RPG Evolution - The AI GM: Your Somewhat Unreliable Familiar

Artificial Intelligence creates a lot of opportunities for game designers and game masters, but it has some serious flaws too.

Picture courtesy of Pixabay.​

The Basics​

One of the most popular text AIs is ChatGPT, but there are many others, particularly in the art space, like Dream.

What is commonly termed "artificial intelligence" isn't quite what is portrayed in science fiction as a thinking robot. Rather, today's AI is more a hive mind, the collective output of humanity from decades of using the Internet. AI is trained to cull this data into something human readable, be it graphics or text, in a way that feels natural. This makes AI very appealing to use and problematic to source.

Because AI is trained on data sets, the potential responses are not known even to the creators. Essentially, while companies can create AI, few can explain how they come to their conclusions because the systems aren't made to share how they work. That's of course part of their competitive advantage, as the AI race has kicked off to produce something everyone on the Internet comes to rely on.

When it comes to creating something creatively, AI can be a wonderful thought starter. It can create art uniquely to match your requirements that might not be available elsewhere; it can write an article about any esoteric topic; it can essentially try its best to do exactly what you ask it to do. But the execution is rarely perfect.

The Problems​

AI has to be trained. That is, it considers all outcomes equally valid until someone tells it otherwise. AI exposed to the Internet have quickly reverted to the Internet's worst instincts, as AI developers have yet to implement guardrails on how the tools should work. Or to put it another way, like all of Internet searches, if you put garbage in, it spits garbage out.

The problem with AI then is that it conceals how it comes to conclusions about content that wasn't reviewed by a human, may not be appropriate to the topic or audience, or otherwise is irrelevant. AI developers who use large swaths of the Internet for their data sets are taking a gamble that there will be more positive than negative outcomes.

Using AI in Games​

The possibilities of using AI to assist in tabletop gaming and publishing is enormous, but they come with some important caveats.

For one, AI is by definition using everyone else's work. This means the quality it produces can be average; in extreme cases, it can potentially violate intellectual property and copyright by copying someone else's work without modifying it enough to make it unique. This is one of the many reasons major publishers are prohibiting AI art and why DriveThruRPG has separate tags to identify when AI art has been used in a product.

Using AI also means giving up control. While some AI can be trained, the free ones are wild beasts that can only be guided in a general direction. Worse, ChatGPT creates content with 100% confidence, which can fool users into thinking their content is accurate or unique when they have no clarity on its source. I've used ChatGPT to create stat blocks and calculate challenge ratings, and it's been flat out wrong on more than one occasion.

Finally, AI takes a lot longer than the outputs you see. My personal experience is that working with free AI tools takes between 10 to 20 attempts (with art), and 3 to 5 attempts (with text) to get something useable. With art, this often means weird fingers or the wrong number of arms and legs. Even when AI gets gets close to what we want, it usually needs to be heavily edited to get it right.

Ready or Not, Here AI Comes​

Whether or not we're ready for it, the AI revolution is here. If we can use it responsibly, there is tremendous upside, but that requires training humans to use the tools responsibly too. In the upcoming series of articles, we'll discuss how game masters and publishers can use AI in creating art, text, and even game design.

 

[SakanaSensei]

Yielding creative expression out of human hands is the end of human culture. That is the terminal point of this technology. When we no longer write our own stories or compose our own songs or paint our own pictures, we won't have a culture any longer.
Quite simply, I won't support or purchase any TTRPG product that makes use of purely AI derived work as part of the project, be it written or visual media.
We all have to have a line. This is mine.

I both find the technology fascinating and find that I agree with you here. It feels like the logical terminus of everything having become “content.”

 

[Professor Murder]

I both find the technology fascinating and find that I agree with you here. It feels like the logical terminus of everything having become “content.”

I also find it fascinating, and it's really not new, but its on the cusp of being good enough to be commercially viable.

 

[doctorhook]

This is a very dangerous road and some folks are just skipping along, throwing posies.

I agree with you. Alas, as people keep saying, the genie is out of the bottle—this technology is available now, and there’s enough expertise with it that it’s impossible to remove it permanently from the market. Our choices now are either prepare for this new wave and try to surf, or we wait for the tsunami to drown us. It’s coming, and we can’t ignore it.

 

[Pauper]

Alas, as people keep saying, the genie is out of the bottle—this technology is available now, and there’s enough expertise with it that it’s impossible to remove it permanently from the market. Our choices now are either prepare for this new wave and try to surf, or we wait for the tsunami to drown us. It’s coming, and we can’t ignore it.

I'm going to disagree with this take -- there are lots of examples of 'technology' that is 'out there' but that nobody bothers with, from modern automotive automation that simply isn't being used to the formula for New Coke. At this moment, it's folly to confidently state whether technologies like ChatGPT and their underlying engines are the future, of if they're just going to be the technological equivalent of LaserDisc -- a curious stop-over while something more efficient and effective takes over.

One thing against ChatGPT, and which helps explain Talien's issues in using ChatGPT for generating stat blocks and challenge ratings, is that ChatGPT doesn't know how to do math. If you give it a math problem that's in its training library, it will probably get it right, but complex problems such as adding two large multi-digit numbers trigger a process of 'interpolation' where the program grabs what data seems relevant to the question and fills in the gaps in its understanding. With respect to addition, as noted in the linked article, this means that ChatGPT often fails to 'carry the 1' when adding large numbers. Maybe with more time the software will figure out how to do that, or more likely, the software will identify more and more cases where it needs to use a slightly different method of interpolation and will never actually 'learn' how to do basic addition.

