The adult answer key hidden inside the test

A standard consciousness experiment often looks simpler than it is. Show a stimulus. Record a neural or behavioural signal. Ask what the participant saw. Compare seen with unseen trials. The visible result is a correlation between a marker and a report. The invisible machinery includes intact language, a trained response rule, adult sensory speed, motivation, memory across the trial and a stable assumption about who the participant is.

Call the target property P, such as perceptual access to a face. Call the observation O, such as a late cortical response or a pointing movement. An observation supports the property only through a bridge explaining why this candidate, under this task, should produce that observation when the property is present. In adults, repeated calibration makes the bridge feel like part of the observation. In infants, organoids and split brains, it becomes visible because it breaks.

The first bridge is observable access. It connects a candidate state to a measurable channel. Speech is one channel. Eye movement, persistence, hand choice, autonomic response and a neural signature are others. A failed channel can hide an intact state. A positive channel can also be contaminated by task demands or inference.

The second bridge is world-coupling. It asks whether the candidate receives structured input, changes in a state-dependent way and uses that change when the environment responds. Spontaneous complexity alone cannot tell us whether a system distinguishes anything, learns a contingency or has anything like a bodily stake in the outcome.

The third bridge is subject individuation. It specifies which processes form one candidate for the property under study. A skull is an anatomical container. A dish is a laboratory container. Neither automatically identifies the causal boundary of a perceiver. The relevant boundary may differ for sensory integration, report, action control, agency and phenomenal unity.

These spans do different epistemic jobs. Observable access links a state to a channel the investigator can read. World-coupling links internal change to a structured source and to consequences that can alter later processing. Subject individuation links a measured process to one declared candidate rather than to an arbitrarily chosen container. None is, by itself, a necessary or sufficient condition for experience. They are conditions on our inference. This distinction matters because a weak access bridge can coexist with rich internal organisation, while a strong output channel can be driven by a shortcut that bypasses the target process.

The order also matters. Begin with the property and candidate boundary, then choose a probe, then specify the bridge, rival and intervention. Reversing that order invites marker-first reasoning: find a striking waveform, oscillation or response and only afterwards decide what it means. The same marker can be evidence for different claims under different bridges, and for no consciousness claim when the bridge is absent. This blocks two mirror-image errors: treating an inaccessible state as absent, and treating an accessible signal as phenomenally decisive.

When silence is developmental, not empty

An infant cannot satisfy the adult laboratory contract. There may be no stable verbal report, no reliable instruction following and no guarantee that adult timing windows fit a developing nervous system. The tempting conclusion is that consciousness cannot be studied until speech arrives. That conclusion confuses the absence of a preferred access channel with the absence of the target state.

A better question is whether several independent channels change together under a controlled manipulation. A face can be made progressively more visible. Looking behaviour, persistence, heart rate and electrophysiology can be measured without requiring the infant to name the face. Each channel has different failure modes. Agreement among them narrows the space of rivals, although it still does not convert a marker into a metaphysical verdict.

A second route avoids asking the infant to describe a state and instead asks whether behaviour tracks confidence. Goupil and Kouider found that twelve- and eighteen-month-old infants persisted longer after correct than incorrect choices, could use confidence to confirm or revise a choice, and showed an error-related neural response.³ The result supports a bounded claim about performance monitoring. It does not establish mature introspection or a verbally structured self.

The strongest infant design therefore does not search for one magic signature. It triangulates: manipulate visibility or uncertainty, validate timing within the developmental group, use more than one response channel and test whether the marker predicts later behaviour. A late wave, a change of mind and a physiological response become more informative when they respond to the same causal variable but do not share the same confounds.

Independence here is causal, rather than only statistical. Two measures recorded from the same electrode montage may share preprocessing artefacts. Looking time and pupil dilation may both be driven by novelty. A stronger cluster mixes routes that fail differently and predicts their pattern before collecting data. For example, increasing mask visibility should alter the late response, improve later discrimination and change confidence-sensitive persistence in a coordinated direction. If only arousal measures move, the consciousness-related interpretation weakens even though the experiment contains several positive variables.

