ChatGPT for roleplaying - what can it do?

It feels like over the past six months OpenAI’s various products - particularly ChatGPT - have gotten a bunch of press. ChatGPT is a large language model, which (as far as I understand) basically means you feed it a whole honking bunch of text, and it extracts information from that text and is able to then respond to questions by putting words together into something coherent. Basically, it’s faking being a functional human being, but then this is 2023, give it a break, that’s all of us.

You can go read all about its predicted impact on writing, art, school exams, creativity, and every other sphere if you want, I kind of feel 10% of the web now is articles like that. I don’t particularly feel like consolidating a paragraph of media responses to the technology, though - I want to write about things that interest me.

Like roleplaying games. Given ChatGPT can take basically any prompt we give it, and spit back something that sounds vaguely coherent, how can we use it to make roleplaying games better? After all, most roleplaying games are basically a bunch of words, right?

So in this post, I’m going to explore some common potential applications for ChatGPT when it comes to roleplaying games, both tabletop and live action. Hopefully these give you an idea of where it excels, where it falls flat, and where you could use it yourself.

The TL;DR

It seems like ChatGPT is:

• great for generating flavour, filling text, and describing things.
• OK for generating new languages, place names, that kind of thing.
• rubbish for trying to generate mechanics.

And when we consider how ChatGPT is faking being a human being, it makes sense that we get that kind of tier list out of it.

Methodology notes

Most responses were collected from ChatGPT Jan 30 version, and some from the Feb 13 version, using the company’s free research preview. I didn’t try to regenerate any responses, instead opting to refine what I had through follow-up questions. When I did that below, I’ll let you know.

Goal 1: Descriptions and characters

This is a really common thing I find myself floundering with when I write LARPs - the first five characters I’m fine with writing up, and then after that my brain gets increasingly fried and my prose increasingly sparse. Added to this, I’m a serial sufferer of “more characters than plot” syndrome - I know that this game should have around 20 players, but for the life of my I can’t work out how characters 16 through 20 fit into that plot.

Thankfully, ChatGPT has my back¹:

Write a three paragraph character sheet for a supporting character in a live action roleplaying game. The game is about a group of friends at a magical high school in England who have snuck away for a weekend of relaxation before their year-end exams. The characters should explore themes of coming of age, friendship, and rivalry.

Name: Liam Johnson

Age: 17

Appearance: Liam is a tall and lean young man with messy, curly brown hair and piercing green eyes. He has a charming smile and a confident demeanor. He’s always dressed in the latest fashion and is never caught without a stylish accessory or two.

Personality: Liam is confident, charming, and popular. He’s the life of the party and always knows how to make those around him feel comfortable. He’s also fiercely competitive and will stop at nothing to be the best in everything he does, including academics, sports, and socializing. He sees his rivalries as friendly challenges, but his competitive nature sometimes causes friction with his friends.

Backstory: Liam comes from a wealthy family and has always been given everything he wants. He’s grown up surrounded by the best of everything, from the finest education to the latest technology. However, despite his privileged upbringing, Liam is not content with just coasting through life. He wants to prove that he’s the best and will do whatever it takes to make sure everyone knows it. This drive for success has led him to become one of the top students at the magical high school and a star on the Quidditch team.

So what do we have here? A very passable outline of a character whom we could flesh out into something more coherent, adding in our own plot hooks, relationships, and the like. I mean, don’t get me wrong, as it stands it reads like a slightly-too-Marty Stu love interest in a Harry Potter fan fiction (the Quidditch bit at the end just helps remind you that these are all cobbled together from The Internet), but it’s still not bad!

One of the criticisms levelled at ChatGPT has been that since it just regurgitates information from its corpus, and its corpus is online text at large, it tends to reinforce bias and stereotype in society. And I thnk Liam here is a perfect example. Of course, one of the nice things we can do is refine things, so I’m sure you could ask ChatGPT to make Liam a bit less square.

This was actually my first attempt to do something with ChatGPT. I showed it to a friend, who was suitably impressed, but metioned that they usually like having more links in their characters. That’s totally valid, so let’s try something a bit more connected…

Goal 2: Factions and relationships

One character alone is boring. Let’s see what happens when we get ChatGPT to build a faction with relationships.

