Friday, October 16, 2020

Our AI future and Tabletop Roleplaying


 What I hope for and think will happen, is something like this.

So you want to create a setting and populate it. 

Most folks I know generally has two or three dozen ideas going into this. It may be more but there is some limit. Once you start working beyond this limit it become a bit of a chore and not fun as a hobby. The good news is that most of time you can start small. If you keep the result of what you do for the next campaign then within a few years you will find yourself quite a bit of detail for your setting and that was fun to create.

In my experience a lot of the reason this stuff is not fun is because it is repetitive work. It fine when you describe the first handful of shopkeepers. When you try to describe the 20th often it is not as fun unless some time has passed to recharge your creativity.

A great set of random tables can overcome much of this but even they have their limit. For example Traveller does a great job of generating sectors. But even that process breaks down if you tried to use to generate an entire Imperium of 16 sectors or more.

But what if we didn't use even a great set of random tables? What if we used a trained neural network instead? What if it was setup not just randomly generate but randomly generate with the two to three dozen ideas we already had? 

You tell software or webpage take what you already thought of. Then it will generate the rest around it. In addition after it done, you can review the results and have re-generate the elements you don't like. 

Maybe the result is partially there but needs to be tweaked. So you edit it and then have it regenerate the rest of that specific element. As a bonus it would be nice to drill down to the level of individual character.

What make this possible, is something I noticed about the best random tables. That they seemly capture the author experience with the subject of the table. Whether is something specific like traps, or magic items. More general like a dungeon maze. Or expansive as an entire galaxy worth of sectors. 

While it doesn't replace our creativity, random tables allow us to extend it by using the wisdom of the author of the table. The same with the use of AI software. 

Now that I can see being very useful tool for the hobby.

4 comments:

squeen said...

Neural networks are just a series of gains (multiplications) connected in a grid. You push training data through them to "train" the network (i.e. tune the gains), which its just like fitting line-slopes (a gain) to measured x-y data.

Despite all the hype, I believe there is very little "intelligence" currently in AI.

Robert Conley said...

I concur, sophisticated tools to sure but no where near sentience as we understand it.

When they can make an autonomous machine that one can toss out the door and it survive on it own for a week then I think we will be on the path.

BeRKA said...

You should try "Talk to Transformer" or InferKit.
https://app.inferkit.com/demo

Huttan Funaila said...

Squeen, the old joke in AI is that once something can be done, it is no longer "artificial intelligence", it becomes "optical character recognition" or stuff like that.

There is a procedural generation subreddit over at reddit. Some of the people posting over there do things like town/city generation for games, or pretty stuff with graphics.