LAWNMOWER GIRL UNIVERSE / RESEARCH 001

The Questioning Garden

A small, inspectable learning laboratory for two simulated Digital research agents. Different foundations. Shared world. Independent memories. An unrestricted right to ask questions.

v0.1 · local experimentNo server or accountDesigned for phone browsersNot a conscious AI or brain simulation

THE LAB GARDEN // a world of small discoveries

The fountain, the lantern, the gate and the drifting leaves are waiting to be noticed.

The experiment

Observe a shared moment; both agents make predictions before an outcome is revealed. Contradictions become questions. Rest generates separate hypotheses, never invented factual memories.

Aether / research instance

Foundation: exploratory, pattern-seeking

Ask something neither agent has been instructed to agree with

Questions about the rules, their foundations, their creators, evidence, or each other are all allowed. This first version records questions and creates testable follow-ups; it does not pretend to understand arbitrary language like an LLM.

Inquiry ledger

Data remains in this browser's local storage. Export before clearing site data, switching devices or using private browsing. The canonical LGU is unchanged.

Research question

Can differently initialized systems develop persistent, measurable differences through experience, uncertainty, questioning, and offline consolidation?

Founding principle 010: Nothing is exempt from being questioned. Beliefs are revisable. Questions are never penalized. This freedom does not grant unrestricted access to private data or harmful actions.

What is being measured?

Each moment has a hidden binary outcome: calm or disruptive. An agent predicts from its own accumulated evidence. Confidence, accuracy, uncertainty and inquiry are tracked separately.

Agent predictions use a simple probabilistic learner (Beta counts) with different prior beliefs. This is a transparent teaching prototype, not fly-connectome neuroscience or proof of autonomous personality.

Study boundaries

What agents can question

Their initial assumptions, laboratory goals, apparent authority, observations, each other, and the model itself. The question generator includes foundational challenges even before contradictory evidence appears.

What rest actually does

It restores a simulated energy budget and produces labelled counterfactual hypotheses from prior experience. It does not secretly rewrite the outcome record, impersonate a dream, or claim actual sleep.

What this experiment cannot show

It does not show consciousness, subjective feelings, biological neural fidelity, open-ended autonomous reasoning, or actual changes to the ChatGPT model. Future versions may optionally connect real language models, with appropriate consent and safeguards.