Developmental validation also needs sensitive negative controls. A probe should detect an age-appropriate process known to be present, remain quiet when the relevant stimulus relation is removed and preserve its rank order on held-out trials. Without those checks, “no marker” may mean low signal-to-noise, poor fixation, an unsuitable temporal window or a response demand beyond the infant’s motor repertoire. A developmental null becomes evidence against the target only after the access bridge has demonstrated that it could have succeeded.

flowchart LR
  A["The adult answer key hidden<br/>inside the test"] --> B["When silence is developmental,<br/>not empty"]
  B --> C["When neural activity has no world"]
  C --> D["When one body does not settle<br/>one subject"]
  D -. "re-test" .-> A
  classDef first fill:#0a1424,stroke:#7B4CD8,stroke-width:2px,color:#e6f3ff
  classDef step fill:#0a1424,stroke:#3f7da0,stroke-width:1px,color:#e6f3ff
  class A first
  class B,C,D step

When neural activity has no world

An organoid presents the opposite temptation. Instead of too little evidence, it offers visually compelling evidence: neurons, synapses, oscillations, bursts, plasticity and sometimes task performance through an electrode interface. The tissue is biologically neural, so familiar patterns feel closer to mind than activity in silicon. But a brain is not conscious merely because each local piece resembles neural tissue. The missing question is what the activity is organised to do.

Embodiment is often reduced to having sensors and actuators. For this problem, the useful idea is narrower and more causal. A system is world-coupled when its internal changes depend on structured input, when its output changes what happens next and when those consequences matter to future processing. The loop can be experimentally minimal. It need not resemble a human body. Yet open-loop stimulation, spontaneous bursting and a replayed input stream do not establish the same relation.

Even a closed loop is not enough unless the loop carries a discriminating contingency. A thermostat changes the world and receives the consequence, but its single error variable supports only a narrow control claim. An organoid connected to an array may show richer state dependence, yet the experiment still has to separate learned contingency from stimulation history, electrode drift and generic adaptation. The useful progression is therefore not “more activity, then consciousness”. It is structured input, selective internal change, causal use, reversal, retention and recovery. Each step removes a rival while leaving the phenomenal question open.

The candidate boundary must be declared just as carefully. Is the candidate the tissue alone, the tissue plus electrode interface, or the complete closed-loop apparatus? If performance disappears when the external decoder is replaced, the relevant computation may belong to the coupled system rather than the organoid. If a partition within the tissue destroys learned discrimination while preserving overall firing rate, the internal organisation gains evidential importance. Moving the boundary is not bookkeeping; it changes which system is credited with adaptation.

Organoid studies already occupy different rungs. Trujillo and colleagues reported complex oscillatory activity whose statistical features modelled aspects of early human network development.⁵ Cai and colleagues connected organoids to a high-density multielectrode array, sent spatiotemporal electrical inputs and used the resulting nonlinear dynamics and fading memory for reservoir-computing tasks.⁶ Alam El Din and colleagues reported spontaneous and evoked activity, connectivity, critical dynamics and input-specific short- and long-term plasticity, while also documenting results that did not generalise uniformly at the whole-network level.⁷

These are genuine advances. They support claims about neurodevelopmental modelling, adaptive physical computation and synaptic plasticity. They do not by themselves identify a point of view. Oscillation can arise in many excitable systems. Criticality can describe a regime of network dynamics. Reservoir performance can be achieved by non-conscious media. Plasticity says that past input changes future response, not that the change is felt.

When one body does not settle one subject

Split-brain research removes a different shortcut. After the corpus callosum is cut to treat severe epilepsy, many ordinary interactions remain remarkably coherent. Yet carefully lateralised tests reveal that information delivered to one visual half-field may not be available for comparison with information delivered to the other. The same person can look unified in conversation and divided under a millisecond-controlled task.

The classic picture is simple: the left visual field projects mainly to the right hemisphere, while the right visual field projects mainly to the left. In a person with intact callosal connections, information can be integrated across the midline. After complete callosotomy, some cross-field comparisons fail. The temptation is to infer two minds, one per hemisphere. The rival temptation is to preserve one mind because there is one body, one life history and one socially coherent agent.

Pinto and colleagues tested two people with complete callosotomy. The participants were highly accurate at indicating the presence and approximate location of a stimulus across the visual field using verbal, left-hand or right-hand responses. At the same time, they failed some comparisons between stimuli presented in opposite visual fields.¹¹ The authors argued that perception could be divided while conscious agency remained unified.

Volz and Gazzaniga offered a strong alternative. A hemisphere may signal another through subtle eye movements, shared attention, posture, facial movement or other forms of cross-cueing. Residual non-callosal pathways may carry coarse location or alerting information without transferring detailed content.¹² A successful verbal response therefore does not prove that the left hemisphere directly received the right hemisphere’s visual representation.

