Monday, 10 October 2022

AI writing partnerships


The topic of using AI/Machine Learning in TTRPGs just won't go away. Most recently, I see that Monsters and Manuals has declared for the Luddite side. I can certainly appreciate this position from an ethical standpoint, and from an aesthetic standpoint I can see why a person would think that computer-created content lacks soul (and I agree, given the current state of the art, though I think that a computer will learn to mimic soulful words and pictures eventually. The question here is how long "eventually" takes to arrive). 

In the meantime there is a way in which AIs can be used both ethically and soulfully, particularly in the field of writing. And that is by putting a human in collaboration with an AI. Human-to-human  collaborations like this already exist - think of the classic comedy writing duo, or the sitcom "writers' room". Even many RPGs and RPG supplements are created by teams - I'm thinking in particular of Tracy & Laura Hickman, who really shook up the D&D world when they wrote the Desert of Desolation adventure series (and went on to write Ravenloft and then Dragonlace, both of which had an even bigger impact).

So, how about a writing duo, one half of which is a computer? It's fairly easy to picture how this would work: the human writer writes some stuff, feeds it into the computer, the AI riffs on the writer's ideas and style, throwing out new ideas which the human can then either incorporate into the original text, or finesse and publish as an entirely new piece of work.

In fact, this is already happening. My friend Charlene Putney who writes scripts for computer RPGs (including games such as the Divinity series and Baldur’s Gate 3) is one of the people behind a piece of software (currently in beta-testing) called LAIKA, which is intended to do just this. Once you have trained LAIKA, it will make suggestions as you write, throwing up options you may not have thought of, and helping you to break out of writer's block. You can also use pre-trained models in LAIKA if you would like to, say, write like Dostoevsky or Dickens, although that's not something that appeals to me at all (but it might help if you were writing a Dostoevskyan or Dickensian RPG). I have had a quick play with LAIKA, but don't really have enough of my own writing on which to train it. I'm working on that though...

I just started thinking again about this collaborative approach to AI because I revisited my earlier, surprisingly popular blog post "Generating random content for RPGs using Artificial Intelligence" and was impressed at rediscovering the surreal creativity of the prompts I wrote (at the very end of the post) in order to provoke the AI to also be surreal and creative. I don't think that the AI-produced random encounters are as good as the ones which I wrote (please, allow me a moment's immodesty), but the computer also produced some gold dust: I would never in a million years have come up with the idea of a "magical amulet, which will allow the wearer to control any sand-dune within 100 miles" and which an evil wizard is using to slowly bury a village in sand. I could very easily run with just that concept alone, spinning it out into a fully fledged adventure, without any further assistance from the computer.

My project Nanodeities keeps getting kicked further and further into the future, but when I eventually get around to it (and I really do hope to do so) I shall certainly be collaborating with computers, probably in several ways including using both machine learning and procedurally-generated content. In fact, part of the conceit behind Nanodeities is that any tiny sentient entities which are already lurking inside my computer hardware and software (and I'm pretty certain such beings do exist, right?) will be gently pushing the content in whichever direction they see fit.

By the way, I am in the process of writing a book "King Arthur vs Devil Kitteh", and if you could visit the Kickstarter page and sign up to be notified on launch, that would help me an awful lot, thanks! 


3 comments:

  1. Ya once again I'm mostly with you on this; I think most people are talking about it in this way that's more like a linear projection of "as things are", and not recognizing that this could alternatively just open up new ways of doing things, such as using it as a tool in tandem with other methods; or new skillsets and even careers open up to facilitate the use of these tools.

    I understand some of the ethical concerns and some of them are pretty unambiguous, but I do think some of them are often not fully considered. For instance, any artist or writer is going to take inspiration from the things around them, and copyright and IP and all that is obviously quite complicated, but we all accept implicitly that there is a degree to which one inevitably is inspired by the things around them or the things they consume. So I agree that a ML model is not the same as a person by a longshot, but it is worth considering what it means for a model to train on data, and whether that should require any more permission to influence a model than you or I seeing a thing and finding inspiration in it, so long as the end result is something sufficiently distinct. Again, not trying to make a strong claim here, just saying that it is worth considering what this means and what the legal standing around that should be.

    I dunno, as a machine learning engineer myself, I do have plenty of thoughts, but I'm also not really looking to get into arguments with people about this, especially if they don't know the first thing from a linear model to classification to clustering and so on, and I've probably already said too much, but anyway, thank you for sharing this.

    ReplyDelete
  2. I saw that post as well. I understand the rejection of the tech, but I haven't yet made my own mind up. I haven't played around with AI writing, but with regards to visual art, my instincts are really something like -"if it helps me instantiate what's in my head, I'll use it." I have a feeling I may be a little less phobic about this stuff because of my musical background, which has a long history of machines providing at least part of the basic makeup of the art itself; yes it is human directed, but so are most of the art programs I have played with!
    That said, for whatever reasons I have misgivings about labeling such a creation "mine." I do think that transparency is needed though I haven't fully determined what that looks like yet, if it is just the program, or if it's the program and prompt or other items that consciously went into the ultimate output or what!

    ReplyDelete
    Replies
    1. Agreed; caution or misgivings and transparency are all probably for the best, but to just outright reject it or fail to consider it in either direction just seems really counterproductive. It's here, and if you ignore it, others will do with it as they please, so better to actually grapple with it in a real and pro-active way.

      Delete