NEXUS ACADEMY · GUIDED LEARNING
Evidence Across Worlds: How to Test a Claim
Six connected lessons on the limits of arguments, observations, simplified models and AI performance.
A complete six-module inquiry course using all six published Nexus Humano guides and three original Labs to distinguish observations, interpretations, validity, model assumptions and evaluation evidence.
Use your browser’s Print command to keep a worksheet with an explained answer key.
Your learning objective
Build a claim-evidence audit that identifies each question, the kind of method used, the observation available, a plausible competing explanation and the limit of what can responsibly be concluded.
Modules
Philosophy & Meaning · MODULE 1
1 · What exactly needs explaining?
Start with the difference between an observable report and a definition or theory of subjective experience.
Read Consciousness and identify where the guide discusses first-person experience, empirical observations and competing explanations. Then visit the deductive-argument Lab as a comparison of inferential structure.
Why this Lab is relevant — and what it cannot test: The Logic Lab belongs to the distinct Deductive Arguments guide in the same published Philosophy world. Here it is a companion method exercise about valid inference, not an empirical measure of experience or a claim that the two topics have a published direct graph edge.
Check your understanding
A fictional person describes seeing a blue circle while a device registers a signal. Does this observation alone establish a complete theory of consciousness?
Reveal the explained answer
No. The imagined association might be relevant to a theory, but the observation needs careful study and does not settle every conceptual or causal question. A recorded report, measurement and interpretation must be distinguished.
Answer key: No. The imagined association might be relevant to a theory, but the observation needs careful study and does not settle every conceptual or causal question. A recorded report, measurement and interpretation must be distinguished.
Three original investigation tasks
Investigation 1
Write two questions about the fictional blue-circle report: one about a measurable association and a second about the concept of first-person experience. Make their evidence requirements explicit.
Compare an expected response
One empirical question requiring an actual study design, and one conceptual question that cannot be resolved by a single co-occurrence.
Example: One empirical question requiring an actual study design, and one conceptual question that cannot be resolved by a single co-occurrence.
Investigation 2
Draft a three-column evidence sheet separating the imagined person's report, the recorded device signal and the commentator's proposed explanation. Mark any unobserved causal claim.
Compare an expected response
The first two entries describe supposed observations in an invented example; the final causal interpretation requires additional support.
Example: The first two entries describe supposed observations in an invented example; the final causal interpretation requires additional support.
Investigation 3
Use the Logic Lab to examine a conditional inference. Explain why a valid conditional form differs from a scientific claim with uncertain premises.
Compare an expected response
Formal validity tests consequence if premises hold; factual truth and the connection to lived experience remain separate questions.
Example: Formal validity tests consequence if premises hold; factual truth and the connection to lived experience remain separate questions.
Read the complete worked synthesis
Worked example: ‘The report and signal occurred together’ is an imagined observation, not a causal conclusion. ‘The signal made the person conscious’ goes beyond what was given. A logical argument could be valid in form but still rest on unsupported premises. A more responsible statement identifies the observation, the proposed interpretation and the evidence that has not been collected.
Worked synthesis: Worked example: ‘The report and signal occurred together’ is an imagined observation, not a causal conclusion. ‘The signal made the person conscious’ goes beyond what was given. A logical argument could be valid in form but still rest on unsupported premises. A more responsible statement identifies the observation, the proposed interpretation and the evidence that has not been collected.
What this activity cannot establish: The scenario contains no real recording or participant data. The companion Lab demonstrates logical form only, and this module neither measures a person's mental state nor establishes whether any artificial system is conscious.
Why the next world connects: In the next module, biological cognition provides a more directly observable set of mechanisms. Do not assume a change of vocabulary answers the original philosophical question.
Life, Body & Mind · MODULE 2
2 · From cognition to neural mechanisms
Relate a question about understanding or attention to a biological mechanism without confusing levels of explanation.
Read Brain and Cognition and inspect the toy neuron-threshold Lab. Compare a behavioural description with a proposed mechanism and note the assumptions needed to connect them.
