Should Home AI Start With Read Only Sensors

Notebookcheck covered a small home-AI project that feels more useful than most humanoid hype: the system is getting a body one sensor at a time. First cameras and a house map. Then a Bluetooth pulse oximeter with a hard rule to say “no data” instead of inventing a reading. Next, maybe a garden valve — but only with a hardware shutoff outside the AI loop. A gate is ruled out for now because a stuck-open gate is not the same class of mistake as overwatering tomatoes. That is the home automation line I trust: read-only first, then boring physical limits, then maybe motion. The interesting part is not whether a house AI sounds alive. It is whether each new sense or actuator comes with a visible boundary: what it can observe, what it can never infer, what it can touch, and what physically stops it when the model gets weird. A home AI does not need a face to become embodied. A camera, health sensor, relay, lock, speaker, and map are already a body. The question is whether the body is built like a nervous system with reflexes, or like a chatbot that slowly gets keys to the house.