for. There isn't going to be "Star Trek the Next Generation" levels of Holodeck adventure building but I can imagine that there are ways this can be pulled off in a reasonable way.

Sovrath · November 2019

blamo2000 said:

But, without devs pursuing the possibility of AI and what it can do, or pushing towards what it hopefully will do in the future, we will never get there.

pretty much this. We need people to be pushing the envelope of what can be done.

Maurgrim · November 2019

They talk about it didnt see any ingame footage which is 10000000000000 times more interesting than just doing selling pitches on a 1.5 hour Q&A video.

agamenn · November 2019

Maurgrim said:

They talk about it didnt see any ingame footage which is 10000000000000 times more interesting than just doing selling pitches on a 1.5 hour Q&A video.

They are early in production, but we get an idea of what they want to make.

Mendel · November 2019

Sounds like these people are willing to take some risks. But, the fact is that presently they "want" to do this, not "have" done it. Whether or not they can accomplish a "DM AI" seems like a very powerful step forward for the genre. Fail and they learn their limits. Succeed and we have a playable game. I equate this with "conceptual game design", much like conceptual art for now. I am curious enough to watch the full video when I have time to spare.

A quick prediction. The real challenge this effort will face will be determining how to abstract user actions. Fundamentally, any game needs to know what inputs to expect and how to respond to those inputs. How do they intend to abstract an action they haven't anticipated and coded?

Amathe · November 2019

I would certainly give this a try.

Sovrath · November 2019

Mendel said:

Sounds like these people are willing to take some risks. But, the fact is that presently they "want" to do this, not "have" done it. Whether or not they can accomplish a "DM AI" seems like a very powerful step forward for the genre. Fail and they learn their limits. Succeed and we have a playable game. I equate this with "conceptual game design", much like conceptual art for now. I am curious enough to watch the full video when I have time to spare.

A quick prediction. The real challenge this effort will face will be determining how to abstract user actions. Fundamentally, any game needs to know what inputs to expect and how to respond to those inputs. How do they intend to abstract an action they haven't anticipated and coded?

I haven't had a chance to watch the whole video but I think it depends on their scope. There are definitely things they can plan for. However, in a real "live" pen and paper game, players can come up with a huge amount of variation, even things that go beyond the scope of the adventure.

It's my guess that they are talking more about "I use a fireball/I go down this hall over that hall/I ingest this potion at this time", etc.

Over players saying "I find a pickaxe and I try to burrow through the wall."

cheyane · November 2019

I always wished to play in a game like the one in the book Dream Park by Larry Niven and Steven Barnes. I really enjoyed that book and often wondered how great it would be to play something like that. I don't want the horrific things that happened but the idea definitely made me imagine things.

agamenn · November 2019

"How do they intend to abstract an action they haven't anticipated and coded?"

In an interview that Julian gave, he said he wants to put in neural network and machine learning to create the AI.

AI would essentialy take action based on what the developers made it learn, and as it is 'dungeon master' the AI would be only limited to certain things it can do.

I guess by 'dungeon mater' they mean the AI would create stories for you, basicaly quests - you go into a town and the AI creates a story for you in that town..... you go into a cave instead of a town and the AI creates a story for you in the cave.

Vrika · November 2019

agamenn said:

"How do they intend to abstract an action they haven't anticipated and coded?"

In an interview that Julian gave, he said he wants to put in neural network and machine learning to create the AI.

AI would essentialy take action based on what the developers made it learn, and as it is 'dungeon master' the AI would be only limited to certain things it can do.

I guess by 'dungeon mater' they mean the AI would create stories for you, basicaly quests - you go into a town and the AI creates a story for you in that town..... you go into a cave instead of a town and the AI creates a story for you in the cave.

The problems with that are:
-There is no way to automatically measure a story. The AI doesn't learn anything unless you've got a human who rates the story
-Machine learning is good when your aim is to have the machine arrive in same or similar solution in similar situation. For creating stories it doesn't work that well because it tends to reach solutions like "Most towers have an imprisoned princess" rather than "Having a story about princess imprisoned in a tower would be good"

Learning AI could work well for placing things like traps and monsters, where you can make automatic ways to measure how well party is doing, and the player is happy enough if the orc camp has 10 slightly different groups of orcs in it. It's not good for stories that can't be automatically measured and the player expects to be different from each other.

Mendel · November 2019

agamenn said:

"How do they intend to abstract an action they haven't anticipated and coded?"

In an interview that Julian gave, he said he wants to put in neural network and machine learning to create the AI.

AI would essentialy take action based on what the developers made it learn, and as it is 'dungeon master' the AI would be only limited to certain things it can do.

