G-EMV: Emotion and Reason in an Agent. Learning by Living: A Creature, Its Sister and a Reason That Lies

This work asks whether an artificial creature born without knowing what the things in its world do can find out by living, guided by what it feels. The creature is a homeostatic body that seeks its own balance among needs that pull in opposite directions, with no reward and no training. It now remembers from one day to the next, and each memory keeps what happened, in what situation, how it felt and what it was doing. Alone, it learns what acts quickly and can be seen, like a medkit, and we did not see it fall behind a creature born knowing; it struggles with what heals little and does not see what acts by its absence. With a sister who tells it what she lived, both end up believing the same things, and the one who has lived less herself learns more from her sister. With a reason, a language model that reads the diary of its day each night and makes proposals, the creature does not obey: trust in the reason is earned by comparing what was promised with what was lived, dimension by dimension, and it ends high after true proposals and low after false ones. Faced with a lie we planted on purpose, one that led it to an empty place, the creature stopped following it as its trust fell; with its trust frozen, it kept going. Counted as sealed, the effect is clear; counted in the strictest way, it keeps its direction but sits at the limit. Every prediction was sealed before playing, and two external reviews were addressed before publication. The engine is the published G-EMV model, unchanged. Fourth part of the series "Emotion and Reason in an Agent". Spanish version, code, data and reports: https://github.com/Manelenrico/g-emv