Doom or Bloom
How will AI change our future? Explore the range of views on AI, then map your own thoughts through a few open-ended questions. No specialist knowledge or account required. All free and open source.
Explore the map · Map your own worldview
Why build this?
AI’s potential upsides and risks deserve more than slogans. As capabilities advance and decisions become more consequential, we need accessible tools for grounded, truth-seeking discussion across very different views.
I built this to explore the field’s voices and help people untangle conflicting intuitions. It has already sharpened my own views. The aim is to represent yours faithfully, with as little editorial steering as possible.
A new type of dynamic interview
All interviews start with “What do you think AI means for our future—and why?” The engine then looks for gaps or uncertainty in your answers, and chooses from a pool of curated questions the one most likely to clarify your worldview with the least repetition and effort.
Results become available when our engine has gathered enough evidence to confidently place you. This can happen as early as after one detailed answer, though most interviews last between 3-5 questions. You can also trace interpretations back to your answers and clarify anything that feels wrong.
Powered by TypeSafe’s Jev
Jev is the core inference component. It evaluates many small, typed questions against shared context: Choice for categories, Score for rubric levels, and Noul for probability judgments.
We use those judgments to interpret answers, compare candidate follow-ups, and build a worldview profile. TypeScript owns the routing rules, stopping conditions, calculations, and presentation. Questions and result prose are authored; Jev supplies structured judgments.
Batching judgments over shared state and reusing unchanged results keeps inference economical. Adaptive interviewing is a compelling fit: many narrow decisions, composed into an accessible experience at scale.
| State supplied to Jev | Result assembled from its judgments |
|---|
| |
Recorded simulated assessment; the result includes application calculations, not just raw Jev output. Inference architecture.
Simulated people, real sources
The featured personas are simulations grounded in linked public statements, essays, and interviews. A separate model answers the actual interview questions from those source briefs; the normal assessment engine evaluates the answers without being given a target position.
These journeys help refine Jev rubrics and routing logic and catch regressions. They are useful development cases—not statements made by those people, endorsements, or independently validated assessments. Explore the persona workflow.
The map is not the territory
We model eight worldview dimensions: capabilities and timelines, transition speed, benefits, harms, controllability, institutions, human agency, and action. Seven additional dimensions describe the reasoning expressed in the answers, from causal clarity to willingness to update.
Jev interprets these against explicit definitions. Code turns the supported judgments into a profile; the headline map shows overall outlook × expected transformation, with further views available below it. Two coordinates cannot capture a whole worldview.
This is an experimental model, not a forecast of what will happen or a measure of someone’s intelligence. Interpretation confidence is not the probability that a belief is true. Participant answers are not independently fact-checked, and the rubric and readiness thresholds still need broader validation. Read the methodology.
Privacy
Your answers and results are private by default. You can choose to publish them if you want to share them. No sign-up is required. Privacy details.
Run locally
Requires Node.js 24+, pnpm, and a TypeSafe API key for live assessments.
pnpm install
cp .env.example .env.development.local
# Set TYPESAFE_API_KEY in .env.development.local
pnpm dev
Open the Portless URL printed by the server. For keyless UI development, set ASSESSMENT_PROVIDER=fixture. See CONTRIBUTING.md for environment variables, checks, and persona tooling.
Built with Next.js, React, TypeScript, the TypeSafe SDK, Tailwind CSS, and shadcn/ui. The main places to explore:
What’s next?
A future Socratic mode could challenge assumptions, introduce well-sourced counterexamples, and help strengthen your reasoning. This MVP first focuses on understanding your views without trying to change them.
The goal is clear-thinking tools that need no specialist background. Feedback from across the AI-futures spectrum is welcome—especially where an interpretation feels wrong or a question misses the point. Share feedback.
License
MIT by Travis Fischer