I guess by 'dungeon mater' they mean the AI would create stories for you, basicaly quests - you go into a town and the AI creates a story for you in that town..... you go into a cave instead of a town and the AI creates a story for you in the cave.

My initial point was that players are inventive, computers aren't. When faced with a chest that they can't pick, players are likely to try to take the entire chest, or burn the chest, or pour acid in the lock. What exactly is the UI icon to suggest pouring acid in the lock? How is that icon distinct from one for pouring water in a lock or pouring acid over the hinges? That's what is going to be difficult. Words can be precise and flexible, graphic icons not so much.

In order for a computer to react to an action, the player needs to first input that action. Without the ability to input a specific action, even the best computer learning machines can't learn.

Palebane · November 2019

Would be cool if they could afford to hire some top notch human DMs, imo. Imagination is key.

Sovrath · November 2019

Mendel said:

agamenn said:

"How do they intend to abstract an action they haven't anticipated and coded?"

In an interview that Julian gave, he said he wants to put in neural network and machine learning to create the AI.

AI would essentialy take action based on what the developers made it learn, and as it is 'dungeon master' the AI would be only limited to certain things it can do.

I guess by 'dungeon mater' they mean the AI would create stories for you, basicaly quests - you go into a town and the AI creates a story for you in that town..... you go into a cave instead of a town and the AI creates a story for you in the cave.

My initial point was that players are inventive, computers aren't. When faced with a chest that they can't pick, players are likely to try to take the entire chest, or burn the chest, or pour acid in the lock. What exactly is the UI icon to suggest pouring acid in the lock? How is that icon distinct from one for pouring water in a lock or pouring acid over the hinges? That's what is going to be difficult. Words can be precise and flexible, graphic icons not so much.

In order for a computer to react to an action, the player needs to first input that action. Without the ability to input a specific action, even the best computer learning machines can't learn.

But that's the thing, that's not really hard. As long as it's within the context of what the object can reasonably receive as actions.

The player comes to the chest. He "can" take it. As long as he/she can carry it. Done. Take it into town. That's not hard to do.

As far as the lock? Assign the "pick/break/acid/whatever" and then some outcomes with die rolls.

My guess is that all the actions a player might do will be there. Not hard. Now, "other things" such as fill the lock with feathers or any number of creative things, won't be there. But that's for players to temper their expectations.

Eladi · November 2019

Impossible venture atm, yes they could do some nifty things whit AI, but cant come near the way a real player could react to a situation. for that you need AI that can almost mimic the human state of mind whitin the moment a human player reacts to a situation. This a AI will never ever be able to do even if it had direct acces to all the world knowlidge and a way to near instandly process sutch data. the only way it will ever come close is when we humans are able to make a direct copy of our brain workings into a mechanical version. (brain in a jar..)

IF we would reach this point of game development in our lifetime we would also have 100% VR and eliminated death more or less.

Its all on the same scale as 42

AlBQuirky · November 2019

For those wanting a tl/dr version, I just watched this video not realizing it was about the same game.

agamenn · November 2019

Eladi said:

Impossible venture atm, yes they could do some nifty things whit AI, but cant come near the way a real player could react to a situation. for that you need AI that can almost mimic the human state of mind whitin the moment a human player reacts to a situation. This a AI will never ever be able to do even if it had direct acces to all the world knowlidge and a way to near instandly process sutch data. the only way it will ever come close is when we humans are able to make a direct copy of our brain workings into a mechanical version. (brain in a jar..)

IF we would reach this point of game development in our lifetime we would also have 100% VR and eliminated death more or less.

Its all on the same scale as 42

We have to wait and see :P

agamenn · November 2019

i.imgur.com/WCpAzoH.jpg

skeaser · November 2019

AI worked well for EQ Next and storybricks...

vegetableoil · November 2019

I'ts called Procedural generated games. ring a bell?

anemo · November 2019

vegetableoil said:

I'ts called Procedural generated games. ring a bell?

Most procedurally generated games only really have the ability to "generate" at the start of world generation. The ability to realize that it needs to increase the number of "crafting encounters" because the player is starting to run out is pretty rare. NPCs realizing that the player likes to get in on "This type of job, so we should get them in on this one", is also pretty rare for procedurally generated worlds.

The closest adaptive systems we have in gaming right now are related to adaptive matchmaking for PvP. Where the matchmaker takes any data available (skill levels, prefered class types, types of players the player has worked with well) and aims to get a targeted win rates.

So it would make sense that a dev somewhere would want to try and play with such a system brought into the PvE space.

Howdy, Stranger!

Worlds first ever RPG with 'dungeon master ai' and 'story engine' in the works

Comments

Howdy, Stranger!

Quick Links

Worlds first ever RPG with 'dungeon master ai' and 'story engine' in the works

Comments