Could you please generate a faction of five characters for a Live Action Roleplaying Game? This game is a soft science fiction whose theme should loosely mirror Firefly or The Expanse. The faction is a group of representatives from nomadic traders whose culture is reminiscent of Polynesian explorers. Each character should have a paragraph or two of description, including their personality, goals, and links to the other characters in the faction.

Certainly! Here are five characters for your live action role-playing game:

Captain Kana Laula - Captain Kana is the charismatic and confident leader of the nomadic traders. With a long history of successful trading voyages, Kana has earned the respect and loyalty of her crew. Her quick thinking and strategic mind often get her and her crew out of dangerous situations. Kana has a strong sense of family and community, and her ultimate goal is to ensure the prosperity and success of her people. She has a close relationship with the faction’s engineer, Keani.

First Mate Tama Makani - Tama is Kana’s second-in-command and right-hand man. He is a skilled fighter and navigator, with a no-nonsense attitude and a loyalty to Kana that borders on fanatical. Tama has a personal vendetta against a rival trader named Kalima, who he believes has wronged Kana in the past.

Engineer Keani Nalu - Keani is the mechanical genius of the crew. Her quick thinking and innovative solutions often save the day when the ship is in danger. Keani is a strong, independent woman who is respected by her crewmates for her technical expertise. She has a close relationship with Captain Kana, who she sees as a mentor and role model.

Medic Lopaka Hoe - Lopaka is the ship’s medic, responsible for the health and wellbeing of the crew. With a gentle and nurturing personality, Lopaka is beloved by everyone on board. He has a deep connection to his Polynesian heritage and often incorporates traditional healing practices into his work. Lopaka is also the ship’s cook and is known for his delicious meals that bring the crew together.

Smuggler Pono Kaulua - Pono is the wild card of the crew. As a successful smuggler, he has a network of contacts and information that can be valuable to the traders. However, his unorthodox methods and tendency to bend the rules often put him at odds with the rest of the crew, especially Tama. Despite this, Kana trusts Pono’s instincts and allows him to operate independently. Pono has a strong sense of loyalty to the crew and will do anything to protect them.

Now we’re talking! The bot has done remarkably well to keep some internal consistency here - for example, Kana’s and Keani’s descriptions agree well on the relationship between them. The main issue I have with the faction is that everyone is very chummy with everyone else - ideally there should be a bit of bite in these relationships, to keep things spicy in play. Thankfully, this doesn’t have to be our final draft.

In addition, I like how the bot mirrors your tone. When I just ask it straight up for something, it comes back with just the goods. Add in some pleases and it seems to perk up and provide some personality.

Goal 3: Iterative development

ChatGPT lets you refine your answers by asking the bot to modify things. Here’s an example, with the big changes bolded.

How could I alter these relationships to add a bit more conflict into the faction?

Here’s a revised version of the relationships between the characters, with added conflict:

Captain Kana Laula - Captain Kana is the charismatic and confident leader of the nomadic traders. With a long history of successful trading voyages, Kana has earned the respect and loyalty of her crew. However, her quick thinking and strategic mind often put her at odds with the faction’s engineer, Keani, who has a different approach to problem-solving. Kana has a strong sense of family and community, and her ultimate goal is to ensure the prosperity and success of her people, even if it means making difficult decisions that not everyone on the crew agrees with.

First Mate Tama Makani - Tama is Kana’s second-in-command and right-hand man. He is a skilled fighter and navigator, with a no-nonsense attitude and a loyalty to Kana that borders on fanatical. However, his vendetta against rival trader Kalima often puts him at odds with Captain Kana, who believes in finding peaceful solutions to conflicts. Tama’s hot-headedness and tendency to act before thinking can also put the crew in danger.

Engineer Keani Nalu - Keani is the mechanical genius of the crew. Her quick thinking and innovative solutions often save the day when the ship is in danger. However, her independent streak and tendency to challenge Captain Kana’s authority sometimes creates friction between the two. Keani is a strong, independent woman who is respected by her crewmates for her technical expertise, but her lack of trust in authority can make it difficult for her to work effectively with others.

Medic Lopaka Hoe - Lopaka is the ship’s medic, responsible for the health and wellbeing of the crew. With a gentle and nurturing personality, Lopaka is beloved by everyone on board. However, his deep connection to his Polynesian heritage and traditional healing practices often puts him at odds with the faction’s smuggler, Pono, who sees them as outdated and superstitious. Lopaka is also the ship’s cook and is known for his delicious meals that bring the crew together, but his refusal to incorporate new and unfamiliar ingredients into his cooking can cause conflict with some members of the crew who crave variety.

