Using AI to create art or texts is all the rage right now. But it’s also a highly controversial topic. These AI models are trained using existing art which more often than not is copyrighted and was used without permission. Creatives of all kinds fear losing their livelihoods to AIs trained by their creations. While I find the technology highly fascinating it’s also pretty problematic and might again shift power away from working people to large corporations.
But it’s hard not to quell the urge to experiment with AI and find out its limitations. As far as I understand it, these “artificial intelligences” can basically only riff on existing material. This becomes quite apparent when you try to let ChatGPT design game mechanics for you. I have played around with text-to-image AI and had very mixed results. Most of the time the images created looked pretty terrible. I have no idea how some people manage to create really awesome looking things with DALL-E or Midjourney.
Inspired by Michael Shorten’s post about using ChatGPT for creating hexmap descriptions, I wanted to try something RPG-related as well and started by asking ChatGPT to design a core mechanic for a roleplaying game.
Design a core mechanic for a tabletop roleplaying game using various dice including a d4, d6, d8, d10 and d12.
A core mechanic for a tabletop roleplaying game using various dice could be based on a combination of skill checks and attribute tests. Here’s how it might work:
- Character Creation: Each player creates a character with six attributes: Strength, Dexterity, Constitution, Intelligence, Wisdom, and Charisma. Each attribute is assigned a die type, based on the character’s strengths and weaknesses:
- Strength: d12
- Dexterity: d10
- Constitution: d8
- Intelligence: d6
- Wisdom: d6
- Charisma: d4
- Skill Checks: Whenever a character wants to attempt a task that has a chance of failure, they make a skill check. The player rolls the appropriate attribute die and adds any relevant skill bonuses. If the result equals or exceeds a target number set by the Game Master (GM), the task is successful. The target number is determined by the difficulty of the task and the character’s relevant skills.
- Attribute Tests: When a character encounters an environmental or physical challenge that tests their inherent abilities, an attribute test is performed. The player rolls the appropriate attribute die and adds any relevant bonuses or penalties. If the result equals or exceeds a target number set by the GM, the character succeeds.
- Combat: Combat is resolved through a combination of skill checks and attribute tests. When a character attacks an opponent, they make a skill check with a weapon. When an opponent tries to dodge an attack, they make a Dexterity attribute test. When a character tries to resist damage, they make a Constitution attribute test.
By combining these mechanics, the game allows players to use their characters’ strengths and weaknesses to determine the outcome of a wide range of actions, from athletic tasks and puzzles to combat and social interactions.
Ok, that’s not too bad actually. It’s probably not surprising that it uses D&D’s set of six attributes since D&D is the most well-known roleplaying game out there. The rest of the text sounds pretty much like every description of core mechanics out there. Again, no surprise there, since this is just the way people generally describe TTRPG mechanics.
In my next attempt I gave ChatGPT more specific instructions:
Design a simple tabletop roleplaying game using pools of six sided dice, three attributes, six skills and nine talents which grant the characters additional abilities.
Here’s a basic design for a tabletop roleplaying game using six-sided dice pools:
- Strength (STR)
- Dexterity (DEX)
- Intelligence (INT)
- Athletics (STR)
- Acrobatics (DEX)
- Stealth (DEX)
- Perception (INT)
- Persuasion (INT)
- Lore (INT)
- Dual Wield (DEX)
- Great Fortitude (STR)
- Evasion (DEX)
- Inspiring Presence (INT)
- Quick Thinker (INT)
- Keen Eye (INT)
- Poisoner (INT)
- Trapfinder (INT)
- Survivalist (STR)
- Character Creation: Players roll 3d6 to determine their starting scores for each of the three attributes. They then assign scores to their skills based on their attribute scores (e.g. Athletics is based on STR, so a character with a high STR score would allocate more dice to Athletics). Finally, players choose three talents for their character.
- Task Resolution: When a character attempts a task (such as climbing a wall or negotiating a deal), they roll a number of dice equal to their skill or attribute score and add any bonuses from their talents. The GM sets a difficulty rating for the task (e.g. 4), and the player must roll equal to or higher than the difficulty to succeed.
- Combat: In combat, players roll their attribute or skill score to attack and defend, with bonuses from their talents potentially coming into play. If a player rolls equal to or higher than their opponent’s defense roll, they hit and deal damage equal to their Strength score.
And that’s it! This simple system provides a solid foundation for a tabletop roleplaying game, and the dice pool mechanic allows for a range of possible outcomes and adds an element of luck to the game. Of course, there are many ways to expand and refine this basic design, but this should give you a good starting point!
Again it’s easily noticeable that there’s some D&D influence here. Some of the talents are named after D&D feats (possible even all of them). The skill list is also clearly D&D-inspired but the choice of skills is actually pretty good. When you have a closer look at gameplay things fall apart pretty quickly. It’s pretty unclear how many points the player can allocate to their skills. With attributes in the three to eighteen range, dice pools become pretty big and a bit unwieldy unless you want to reinvent Shadowrun. The task resolution mechanic doesn’t make that much sense under the circumstances especially with the example difficulty rating of 4.
It’s obvious that ChatGPT is quite good at creating texts that look pretty good on a first look, but under closer scrutiny you realize that it has no clue of what its writing about. At this point I don’t think game designers will be replaced by AI any time soon. But there’s some potential here. You could have it pick skill lists for you if you can’t make up your mind, or ask it to name some attributes for you and come up with fitting descriptions. In time it could help game designers with some of the more tedious stuff.
On the other hand there’s a risk. ChatGPT learns and the more we use it, the better it potentially gets. While we are using it, it’s using us to get better. At least that’s how I understand how this works. The feedback we provide improves future results. It’s not totally inconceivable that at some point ChatGPT could create a working RPG from scratch or rather create a close simulacrum of an existing one. I doubt it will be very good or include creative ideas since it’s just a rehash of existing ones, but a lot of human-made games out there are not much different. My own Warrior, Rogue & Mage is just a remix of ideas which existed for a very long time.
ChatGPT will probably not totally revolutionize the RPG industry, but I am sure it (or other AI systems like it) will have an impact. By the way, I tried using DALL-E to create a fitting featured image for this post, but the results were – to put it mildly – crap. So I relied on something created by a human.
Interesting stuff! I’ve played with chatGPT for writing code and I think it does better when there is a structure to adhere to, so I’m not surprised that it seems OK at writing game stuff.
I would never have thought of this, but I have been toying with giving it a spark table to work from. The complexity of the prompts in the linked post is pretty intriguing, however…