Causal grammar

Change a condition. Cut an arrow.
Follow the consequence.

What changes when you act?

World

The declared world

1 active 0 inactive interventionClick a node to intervene · dotted ring = changed

Starting conditionstap to flip

do =

Your causal program

SET A = 1A CAUSES BA PREVENTS BA AND B CAUSE C

The same sights leave three answers.

Outside input do =

A causes B

Observed: 00 · 11

A · B

B causes A

Observed: 00 · 11

A · B

A common cause

Observed: 00 · 11

A · B

Observation is shared. Intervention reveals a difference.

One distribution. Two dictionaries.

Toy bigram model · add-one smoothing

After P(next token)

Next \| previous	Count / total	ln p

Shared corpus · edit the evidence

Signal do(A = )

Dictionary one

dax → cause
wug → prevent

Dictionary two

dax → prevent
wug → cause

What would count as evidence?

Selected literature through 2024

01 / Humans & other animals

Learning and acting

Prediction & actionmeasurable

Causal task performancetestable

Experienced causationunsettled

Seth & Bayne · 2022 ↗

02 / Plants

Signaling and growth

Adaptive responsemeasurable

Neuronsabsent

Subjective experienceunestablished

Taiz et al. · 2019 ↗

03 / This lab

A declared construction

Boolean consequencescomputed

Organism measurementsnone

Experience in any systemunestablished

Pearl · 2009 ↗

Prediction · action · report · experienceDifferent questions require different evidence.

About this experiment

The exactness belongs to a stipulated Boolean grammar. This is a computational construction, with no neural or organism measurements.

What changes when you act?

SET supplies root conditions. Incoming rules determine their targets. An intervention replaces a target's incoming equation. The graph shows this by cutting its incoming edges. Conjunction means every input must be one; prevention means B equals one minus A.

The same sights leave three answers.

Each model produces A = B when its outside root is zero or one. Perfect agreement with those observations does not choose a causal direction. An intervention can expose disagreement, though these particular interventions do not distinguish all three worlds.

One distribution. Two dictionaries.

The count table is a toy bigram model, not a neural language model. Identical corpora yield identical token probabilities. The two dictionaries are stipulated externally; they assign opposite Boolean operations to each signal. This construction separates token distribution from intervention semantics without claiming that a trained model lacks causal structure.

Learning and acting

Predictive processing, active inference and causal learning address different aspects of organisms' prediction and action. Consciousness theories remain competing frameworks. Task success requires controls for simpler cues and does not certify experienced causation.

Signaling and growth

Electrical and chemical signaling, physiological memory and adaptive growth establish complex plant responses. They do not by themselves establish subjective causal experience. Richer intelligence vocabulary remains contested in the selected literature.

A declared construction

Formal exactness here is finite and stipulated. No brain, animal, plant, chemical system or language-model activation was measured. The visual networks are diagrams of a construction; their resemblance to living organization proves nothing about experience.

What changes when you act??

The same sights leave three answers.?

A causes B

B causes A

A common cause

One distribution. Two dictionaries.?

What would count as evidence?

What changes when you act?

The same sights leave three answers.

One distribution. Two dictionaries.