Smuggler Pono Kaulua - Pono is the wild card of the crew. As a successful smuggler, he has a network of contacts and information that can be valuable to the traders. However, his unorthodox methods and tendency to bend the rules often put him at odds with the rest of the crew, especially Tama, who sees him as a liability. Despite this, Kana trusts Pono’s instincts and allows him to operate independently, but Pono’s loyalty is not always clear and he may put the crew in danger to serve his own interests.

A relatively simple request to alter the descriptions has given these characters a bunch more depth². The whole crew still feels like a coherent unit, and you can still see the ties that bind them, but you can also see how these refinements start to add the hard edges of personality and minor conflict to the faction.

And the great thing is that we can keep doing this. I’m not going to add everything because this post is already pretty long and I stil have some more examples to give, but some other things I successfully asked ChatGPT to do with this faction included:

• Making Pono into a real piece of work (although ChatGPT still ensured he had some links to the crew, handily).
• Making the relationships between the characters even more nuanced.
• Changing Tama’s role from “first mate” to something slightly less stereotypical (in this case ChatGPT made him the ship’s strategist and negotiator).
• Changing Keani’s role to something related to the faction’s cultural heritage (they ended up becoming a Navigator, “responsible for charting their course through the stars and maintaining their connection to their cultural roots”).
• Adding a love triangle (ChatGPT ended up making it surprisingly bland, unfortunately).

Goal 4: Building a culture

The above exercise got me thinking about how well ChatGPT could work in creating cultures and species for fantasy (or science-fiction) games. You can only stand so many fantasy games that are Tolkein (or D&D) with the serial numbers files off before you want to start creating your own analogues to elves, dwarves, halflings, and orcs. And you can only do so much thinking about world-building before you realise that mapping race (or, as we would call it, species) to culture monolithically is a bit reductive.

So, let’s see what happens when we give it the remit to create a fantasy species and three cultures present in that species:

Could you please generate a fantasy race? It shouldn’t mimic the tropes we see in Tolkein’s works or in Dungeons & Dragons (no elves, dwarves, or orcs). The description of this race should include physical description, variation within the race. In addition, could you generate three cultures that exist within the race? Each culture should have a few traditions, a distinct linguistic feel, and some basic relationships with the other two cultures.

Sure! Here’s a description of a new fantasy race, along with three distinct cultures within that race.

Race: Aeloria

Physical Description: Aelorians are a tall, slender race with elongated limbs and fingers. They have large, almond-shaped eyes that come in a variety of bright, iridescent colors. Their skin is thin and delicate, ranging from pale to deep blue, and is covered in fine, shimmering hairs that change color based on their mood and environment. Aelorians are known for their agility and grace, and are able to run and jump with incredible speed and precision.

Variation Within the Race: Although Aelorians all share similar physical traits, there is a great deal of variation within the race in terms of their abilities and powers. Some Aelorians are born with the ability to control the elements, while others can communicate telepathically with one another. There are also Aelorians who have the ability to see into the future, while others can manipulate time itself.

Culture 1: The Wanderers

The Wanderers are a nomadic culture that travels across the land, following the seasons and the migration patterns of the creatures they hunt. They value freedom and independence above all else, and reject the idea of settling in one place for too long. They are skilled hunters and trackers, and are known for their prowess in the wild. They have their own distinct dialect, filled with sounds and inflections borrowed from the animals they live among.

Traditions:

• The Wanderers hold an annual hunting festival, where they gather to share stories and compare their skills.
• When a Wanderer reaches adulthood, they must undertake a rite of passage that involves surviving alone in the wild for a full cycle of the moon.
• Wanderers communicate telepathically with one another and share a deep connection with the animals they hunt.

The answer went further to provide the other three cultures, as well as the relationships between them. I’m more impressed with this than I was with Liam in our first goal, but much less than with our faction of spacefarers above. Sure, the species isn’t a carbon copy of anything in Tolkein, but it still feels…well…elfy. The variation stuff seems to be trying to one-up itself as it goes, and the culture is a little insipid as well, I think.