Why this Lab is relevant — and what it cannot test: The threshold Lab is published with the Neuron Signals guide, not Brain and Cognition. Their existing published part-of graph relation justifies studying the narrower mechanism alongside the broader cognitive question; a toy input cannot measure an individual's understanding.
Check your understanding
If an invented model produces a signal at one threshold, can we infer that a person will understand a difficult text?
Reveal the explained answer
No. The invented signal is an output of a simplified arithmetic rule. A learner's performance involves many processes, contexts and forms of evidence not represented by the Lab.
Answer key: No. The invented signal is an output of a simplified arithmetic rule. A learner's performance involves many processes, contexts and forms of evidence not represented by the Lab.
Three original investigation tasks
Investigation 1
Write an observation that could be made about a fictional learner answering a question, then identify a neural mechanism one might wish to investigate independently.
Compare an expected response
A behaviour can be described without assuming a specific cell-level cause; hypotheses about mechanisms require appropriate measurements.
Example: A behaviour can be described without assuming a specific cell-level cause; hypotheses about mechanisms require appropriate measurements.
Investigation 2
Change the Lab's invented excitation and inhibition values on opposite sides of the boundary. Record its internal arithmetic, then list at least three biological variables it omits.
Compare an expected response
A recorded threshold decision follows the invented rule, while real neurons require data about voltage, timing, cell interactions and other biology.
Example: A recorded threshold decision follows the invented rule, while real neurons require data about voltage, timing, cell interactions and other biology.
Investigation 3
Propose one fair comprehension task and explain why neither the toy Lab's output nor a single correct response establishes a person's general intelligence.
Compare an expected response
A limited task can examine a defined capability in context but cannot support broad diagnostic or personal claims.
Example: A limited task can examine a defined capability in context but cannot support broad diagnostic or personal claims.
Read the complete worked synthesis
Worked example: In the toy model, six excitation units less two inhibition units gives four; if the invented threshold is four, the displayed rule is satisfied. That output can teach what a threshold is without recording real membrane voltage. A person's reading comprehension is a different outcome at a different scale, demanding independent observations and a relevant research design.
Worked synthesis: Worked example: In the toy model, six excitation units less two inhibition units gives four; if the invented threshold is four, the displayed rule is satisfied. That output can teach what a threshold is without recording real membrane voltage. A person's reading comprehension is a different outcome at a different scale, demanding independent observations and a relevant research design.
What this activity cannot establish: Fictional learner outcomes are not psychological data, and the threshold simulator is not a neural recording or clinical instrument. Do not infer a person's intelligence, diagnosis or learning needs from this exercise.
Why the next world connects: Next, compare modelled task performance in AI. Similar mathematical notation does not imply identical causes or kinds of experience.
Technology & Futures · MODULE 3
3 · What does a machine output show?
Evaluate observed AI behaviour without treating fluent words as a complete account of subjective experience.
Read AI and Consciousness. Use the classifier-tradeoffs Lab as a separate example of a measurable artificial-system outcome, not a consciousness detector.
Why this Lab is relevant — and what it cannot test: The classifier Lab belongs to the distinct Model Evaluation guide in the same published Technology world. The published AI-and-Consciousness versus Model-Evaluation graph link explicitly says measured prediction accuracy alone cannot establish subjective experience.
Check your understanding
A fictional chatbot writes a convincing description of fear. Which claim is directly supported, and which question remains open?
Reveal the explained answer
Its output can be described and evaluated as text. A conclusion that the system possesses a specific subjective experience requires further conceptual argument and evidence beyond fluent language.
Answer key: Its output can be described and evaluated as text. A conclusion that the system possesses a specific subjective experience requires further conceptual argument and evidence beyond fluent language.
Three original investigation tasks
Investigation 1
Describe only the observable features of a fictional machine-generated paragraph: the prompt, the output and whether it appears relevant. Avoid claims about unobserved feeling.
Compare an expected response
An output report can support a narrow behavioural statement without supplying a directly observed subjective state.
Example: An output report can support a narrow behavioural statement without supplying a directly observed subjective state.
Investigation 2
Write two rival interpretations of fluent output and identify the missing evidence for each. Explain why repeating the same fluent example would not resolve the disagreement.
