[FINISHED] DOT Fuzzy Logic AI Engine Pre-Alpha (Open Source)

[anonymous_user_b7c8cba0]

@Mugluck;206166

Hey. This is really pretty interesting stuff. I’ve always been a huge fan of procedural stuff but pretty much everyone I know is terrified of the idea for things like people. It looks like you guys are implementing a whole bunch of stuff I’ve been considering for quite some time but being a designer rather than a coder really puts a damper on making stuff happen, as it were. Nonetheless, most of what you’re discussing makes sense. I do have a few questions though:

I realise the discriptions are very much simplified, but exactly how nuanced and complex is the system? Basically how many individual variables actually interact in any given system. For instance you might have a character, and that character is defined by say ten variables defining their goals, emotions, ambitions and so on.

Using stochastic linguistics is quite a clever idea, but exactly how much does it modify itself according to dialect inflection and individual nuance. Essentially, how much gibberish does it produce or not produce?

How do you handle offscreen events? So let’s say there’s a city down the road that’s important to the current city you’re in, what kind of simplification do you do to handle that cities events?

Why single story? I’m poking at the limitations you’re probably running on here. But why do the events necessary need to be set up prior to the game start, rather than setting initial states and allowing cause and effect to take care of the rest? Or heck, in the case of a game with a linearish storyline, have you considered magnetic inevitability? That is to say, rather than forcing the player into the rut, move the rut under the player. Or, another way to put it, the player feels like they’re at the centre of the story, and they go haring off ala Skyrim into the wilderness, the game notes this, but also notes that other things are more important, and so it essentially directs the events of the players next actions to put them back onto their designated path. I suppose an example of that would be where the player is supposed to slay the evil king, but they decide they prefer hunting trolls and completing all the guild quests and basically ignoring the plotline altogether. The game decides this is bad, so it puts a motivation flag up, and sends a mercenary to ‘motivate’ the player, Morrowind Tribunal style. The player, being a player, ignores this. So the game puts another motivation flag on the player. This time they run into a group of knights from the king sacking the village. They kill the knights, and continue on their merry way. Third motivation flag up. The king declares the player a fugitive for killing his knights. The player is now unable to interact easily with law abiding citizens without being called for arrest. They still ignore the king, so fourth motivation flag up. This time a companion is sent to join the player, their mission being to bring the player to the king. Because nothing works better than a damsel in distress - Actually no that’s not right. No actually they steal the player’s equipment and goods and leave a note saying they liked the player so they couldn’t kill them, please see them in the kings palace, or to that effect, because nothing motivates a player more than getting their sweet loot back.

So that’s the Deus Ex Machina version of a motivation. A more complex one would make each action the player takes drive them towards their goal. Which begs the question. What motivates a player? Aannd too much talking. Nonetheless, more questions, but I’m interested in what your answer would be.

First off for complexity, that’s entirely dependent on the designer. Most test scenarios we’ve done so far require 2K - 3K different variables interacting within a given system, and an individual entity’s single decision rely on about 30 - 40 different variables. A lot of this is entirely automated however. The designer would really only need to create about 10 or so different variables, and DOT will manage the rest.

We have very little idea on stochastic linguistics We’ve created prototype projects that produced very interesting results from a predefined training set, but we haven’t done the full implementation of it yet. I would imagine that 90% of it would be very well managed, as dialogue is only created in accordance to the Dialogue Games paper that was published back in the 70s. I don’t think there would be much gibberish. It would start at an okay, and readable level and only get better from there.

I don’t recall ever saying single story. You define factions, and give them a set of plotpoints they need to hit. Each faction can have 256 different plot points, and it chooses the one that interests it the most. The world is started fairly far into the game since a lot of the plot generation is done as deep learning optimization problems through a blackboards system and a multiclass SVM sytstem for constructing the network itself. Every lower node within a blackboard is an instance of a faction specified by the SVM system and the requirements of the node above it. (No we won’t be using UE4’s blackboard. Yes its because we feel they didn’t do a good job optimizing it for our particular intentions.) If we didn’t start it fairly far into the world’s history, none of the network would be optimized and the player would be waiting for a really long time for anything interesting to happen. DOT automates when the world is ready for use by calculating a bayesian efficiency value over the entirety of its tree. When its greater than a value that’s predefined by the designers, its ready. We typically use 0.7. That implies that the SVM tree has 70% efficiency. In turn this translates to, if the king asks for iron, he will have iron delivered to him 70% as efficient as possible. Therefore, it takes 43% longer than the absolute minimum time it will take to deliver iron. The system will continue to optimize this and redraw sections of the network as time progresses.

As I said every faction can have a set of plot points. There is magnetic inevitability to the plot point that the upper most node selects.

So if a king wants to invade a country, the plot will continuously get pulled towards that as the tree attempts to optimize for it. (Of course pulling the player along for the ride as well); however, if that plot point is finished, the king will select a plot point that would require the closest tree to the one that it already has.

I’m not sure if what I said made sense. If it didn’t, let me know.