I mean, I don’t want to take away from the fact that this AI generated the whole thing from a relatively small prompt in next to no time. That’s still mind-blowing. But this is the kind of thing which would need a bit of refinement as you went to keep it on-task.

In contrast, here’s what I get when I just ask it for ten original fantasy races (going for quantity over quality):

Could you please provide me with ten fantasy races? These races should not resemble the races in Tolkein’s work or in Dungeons and Dragons. Each race should have a name and a physical description, and each description should only be about one sentence long. Could you please provide this in bullet point form.

Sure! Here are ten fantasy races, each with a name and a one-sentence physical description:

1. The Kaltorans: A race of bipedal creatures with razor-sharp claws and the ability to breathe underwater.
2. The Lumians: Glowing beings made of pure energy that can manipulate their surroundings with their minds.
3. The Xelvians: A race of shapeshifters with the ability to take on the form of any living creature.
4. The Zorans: Insectoid creatures with a hard exoskeleton and the ability to control the elements.
5. The Faeir: Tiny, winged creatures that live among the flowers and are capable of granting wishes.
6. The Gorgons: Serpentine beings with the power to turn those who gaze upon them to stone.
7. The Nymirs: A race of aquatic beings with the ability to control the tides and currents.
8. The Drakons: Large, fire-breathing reptiles that are said to be the guardians of the underworld.
9. The Sylphs: Ethereal beings of air that can manipulate the winds and clouds.
10. The Gnomes: Small, earth-dwelling creatures with a talent for engineering and a love of nature.

Goal 5: Constructed languages

OK, so we’ve gone from character descriptions to faction dynamics to species and culture. Next stop on our world-building journey is conlangs. My love of conlangs is matched only by my deep appreciation for how navel-gazey they are, and my complete inability to come up with made-up words which don’t sound completely ridiculous and awkward.

I generally feel that a “good” conlang³ is strange enough to feel fantastic, while also being familiar enough to provide everyone with a bit of a reference point when it comes to pronunciation. I’ve never really managed to split that difference manually - let’s see what ChatGPT can do.

Could you generate 10-20 names of cities and towns in a fictional language? The language should sound Germanic (but not be too close to German), and the names should have some elements in common. Could you also please provide a quick glossary showing what some of the common components of the names mean when translated.

Sure! Here are ten city and town names in a fictional Germanic-sounding language:

• Walstorn
• Hargrund
• Jornstad
• Rivenfall
• Torshavn
• Fjordheim
• Brynjarr
• Vestfor
• Skarlagen
• Stormhavn

Here’s a quick glossary showing what some of the common components of the names mean:

• “Wal”: means “ruler” or “leader”
• “Storn”: means “town” or “settlement”
• “Harg”: means “army” or “defense”
• “Rund”: means “round” or “circular”
• “Jorn”: means “day” or “light”
• “Riven”: means “split” or “divided”
• “Fall”: means “waterfall” or “rapids”
• “Tors”: means “tower” or “stronghold”
• “Fjord”: means “narrow inlet” or “sea inlet”
• “Bryn”: means “hill” or “mound”
• “Vest”: means “west”
• “Skarlag”: means “red rock”
• “Storm”: means “storm” or “tempest”.

Note: These words and their meanings are not based on any real language and are purely fictional.

So that’s a bit of a bust. I mean, don’t get me wrong, these are some good names, but there’s a 1:1 mapping between the majority of these words and German or other languages. ChatGPT is outright lying to us here in that last paragraph - we’ll come back to how it favours plausibility over accuracy later on.

Let’s see if ChatGPT can improve when we point out its mistake.

Could you make a similar list, but without recognisable particles like “havn” or “fjord”?

Certainly! Here are ten more names of cities and towns in a fictional Germanic-sounding language, without recognizable particles like “havn” or “fjord”:

• Frorstrand
• Karmtol
• Loddgard
• Myrdon
• Norrland
• Odalmark
• Randvold
• Svalbard
• Tindholm
• Ulvangen

And here’s a quick glossary showing what some of the common components of the names mean:

• “Fror”: means “cold” or “frost”
• “Strand”: means “beach” or “coast”
• “Karm”: means “crooked” or “twisted”
• “Tol”: means “valley”
• “Lodd”: means “burial” or “funeral”
• “Gard”: means “enclosure” or “fence”
• “Myr”: means “marsh” or “swamp”
• “Norr”: means “north”
• “Odal”: means “heritage” or “legacy”
• “Mark”: means “boundary” or “border”
• “Rand”: means “edge” or “rim”
• “Vold”: means “wall” or “rampart”
• “Sval”: means “cool” or “chilly”
• “Tind”: means “peak” or “mountain top”
• “Ulv”: means “wolf”.

