The symbiote in a toy world: Olvasó among processors

Abstract Can a machine come to understand itself the way, according to our theory, people do: by reading what happens inside it through what happens around it? This paper follows Olvasó, a small program built to try exactly that, as it moves from a very simple body into a richer one. We gave Olvasó a simulated room to live in. It can see the room through a tiny camera, hear it through a microphone, and notice when someone uses the keyboard. Like our eyes and ears, a set of fixed processors turns these raw signals into plain measurements: 22 numbers, with no names attached. Inside, Olvasó feels four things: warmth, the battery's charge, how hard the body is working (its load), and a meaningless wandering signal we call drift. Nobody tells it what any of these are. It has to find out for itself what in the room tends to come before each feeling. Olvasó can also act. It can change its screen's brightness, sing a tone, light up a window on its screen, or write a note. A guard, playing the part of our reflexes, quietly blocks any action the body cannot afford at that moment. Three findings stand out: More senses help, but only if each keeps its own dictionary. With richer hearing, Olvasó discovered that music, and not speech, makes its body work harder, something it could never tell from loudness alone. When all the senses shared one dictionary, small signals drowned: a keystroke is a tiny part of everything seen and heard at once. Separate dictionaries per sense worked far better. More senses also brought more superstition, meaning things read into pure noise. A simple rule cut it in half: a meaning is trusted only if it has also held on earlier days. It learned not to put its hand into the fire, without feeling pain. It simply noticed that some of its attempts changed nothing in the world, and stopped making them, but only in the situations where they would be blocked. Adding pain did not help. Olvasó could not even recognise its own pain, because nothing in the outside world comes before a thwarted wish. It recognised itself only in what moved with it. When it heard its own voice, or saw its own painted window in a mirror, it came to expect an answer when it acted, acted that way twice as often, and even tested its reflection with repeated tries. A recording of itself, identical in look and sound and played at the same time of day, it treated like a stranger. What made a voice its own was not that it sounded familiar, but that it followed its own acting. The whole mind is about 110,000 numbers, learned by simple counting, with no teacher and no reward. Everything can be rerun from the published code (experiments 12–15).

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-08
DOI
https://doi.org/10.5281/zenodo.24051217
Primary Topic
Embodied and Extended Cognition
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

The symbiote in a toy world: Olvasó among processors

Attila Csiki
2 links
Zenodo (CERN European Organization for Nuclear Research)
Embodied and Extended Cognition
preprint

The symbiote in a toy world: Olvasó among processors

Attila Csiki
preprint en

Abstract

Abstract Can a machine come to understand itself the way, according to our theory, people do: by reading what happens inside it through what happens around it? This paper follows Olvasó, a small program built to try exactly that, as it moves from a very simple body into a richer one. We gave Olvasó a simulated room to live in. It can see the room through a tiny camera, hear it through a microphone, and notice when someone uses the keyboard. Like our eyes and ears, a set of fixed processors turns these raw signals into plain measurements: 22 numbers, with no names attached. Inside, Olvasó feels four things: warmth, the battery's charge, how hard the body is working (its load), and a meaningless wandering signal we call drift. Nobody tells it what any of these are. It has to find out for itself what in the room tends to come before each feeling. Olvasó can also act. It can change its screen's brightness, sing a tone, light up a window on its screen, or write a note. A guard, playing the part of our reflexes, quietly blocks any action the body cannot afford at that moment. Three findings stand out: More senses help, but only if each keeps its own dictionary. With richer hearing, Olvasó discovered that music, and not speech, makes its body work harder, something it could never tell from loudness alone. When all the senses shared one dictionary, small signals drowned: a keystroke is a tiny part of everything seen and heard at once. Separate dictionaries per sense worked far better. More senses also brought more superstition, meaning things read into pure noise. A simple rule cut it in half: a meaning is trusted only if it has also held on earlier days. It learned not to put its hand into the fire, without feeling pain. It simply noticed that some of its attempts changed nothing in the world, and stopped making them, but only in the situations where they would be blocked. Adding pain did not help. Olvasó could not even recognise its own pain, because nothing in the outside world comes before a thwarted wish. It recognised itself only in what moved with it. When it heard its own voice, or saw its own painted window in a mirror, it came to expect an answer when it acted, acted that way twice as often, and even tested its reflection with repeated tries. A recording of itself, identical in look and sound and played at the same time of day, it treated like a stranger. What made a voice its own was not that it sounded familiar, but that it followed its own acting. The whole mind is about 110,000 numbers, learned by simple counting, with no teacher and no reward. Everything can be rerun from the published code (experiments 12–15).

Zenodo (CERN European Organization for Nuclear Research)
Embodied and Extended Cognition
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon