In keeping with my self-imposed constraints, I started the project with a simple query.
The opening prompt
If I were to build a hypothetical path from today to a world in which AI gains sentience and is an autonomous entity of itself, could you help me identify a series of crucial incidents or events that need to happen along the way. I am trying to flesh out a science fiction story about how AI achieved sentience and I thought a series of separate incidents like this might be a good way to go about it.
— responded with a set of ten scene sketches, each gesturing at a different facet of the question.
It was a good list, but I pushed back suggesting instead that we create a set of stories that are grounded in reality so that each story feels completely plausible to the reader in the light of current information that we have and the existing technological capabilities of the modern world. This became the central constraint and core instruction of the project. Each story would have to earn its place by being a plausible extrapolation from what is currently possible and should follow from the one before it, the accumulation of them all finally producing the final outcome.
The other constraint that I thought would be useful was to make sure that each beat represented an evolution in the capabilities of AI that humans welcomed. This allowed us to refuse the easy story—there would be no Skynet, no rebellion, no moment where the public recoils. Every step of the way, the world would be getting better so that by the time the question was whether the thing we are speaking about is conscious, the world has become unimaginable without it.
The pushback
These are nice but I want them to be grounded in scientific rigour. They must all be 100% possible in the light of current computer science or a reasonable extrapolation of it.
Each successive incident must be incremental so that the world in the subsequent incident reflects a subtle change from the previous world — one which humans don't really object to but instead sort of embrace as more convenient, or more enjoyable or more interesting — and it is in the context of this new world that the next incremental evolution takes place.
Can you redesign the arc based on this. Feel free to be more detailed in your articulation of each beat.
This is a much better constraint. Let me rebuild from the ground up, starting from where we actually are today and making each step something that follows from real research trajectories — and critically, something people actively welcome.
Based on these instructions, Claude rebuilt from the ground up, starting from where we actually are today and making each step something that follows from real research trajectories and something people actively welcome. It came up with 10 beats, world states that are each slightly advanced from the previous in terms of technological capabilities, but not exponentially so in a way that would make it seem far-fetched. While the world in Beat 10 might be unimaginable today, if the reader gets there one story at a time, it would feel completely plausible.
Based on this Claude came up with the first version of the story bible—the central scaffold that, story after story, would iterate across 23 versions before the final word was written.