I made tiny prototypes of two ideas, and I will record my observations for both:
Guided by plot (previous idea about dystopia where only elites can buy cognitive-enhancement devices)
The idea for this prototype was to guide the player through a plot and story I had already written (detailed in my previous blogpost), in a parser-fiction style, and use GPT to take the character’s action and guide it along a prespecified narrative arc to a prespecified set of outcomes. I’d say it worked okay but not amazingly.
- The early plot points were exciting, but GPT would get into “loops” where it asks the player for too much input, slowing the progression of the game
- Not clear enough what the goal is. Make the player’s “goal” in a scene very clear up front, otherwise not exciting
- Not enough time to explain the plot — the story needs to be shorter, more like a flash fiction than a novel, because there’s only so much a player can get through.
Guided by “goals” + plot
From the above 3 observations from the playtest I decided to recraft my game around a new shorter story. The idea is that a rogue AI has taken over the world, and is now judging humans based on whether they are deemed worthy or not to live. The player is assigned the role of a human who has lived in this world, and has 1 minute to convince the AI they are worth sparing. There is a running score (starting at -5), and if the player gets enough points (earned or lost at each conversation step), they are spared and win the game.
- Instructions weren’t clear enough. Folks didn’t know to include short messages, otherwise they’d keep running out of time
- Not enough narrative immersion; hard to guide someone through a story with this setup
- Still very fun to roleplay a character, and the specificity + snarkiness from the custom suggestions is surprisingly delightful
Conclusion
My observations point me towards the idea that I need to come up with a) a more concise and simpler to explain plot, b) if I do use GPT to expand the number of actions a player can do, use it to “fill in the gaps” rather than drive everything, and c) seriously consider a clever technical architecture blending GPT storytelling (for its dynamicism) with my writing (for quality / retentiveness that is superior to GPT’s, for now) in a dynamic way.