Feed water through the brass pipe.
Learning to keep something alive.
Water the bed. Shelter it from heat. Release what it cannot hold.
Cover the bed to halve evaporation.
Drain up to 3.5 moisture points this step.
Interventions pause automatic care. Lumen travels, operates, then observes the result. Travel animates an action; it adds no physics steps.
What did it think would happen?
A changing balance
A local view
64 × 64 RGBThe pixel baseline sees this small instrument view. The live model uses five structured readings.
Observation boundary ↗
Waiting for an observation.
—
Six-step model rollout—
Live demonstration: guard offDoes experience earn its keep?
Ten controls. The same training and evaluation budgets. Memory, sensing, action access, planning depth, and assistance are tested separately.
The service continues a study if you close this page.
| Control | Viable time | Water used | Guard assists | Learner actions |
|---|---|---|---|---|
| A frozen study will make this comparison concrete. | ||||
The same season, two memories
Complete a study to inspect this journey.
The same weather and initial conditions.
Prediction probes, perturbations, and reusable sequences
Results appear after a study completes.
Three ways to question the result.
Small claims.
Evidence that travels.
Ashby gives us regulation. Lenski suggests useful stepping stones. Lorenz motivates perturbation tests. These experiments connect observation, prediction, intervention, and cost to the image-steering research.
Finite tests can verify a replay or expose a failure. They do not establish general intelligence, sentience, formal continuous-time safety, or open-ended evolution.