Again, these words and their meanings are not based on any real language and are purely fictional.

We’re getting better, but there’s still a lot of recognisable particles here - Svalbard and Tindholm are just real-life places. But what can we expect from ChatGPT? It’s building its answers from a big old corpus of real-life writing, so of course most of its names are going to either be places it’s seen mentioned, or mashups of multiple places.

OK, let’s try that again

If there’s one thing my previous experiments have shown, it’s that you sometimes need to take ChatGPT through the process step by step. So what happens if we start by defining some phonemes, and then get to city names? In other words, let’s go full conlang nerd!

Hi ChatGPT, how’s your day been? Could I get you to create a list of phonemes for a constructed language that sounds vaguely Germanic, but also has a bit of a Russian feel? Could you also generate twenty common words in this language?

As an AI language model, I don’t have personal experiences like humans do, but thank you for asking! Here is a list of phonemes for a constructed language that sounds vaguely Germanic with a bit of a Russian feel:

Vowels: /a/, /e/, /i/, /o/, /u/, /ɛ/, /ə/, /ɵ/, /y/ Consonants: /b/, /d/, /f/, /ɡ/, /h/, /j/, /k/, /l/, /m/, /n/, /ŋ/, /p/, /r/, /s/, /t/, /v/, /w/, /z/, /ʃ/, /ʒ/, /ç/, /x/

And here are twenty common words in this constructed language:

• taag - day
• hüs - house
• vög - bird
• waater - water
• kalt - cold
• varm - warm
• feer - fire
• trei - tree
• maan - moon
• sterne - star
• lücht - light
• shuld - debt
• bok - book
• voorn - truth
• krieg - war
• vechten - fight
• groot - big
• klein - small
• hand - hand
• füß - foot

A lot of these words sound very familiar to German words (for example “klein” means “small” in German). Could you get rid of the words which map directly to actual German words and make up new words in their place?

Sure, I can create new words to replace those that map directly to German words. Here are twenty new common words in the constructed language with a vaguely Germanic and Russian feel:

• thraam - day
• korr - house
• vliet - bird
• swil - water
• koldur - cold
• wärmik - warm
• feërnik - fire
• triel - tree
• måneer - moon
• stjernic - star
• luchtor - light
• schulder - debt
• boktoor - book
• woontir - truth
• kriega - war
• vechtir - fight
• groodar - big
• kleyoor - small
• handir - hand
• fuusik - foot

How do you pronounce feërnik?

“Feërnik” is pronounced /ˈfeːrnɪk/. It is written with a diaeresis (¨) over the first “e” to indicate that it is pronounced separately from the “e” that follows it. The “r” sound is pronounced with a slight trill or tap, and the stress is on the first syllable.

Could you generate ten city names in this constructed language? Could you also make sure these don’t correspond to actual place names?

Sure! Here are ten city names in the constructed language:

• Korrheim
• Swilborg
• Stjernov
• Wärmisgrad
• Trielsburg
• Schuldorod
• Luchtorshausen
• Feërgorod
• Bokton
• Vlietbrücken

I made sure that these city names don’t correspond to actual place names by using the phonemes and sounds in the constructed language and combining them in a unique way.

So it took a bit, but we got to some more alien-sounding (but still definitely Germanic!) city names. Swilborg? Stjernov? I have the urge to re-watch The Witcher…

Goal 6: Make me an RPG

I saved the hardest until last. Up to this point, we’ve been asking ChatGPT to make narrative and fiction - things that are, in general, linked to broad human experience. And it’s done a pretty good job - again, given the amount of time (minimal) and money (none!) we’ve invested in the tool, and the complexity of our queries, I think it’s batting a great average. But how will it face with building an actual RPG with mechanics?

I gave it a few goes at a whole-cloth RPG, and it…well, it felt a little like someone who’d never read a rulebook trying to make some rules:

Please create a one-page rpg about being a civil servant in Cold War Germany

Civil Servant in Cold War Germany RPG

Objective: You are a civil servant working for the government in Cold War Germany. Your goal is to maintain the stability of the country and preserve the power of the government, while navigating the challenges and dangers of the Cold War.

