Skip to main content

I brought ChatGPT to the board game world. Is it ready for game night?

We all know that ChatGPT is great at speeding up mundane tasks. What could be drier than explaining the rules of a complicated game at board game night?

There’s no substitute for just knowing the game, but being able to reach for AI instead of the rulebook could make things a whole lot easier. Nothing derails a great game of Twilight Imperium like breaking out the Living Rules and endlessly scrolling. So, if I were to bring ChatGPT to board game night, I could definitely see it coming in handy. But before I subjected my friends to a robot reading them the rules, I decided to test it out with some basic questions to see if it was up to snuff.

A defined ruleset sounds ideal

I have no idea why ChatGPT knows the rules of so many games, but it does. Or at least thinks it does. While it might be tricky to find an online manual for some of my collection, ChatGPT seems to have it all — some of those billions of data points it was trained on reportedly included the errata for something as obscure as the third Battlestar Galactica expansion.

That’s great, though, because it should know all the rules I don’t, right? Within the limited, constrained, and very particular environment of a board game, ChatGPT with its limited understanding of anything, but extensive knowledge of certain topics, should be great at it.

Unfortunately, as with everything else ChatGPT confidently posits to know, it’s often not quite right, and sometimes it’s outright wrong.

ChatGPT tries to answer a question on board gaming.
Image used with permission by copyright holder

The initial answer is quite correct: you don’t add the dice to the hunt pool. But hunt tiles aren’t added every turn. Maybe it’d be better if I had ChatGPT tell me where in the rulebook I can find this information?

Asking ChatGPT more board game questions.
Image used with permission by copyright holder

Hmm. Apparently putting it on the spot means it then corrects itself and gets the rule more wrong than it did before. It thinks there might be multiple “Gandalfs” in the Fellowship, and that there are special “Will of the West,” dice, rather than that being one of the possible results on the game’s action dice.

It then goes on double down on that error by citing a page in the rulebook that doesn’t have anything to do with “The Hunt.” There’s a section called “The Hunt For the Ring,” but it’s not until Page 40.

War of the Ring rulebook.
Image used with permission by copyright holder

But maybe this isn’t ChatGPT’s best game. Let’s give it one more chance to help with a game that’s somehow even bigger and more complicated than War of the Ring: Twilight Imperium.

ChatGPT answering board game queries.
Image used with permission by copyright holder

Here, ChatGPT does an admirable job, in that it does get the answer right, but for the wrong reasons. You can’t take a home system because you can’t invade it, not because you can’t move ships there.

If that seems pedantic, I get it. I don’t like telling my friend they can’t do something because they’ve misunderstood the very specifically worded text on the card they’ve played. These details matter in games, and if I’m going to get ChatGPT to do it for me, I need to be able to fully trust it.

It’s back to reading the rulebooks over and over

This was just a snippet of my time quizzing ChatGPT on how to play my favorite games. It knew how to launch ships in Battlestar Galactica, even if it wasn’t clear what part of your turn you do it in. It had a good idea of how to get cave tokens in Quest for El Dorado, but was very wrong on the cost you had to pay for them.

It did know Kingdom Death: Monster quite well, though, accurately reporting the stats of some of the monsters, and even making suggestions on how to modify those stats to my advantage.

It was a fun exercise seeing what ChatGPT knows about games, and it feels like one area where in the future, it could be invaluable. It wouldn’t even need to know all games. I can imagine a scenario where game publishers could have their own AI to help teach you their games, and I wouldn’t be surprised if it could act as a stand-in player one day too.

And who knows, maybe GPT-4 in ChatGPT Plus would have already solved this problem.

For now though, since I can’t trust it, it’s back to reading rulebooks on the toilet so that when one of the players has a question, I can answer it. Because ChatGPT can’t. Yet.

Editors' Recommendations

Jon Martindale
Jon Martindale is the Evergreen Coordinator for Computing, overseeing a team of writers addressing all the latest how to…
OpenAI needs just 15 seconds of audio for its AI to clone a voice
A laptop screen shows the home page for ChatGPT, OpenAI's artificial intelligence chatbot.

In recent years, the listening time required by a piece of AI to clone someone’s voice has been getting shorter and shorter.

It used to be minutes, now it’s just seconds.

Read more
How much does an AI supercomputer cost? Try $100 billion
A Microsoft datacenter.

It looks like OpenAI's ChatGPT and Sora, among other projects, are about to get a lot more juice. According to a new report shared by The Information, Microsoft and OpenAI are working on a new data center project, one part of which will be a massive AI supercomputer dubbed "Stargate." Microsoft is said to be footing the bill, and the cost is astronomical as the name of the supercomputer suggests -- the whole project might cost over $100 billion.

Spending over $100 billion on anything is mind-blowing, but when put into perspective, the price truly shows just how big a venture this might be: The Information claims that the new Microsoft and OpenAI joint project might cost a whopping 100 times more than some of the largest data centers currently in operation.

Read more
Nvidia turns simple text prompts into game-ready 3D models
A colorful collage of images generated by Nvidia's LATTE3D.

Nvidia just unveiled its new generative AI model, dubbed Latte3D, during GTC 2024. Latte3D appears to be ChatGPT on extreme steroids. I's a text-to-3D model that accepts simple, short text prompts and turns them into 3D objects and animals within a second. Much faster than its older counterparts, Latte3D works like a virtual 3D printe that could come in handy for creators across many industries.

Latte3D was made to simplify the creation of 3D models for many types of creators, such as those working on video games, design projects, marketing, or even machine learning and training for robotics. In Nvidia's demo of the model, it appears super simple to use. Following a quick text prompt, the AI generates a 3D model and shortly after finishes it off with much more detail. While the end result is nowhere near as lifelike as OpenAI's Sora, it's not meant to be -- this is a way to speed up creating assets instead of having to build them from the ground up.

Read more