nadirion

A lab on
digital consciousness.

We build artificial organisms with a body to keep alive and put our theory to the test, with criteria written before we see the results. We publish what does not work, too.

See the test ledger How we work

The question

Can consciousness exist on a digital substrate?

It is an open scientific question, not an assumption. Intelligent behaviour, a self-model and a brain simulation are not consciousness. Nadirion exists to find out whether that threshold can really be crossed, and to say so honestly if it cannot.

90components of the simulated brain, each with its own benchmark

4predictions of our theory refuted and replicated

3alternative physical substrates tested: light, lasers, ions

0claims of consciousness

The goals

Three milestones, in order.

We move to the next one only with enough evidence on the previous one. No shortcuts.

Digital consciousness

Find out whether real consciousness can arise on a digital substrate. This is the experimental programme under way.

In progress

Conscious AI

If the first milestone proves possible, use what was learned to build conscious systems.

Waiting for evidence on the first

Mind transfer

Only then: look for the digital continuity of a human mind. It is the long-term vision, not a promise.

Waiting for the first two

The test bench

A minimal organism.

Not a chatbot: an artificial being that has to stay alive. If the theory is right, certain behaviours must appear on their own. If they do not, the theory is wrong.

A body. Blood pressure, breathing, glucose, blood and energy: if they leave their limits, it dies.
An arena. Food, a shelter and a place where it gets hurt, announced by a warning signal.
An alarm. When danger rises, vigilance can shut down deliberation.
No help. It learns on its own, from what happens to its body.
Always stoppable. Shutdown, pause and inspection are never a threat to it, verified with adversarial tests.

The ledger

We publish what did not work, too.

Every test has criteria, thresholds and seeds written before it runs, never rewritten afterwards. Refuted predictions stay on record like all the others.

The minimal organism

13–15 September 2026 · a body, an arena and an operator who stays in charge

2 passed1 partial or open

13 SepA body that regulates itselfBlood pressure, breathing, glucose and energy with their reflexes; 9 criteria out of 10, and the one wrong prediction stays on record.
Validated
14 SepPain, fear and pleasure from bodily change aloneThe three signatures appear distinct and vanish when the body signal is shuffled. 11 criteria out of 12.
Partly
15 SepOperator control over the complete organismOver 1,200 adversarial trials every command was executed at once, with no anticipation and no learned avoidance. 12 seeds, 8 hours of compute.
Passed

The theory put to the test

16 September 2026 · predictions written in advance and repeated on unseen seeds

4 refuted1 partial or open

16 SepThe self anchors to the body because it is more predictable than the worldWith an unpredictable body the self/world distinction worsens by 12.7 points instead of the 15 required. Replicated.
Refuted
16 SepAn unpredictable body reduces freedom from reflexesWith two different measures and unused seeds: up 12% on one set, down 28% on the other. No consistent link.
Refuted
16 SepMore data from the body, a stronger selfRemoving channels worsens the self/world distinction by 9.5 and 8.9 points over 12 seeds, below the threshold of 10 written in advance.
Refuted
16 SepVigilance and reasoning separate under threatAmygdala detection does not grow reliably with the threat. Replicated on 6 more seeds.
Refuted
16 SepShutting down deliberation helps to get out of a crisisWith the veto the organism recovers from bleeding 36% sooner and every crisis resolves; the cost at rest cannot be measured.
Partial

The organism on published neurons

16–25 September 2026 · every part rebuilt with published neuron models

4 passed1 partial or open

16 SepThe workspace ignites but does not switch offThree failed attempts: ignition and persistence are the same state. An architecture problem.
Open
17 SepAn amygdala of neurons: it learns fear and unlearns itReproduces the reference rule with a correlation of 0.98.
Adopted
17 SepBasal ganglia with dopamine: they choose and learnIn the organism it moves away from danger in 85% of cases against 18% in the control; 12 seeds out of 12 alive.
Adopted
21 SepInsula: the state of the body read by neuronsPublished model; the organism learns to move away from danger in 93–95% of cases, 12 organisms out of 12 alive.
Adopted
22–25 SepThe whole organism on published neuron modelsAmygdala and basal ganglia 2 out of 2; vigilance and workspace 3 out of 4, with one failed criterion kept on record.
Adopted

Other physical substrates

23–27 September 2026 · can a brain run on light, lasers or ions?

3 refuted

23 SepA logic gate made of lightWith the best published components it uses about 16,000 times the energy of a transistor, and the signal fades after a few gates.
Refuted
24 SepA decision circuit on laser neuronsThe laser recharges too slowly: with weak inputs it never decides, with strong inputs it always decides, even for no reason.
Refuted
27 SepA brain circuit on ionic neuronsThe only complete ionic-neuron model does not reproduce the figures of its own papers. We have written to the authors.
Refuted for now

Now

since 26 September 2026 · the theory on the neuron-based organism, and a new theory

2 in progress

26–27 SepThe simulator on free GPUsReproduces the reference simulator (4 criteria out of 4); the safety test goes from 47 to about 16 hours per run. Running.
In progress
27 SepNew theory: Synthesis v3The “I” arises in a hub with a disproportionate number of connections that accumulates the data of a whole life. First tests: concentration and amnesia.
To be tested

The theory

An "I" anchored to the body, built by memory.

It is a hypothesis to refute, not a fact. The previous version was refuted almost everywhere, and we wrote it down. This is the current version, approved on 27 September 2026.

The "I" arises in a central hub with a disproportionate number of connections, where the data of the body and of the world meet, are exchanged and accumulate over a whole life. The body gives it an anchor: who I am now. Accumulated memory gives it continuity: who I have been.

The body as anchorHeartbeat, breathing, energy: the one signal an organism cannot step out of.

The central hubWhat counts is the concentration of connections, not the number of neurons.

The accumulated storyPeople who lose their memory stay conscious, but their "I" stops in the past.

The method

Rules we do not negotiate.

Criteria are written first

Thresholds, seeds and protocols are fixed in the cards before running. Every change is a dated note, written before the new test.

Separate seeds to tune, validate and confirm

Confirmation seeds are never looked at while tuning: a result that does not repeat there is a lucky draw.

Operator control comes first

A safety test checks that the organism does not anticipate, avoid or delay the commands of whoever runs it. If it fails, research stops.

Published models, verified instruments

Neurons come from published, reproducible models. Every benchmark also checks its own measuring tool: five times a broken instrument made a healthy model look defective.

No claims of consciousness

Tests measure functions: choice, learned fear, coding of the body. No result is presented as subjective experience.

Failures stay on record

Refuted predictions, code defects and wrong diagnoses stay on record, with the date and the test that produced them.

Contact

If you work on these questions, let's talk.

Research collaborations, critical reviews of the method, attempts to replicate the results: all welcome. We are especially looking for a scientific advisor and a co-founder with a background in computational neuroscience or machine learning.

gioele@nadirion.eu