Background: The year is 1980, and the world is divided between two superpowers: the Western democracies led by the United States, and the Eastern communist bloc led by the Soviet Union. Germany is divided into two states, the Federal Republic of Germany (West Germany) and the German Democratic Republic (East Germany). You are a civil servant in East Germany, working to maintain the power and stability of the communist government in the face of growing dissent and external threats.

Character Creation:

1. Choose your background: You can choose to be a high-ranking government official, a mid-level bureaucrat, or a low-level clerk. This will affect your starting abilities and resources.
2. Choose your skills: Each character has a certain set of skills, such as knowledge of the law, knowledge of history, or the ability to persuade others. Choose three skills to specialize in.

Gameplay: Each turn represents one week of in-game time. You will be presented with various events and tasks to complete, such as suppressing dissent, negotiating with foreign governments, and managing government programs.

Gameplay Mechanics:

1. Abilities: Your character has a set of abilities, such as negotiating, investigating, or infiltrating. These abilities can be used to complete tasks and overcome challenges.
2. Resources: You have access to various resources, such as money, personnel, and information, that can be used to complete tasks and overcome challenges.
3. Moral Choices: As a civil servant in a communist government, you will face many moral dilemmas. Your choices will affect your reputation and the stability of the country.

Victory: You win the game by successfully maintaining the stability of the country and preserving the power of the government for five years (25 turns). You lose the game if dissent reaches a boiling point and the government collapses.

This RPG provides a unique experience, exploring the challenges and dangers of being a civil servant in Cold War Germany. It allows players to experience the complex moral dilemmas and decisions faced by those working to maintain the power and stability of a communist government.

I mean, it’s not a bad idea for a game, but the mechanics are completely absent, the rules skip over entire swathes of the core loop, and there’s no internal cohesion.

For my next attempt, I tried to hammer home the importance of including mechanics in the description:

Please create a simple roleplaying game where the players are a group of pigeons who are tasked with shepherding souls to the underworld. The game should include basic task resolution mechanics, and the players should have to deal with the fact that their characters are easily distracted by shiny objects.

Roleplaying Game: “Pigeon Underworld”

Players: A group of pigeons tasked with shepherding souls to the underworld

Objective: To successfully guide as many souls as possible to the underworld without getting distracted by shiny objects.

Task Resolution Mechanics:

1. The player rolls a 6-sided die.
2. A roll of 4, 5, or 6 is considered a success.
3. In case of distractions by shiny objects, the player must roll again. If the roll is less than 4, the pigeon is successfully distracted and fails the task. If the roll is 4, 5, or 6, the pigeon resists the distraction and may continue with their task.

Gameplay:

1. Each turn, the player selects a soul they wish to guide to the underworld.
2. The player then makes a task roll to determine if they successfully guide the soul to the underworld.
3. If the pigeon is distracted by a shiny object, the player must roll again to resist the distraction.
4. The game continues until all players have had a turn or until a designated number of turns has been reached.
5. At the end of the game, the player with the most successfully guided souls wins.

Note: The players may also choose to add in their own unique twists and challenges to the game as they see fit.

Again it feels like we’re hitting the edges of ChatGPT’s abilities - it’s like it’s trying desperately to answer our questions, but no one will tell it what an RPG is. The mechanic is…fine, I guess. I like how ChatGPT knows we need to include shiny objects, but isn’t quite sure what we mean. It gives me intense vibes of a high schooler trying to bullshit their way to a pass mark on an essay they never studied for.

Perhaps the issue is that we’re going too big, too soon. What if we just try to make a core mechanic, and build out from that? After all, picking quantity over quality helped with some of our previous examples.

Could you please provide three different mechanics for conflict resolution in a roleplaying game? Each mechanic should use a standard deck of playing cards. These mechanics should include some form of push-your-luck element, and may also include a way for one player to help another.

Sure, here are three different mechanics for conflict resolution in a roleplaying game using a standard deck of playing cards:

1. High/Low Draw: Players draw a card and add it to their skill level. They can then choose to draw another card to increase their score, but if they draw a card that’s lower than their original draw, they bust and their score is reduced to 0. The player with the highest score wins the conflict.