This is why I ultimately see the refusal to use RPG products built by 'AI' (as opposed to using products designed with the assistance of 'AI'), is not as much a moral issue as a practical one: I could certainly publish a game that uses Aces and Eights's shot clock, GURPS's Quirks, and Call of Cthulhu's Sanity systems as originally published, but without doing the work to make those different mechanics work together, it's likely no one will find my game playable much less enjoyable. But this process of grabbing nuggets and shoveling in material in between the nuggets to smooth out the texture is basically all ChatGPT is doing whenever it 'creates' anything. By definition, it can never produce anything of higher quality than the material in its library.

Feel free to make use of what you can, if you're so inclined. But until the algorithm 'gets better' (or more likely is replaced by something fundamentally different that produces superior results), I don't really see any need to be an early adopter.

--
Pauper

 

[JAMUMU]

It's like watching a newborn god learning in real-time. It's fascinating, it's awesome and it's terrifying.

But eventually it'll be happy little gun-drones, singing in the sky, only shooting the people with the wrong brainwave patterns.

 

[WageMage]

Yielding creative expression out of human hands is the end of human culture. That is the terminal point of this technology. When we no longer write our own stories or compose our own songs or paint our own pictures, we won't have a culture any longer.
Quite simply, I won't support or purchase any TTRPG product that makes use of purely AI derived work as part of the project, be it written or visual media.
We all have to have a line. This is mine.

Not to mention that a few of the art AIs were trained by scraping the internet and using artists' work without consent.

A lot of these AIs would probably not be as good as they are if artists had a chance to opt out from having their work being used by AI.

 

[GrahamWills]

So ... this article is generally correct about some sections of AI. Specifically the sorts of diffusion model that generative tools such as DALL-E, Midjourney and ChatGPT use.

However there is a lot more to AI than these tools. In fact, most AI in use nowadays is not described by this article. To correct this, whenever the article uses the term "AI" replace it with "large language and diffusion based AI".

 

[firstkyne]

I agree with you. Alas, as people keep saying, the genie is out of the bottle—this technology is available now, and there’s enough expertise with it that it’s impossible to remove it permanently from the market. Our choices now are either prepare for this new wave and try to surf, or we wait for the tsunami to drown us. It’s coming, and we can’t ignore it.

I'm not anti-technology in any way; I believe that most creative developments have been progressed by the discovery of new tech, be it Bob Dylan's use of electric guitar or Adobe's enabling of personal expression. And it is a fool's errand to try and put the a tech genie back in the bottle. It has never worked.
But. As an artist working in illustration and music, AI is very concerning. Many (in fact most) people do not intrinsically value a piece of artwork that is made by a human any more than one made by AI. This will devalue the work of artists in any medium it intrudes into. I have seen this first hand already, even at this embryonic stage.
To deny our children the privilege of engaging their brains in artistic endeavour, which has many benefits (for mental wellbeing, self-expression, exploration of the ongoing human condition and social cohesion to name a few) and denying them the chance to receive validation, encouragement and the enjoyment, wonder and pain of the 'Artists's Journey' has serious implications for future generations. Children's worlds will feel the impact of AI in one way or another within just a couple of years, and the extent of the impact in the longer term is unexplored at present.
When I read about people being excited about what AI can offer to creative endeavour, frankly I assume this is someone on the periphery of the creative industry. Because of those who utilise it, the main beneficiaries will be the richest studios who can bring expensive resources into play to create thousands of iterations and swamp creative markets, outgunning small organisations and individuals. They'll do it because they will feel they have to, to compete. Because if they don't do it, one of their competitors will.
Tech scientists seem not to talk to or consider the work of their colleagues in the medical, psychology and sociology fields. I think there needs to be a global discussion about what is important for human beings to do for themselves. But of course, there won't be. Christ! Keep it light. Sorry! x

 

[Darth Solo]

The better video games have used a form of AI for decades so this is old news to me at least. I always expected it to get better and it has but I stopped playing video games a while back just because in my opinion the industry really wasn't pushing the envelope with what computers can be programmed to do.

An AI GM would be okay but what's better: a program designed by someone with limited actual GM experience or having an experienced GM? At best I'm expecting an "arcade version" of D&D and that'll be enough for a lot of people.

 

[FoolishFrost]

As a game developer, with a background in game AI:

Game AI is manufactured according to the game. In most cases (not all), we do NOT make the AI 'smarter' deliberately. To be blunt, it is pretty darn easy to make an AI utterly trounce a player with nothing more than reaction time. Add in functional tactics, and what you have is a bunch of players complaining the game is unwinnable.

What we do strive for, in the limited processing power of a game, is an illusion of intelligence. In most cases, we use semi-complex reactions that allow the entity (dinosaur, human, alien, whatever...) to seem like decisions are being made, even when it's pre-baked.

Things like hesitations. Moments where it just looks at you, then around the room. Even something as simple as adding vocal responses to your presence that trigger other character to notice you as a target.

But as far as expert system AI in a First Person Shooter? Not happening for a simple reason: Your CPU is not up to the task of adding super complex thinking, while at the same time rendering the game. I mean, we still have problems on how much resources we use for pathfinding (a simple AI made to find a path from point A to point Z).

I think the best we got so far was Warframe using recordings of other players moving through levels, and using an AI to process those player movements for use in the game for NPCs.

Keep in mind, I'm oversimplifying alot, but to sum it up, Game AI is nothing like the AIs discussed in this thread. They more like a stack of simple [If this/Then do that] options. Easily readable in code.