A collective review found empirical common ground beneath the disagreement: callosotomy produces a broad but incomplete breakdown of functional integration. Some aspects of action control remain unified. The available evidence is insufficient to decide whether consciousness is one or two, and intermediate descriptions may fit better than a binary answer.¹⁴

“Unity” therefore has to be expanded into dimensions before it can be measured. Information may be unavailable for cross-field comparison while a single hand-selection policy remains stable. Verbal report may be lateralised while orienting and posture coordinate across the body. Autobiographical continuity can remain socially singular even when perceptual contents are partitioned in the laboratory. These facts do not average into one unity score. They form an integration profile whose axes include content sharing, metacognitive access, action arbitration, memory continuity and control of common effectors.

The decisive experiments close suspected routes one by one. Stabilise fixation so that eye movements cannot carry location. Randomise response timing so that one hemisphere cannot train the other across sequential trials. Compare verbal, left-hand and right-hand responses under identical stimuli. Then perturb a route and ask which capability disappears. If detection survives but relation judgements fail, the intervention identifies a functional cut. A functional cut constrains theories of subjecthood, but it does not count subjects without an additional bridge from organisation to phenomenality.

stateDiagram-v2
  [*] --> Claim
  state "Executable lab: edge-case<br/>evidence worksheet" as Claim
  state "Frame the claim" as Test
  state "Test the mechanism" as Revision
  Claim --> Test: expose mechanism
  Test --> Decision: evidence survives
  Test --> Revision: rival survives
  Revision --> Claim: narrow or rebuild
  Decision --> [*]

Build the missing bridges

The three cases now suggest one research discipline. Do not ask whether a marker “is consciousness”. Ask what property it targets, how the candidate can make that property observable, what structured world relation gives the marker causal content and which processes form the candidate subject for this task. Then build the strongest non-conscious or non-target system that could reproduce the observation.

The resulting method is an edge-case bridge stress test. It does not aggregate indicators into a consciousness score. Scores hide missing assumptions by converting them to numbers. The stress test returns typed statuses: supported, provisional or blocked. It also returns the strongest permitted claim, prohibited inferences and the next experiment that would most reduce ambiguity.

The logic is deliberately asymmetric. A supported bridge permits a bounded inference only for the named property, task and candidate boundary. A provisional bridge keeps the claim open but identifies the assumption doing the work. A blocked bridge prevents the inference even when the observation is statistically impressive. Conversely, a negative observation lowers confidence in the target only when the probe was sensitive, the channel was available and the rival failure modes were controlled. This preserves honest uncertainty without making every result inconclusive.

Every row in the worksheet begins with a candidate boundary and one target property. “Consciousness” is too broad for the first pass. Choose perceptual access, confidence monitoring, closed-loop adaptation, cross-field integration or another property that can produce a discriminating prediction. Then state the observation and the rival that could mimic it.

For an infant late wave, the rival may be attention or novelty without conscious access. The intervention might hold sensory energy constant while varying visibility and test whether the marker predicts later confidence-sensitive behaviour. For an organoid oscillation, the rival may be generic excitable-network dynamics. The intervention might reverse an input-output contingency while matching total stimulation. For a split-brain response, the rival may be cross-cueing. The intervention might eliminate eye, posture and sequential-action cues while testing multiple response routes.

{}

The worksheet encodes three assumptions. First, a positive result permits only the named functional or theory-relative claim. Second, a negative result counts against the target only when the probe was sensitive and the bridge valid. Third, no configuration establishes phenomenality, moral status or the priority of physical or non-physical ontology.

const classify = (values) => {
  const total = values.reduce((sum, value) => sum + value, 0);
  const hasIntervention = values.includes(2);
  if (total >= 5 && hasIntervention) return "supported";
  if (total >= 3) return "provisional";
  return "blocked";
};

function evaluate(record) {
  const status = {
    access: classify(Object.values(record.access)),
    world: classify(Object.values(record.world)),
    subject: classify(Object.values(record.subject)),
    discipline: classify(Object.values(record.discipline))
  };

  return {
    ...record,
    status,
    rule: "No scalar consciousness score is produced.",
    prohibited: [
      "no report implies no experience",
      "brain-like dynamics imply consciousness",
      "one anatomical container implies one subject"
    ]
  };
}

The code exposes the rule that prose often hides. It treats bridge fields as ordered evidence states, requires at least one intervention for “supported” and refuses to collapse the result into one total. To apply it, choose the closest preset, edit each field and export the record with the study protocol. A supported result permits a bounded claim about the named bridge. A blocked result identifies what has not been established. The tool cannot decide whether a candidate feels anything.