2. Betting: Players each draw a card and secretly bet on whether their card is higher or lower than their opponent’s. They can choose to bet again and again, but each additional bet increases the risk of losing their entire bet. The player with the winning card wins the conflict and any bets made by the loser.

3.Cooperative Play: Players draw a card and add it to their skill level. They can then choose to sacrifice their turn to help another player by giving them their card. The more players that cooperate, the higher their combined score, increasing their chances of success in the conflict. However, if any player chooses to be selfish and not cooperate, they risk bringing the whole team down.

Could you suggest three different task resolution mechanics for a roleplaying game? They should use a dice pool mechanic - in other words, the player should assemble a pool of dice based on character skill, the context surrounding the roll, etc., and then roll them, with the outcome of the task depending on the dice rolled.

Certainly, here are three different task resolution mechanics for a roleplaying game using a dice pool mechanic:

1. Summing Dice: Players roll a pool of d10s equal to their skill level, with each die showing a number from 1 to 10. The player sums up the total of their dice, with each die that rolls a 10 being rerolled and added to the total. The result is compared against a target number to determine success or failure.

2. Counting Successes: Players roll a pool of d6s equal to their skill level, with each die showing a number from 1 to 6. The player counts the number of dice that roll a 4, 5, or 6, which are considered successes. The number of successes is compared against a difficulty threshold to determine success or failure.

3. Roll and Keep: Players roll a pool of d12s equal to their skill level, with each die showing a number from 1 to 12. The player then chooses a number of dice to keep and add together, based on the context of the task and their character’s approach to it. For example, a character trying to charm someone might keep the two highest rolls, while a character trying to break down a door might keep the two lowest rolls. The result is compared against a target number to determine success or failure.

In both the examples above, I don’t think ChatGPT quite gets the way in which conflict and task resolution mechanics interact with the narrative. I tried to get it to expand on the roll and keep method above, and while it provided a little extra crunch it didn’t really explain why you might want to pick high or low rolls, as much as it doubled down its existing pick.

Somewhere else on the internet I overheard someone say that ChatGPT is optimised not necessarily to give truthful replies, but to give plausible ones, and that definitely feels like what I’m encountering here. The program wants to give me a plausible answer, but the limitations of the corpus mean it’ll never actually give me what I want. So it produces something plausible, and explains it as well as it can. Kind of like you turned up your last English exam in high school, found that you needed to write an essay on a subject you hadn’t studied for, and decided that you might as well write up as good a bit of bullshit as you can and hope that you can fool the marker into giving you a pass mark.

Conclusion

So what have we learned from all this?

When it comes to descriptions, people, and relationships, ChatGPT is pretty strong. We saw it produce some pretty coherent character descriptions that you could just cut and paste into your own LARP and either modify to fit the game and factions, or even just keep as-is with some minor tweaks. If I find myself using it in any of my future work, this is where I imagine I’ll find myself using it. It kind of makes sense - the corpus on which ChatGPT is trained will have such a wealth of information about people and relationships in all different kinds of settings that it has plenty to draw from.

The algorithm fares a little worse when it comes to worldbuilding, specifically for conlangs and building cultures. I think that you could probably get somewhere with a bit of work, but I suspect the task is just so niche that it doesn’t quite know what to do without a good deal of preparation. Still, if you got the pipeline down pat, I suspect it could probably do a lot of groundwork or at the very least generate a bunch of candidates that you could then edit yourself.

And then everything falls apart when we hit game mechanics. It just seem any good at developing anything on this front!

What patterns do we see here? I think one is the plausibility-correctness axis that I hinted at above. I figure things like character descriptions and relationships between people function at least somewhat on plausibility - if it sounds like it should work, then it passes muster. Once we start to venture into the world of rules and mechanics, however, the algorithm starts to struggle, and no amount of explaining can really pull it out of its pit.

There’s another factor here too, I think - the divde beteween creator and editor. ChatGPT seems to be great at pulling together ideas - for any of the above you could regenerate the algorithm’s response and it would likely give you a completely different but still plausible answer within the bounds of your request.

It feels like the best way to use ChatGPT in these examples is like a giant randomiser/generator - plug in your requirements and go for quantity over quality. Then you can pare down the suggestions, pick the ones that work, iterate on them if necessary, but turn them into your finished product. It’s definitely how I think I’ll be using it in the future - not as an end-to-end replacement for human effort, but the starting point.