Compare an expected response
Competing interpretations require explicit concepts, different evidence or arguments, not merely an emotional response to phrasing.
Example: Competing interpretations require explicit concepts, different evidence or arguments, not merely an emotional response to phrasing.
Investigation 3
Run the classifier Lab and record what its confusion counts actually evaluate. Compare its error metric with the claim that a machine has a subjective experience.
Compare an expected response
A classifier metric concerns labelled fictional outcomes, not consciousness, personhood or universal safety.
Example: A classifier metric concerns labelled fictional outcomes, not consciousness, personhood or universal safety.
Read the complete worked synthesis
Worked example: The fictional chatbot generated a coherent description of an emotion. This observation demonstrates a particular kind of language output. It does not determine whether the system has subjective experience, any more than a high fictional classification score establishes all properties of a model. The responsible conclusion identifies which behavioural task was measured and keeps philosophical questions explicitly unresolved.
Worked synthesis: Worked example: The fictional chatbot generated a coherent description of an emotion. This observation demonstrates a particular kind of language output. It does not determine whether the system has subjective experience, any more than a high fictional classification score establishes all properties of a model. The responsible conclusion identifies which behavioural task was measured and keeps philosophical questions explicitly unresolved.
What this activity cannot establish: This module makes no assertion about the consciousness, safety or subjective experience of any actual AI provider's systems. The cases, labels and chatbot are fictional and are not a validated benchmark.
Why the next world connects: The following modules turn back to tools that check inference, simplified dynamics and error rates; each method answers a narrower question.
Philosophy & Meaning · MODULE 4
4 · Can the premises be true while the conclusion fails?
Use formal counterexamples to test whether a conclusion follows from an argument's structure.
Read Deductive Arguments and use the truth-table Lab to compare modus ponens and affirming the consequent. Note separately whether the premises describe a true fact.
Why this Lab is relevant — and what it cannot test: This original Logic Lab is directly linked to Deductive Arguments. Its truth tables can establish the logical result of a stated propositional form but do not verify any real cabinet or artefact evidence.
Check your understanding
If every item from cabinet A has a green label and this item is green, must it have come from cabinet A?
Reveal the explained answer
No. Other cabinets may also use green labels. The inference affirms the consequent, and a possible counterexample has true premises but a false conclusion.
Answer key: No. Other cabinets may also use green labels. The inference affirms the consequent, and a possible counterexample has true premises but a false conclusion.
Three original investigation tasks
Investigation 1
Translate the cabinet example into symbolic conditional form and name the antecedent and consequent. Keep the observation separate from the premise.
Compare an expected response
If A then G; G; therefore A is affirming the consequent, not a deductively valid form.
Example: If A then G; G; therefore A is affirming the consequent, not a deductively valid form.
Investigation 2
Build a possible world in which the cabinet statement and green-label observation are true but the claimed cabinet origin is false.
Compare an expected response
A second cabinet also using green labels supplies a counterexample to the inference.
Example: A second cabinet also using green labels supplies a counterexample to the inference.
Investigation 3
Use the Logic Lab to compare this with a valid form. Add a sentence explaining why valid logic still cannot tell you whether the premises are factually accurate.
Compare an expected response
Modus ponens is valid when its premises hold; verifying the actual premises is a separate empirical task.
Example: Modus ponens is valid when its premises hold; verifying the actual premises is a separate empirical task.
Read the complete worked synthesis
Worked example: Let A mean ‘the item came from cabinet A’ and G mean ‘the item has a green label’. A implies G. Observing G does not entail A, because G could be true for an item from cabinet B. The truth-table Lab demonstrates the structural problem. Whether a real item came from a specific archive would require provenance evidence and cannot be inferred from the imagined colours.
Worked synthesis: Worked example: Let A mean ‘the item came from cabinet A’ and G mean ‘the item has a green label’. A implies G. Observing G does not entail A, because G could be true for an item from cabinet B. The truth-table Lab demonstrates the structural problem. Whether a real item came from a specific archive would require provenance evidence and cannot be inferred from the imagined colours.
What this activity cannot establish: This is a fictional logical example. It is not a real archaeological provenance assessment, a legal proof standard or a demonstration of how often mislabelling occurs.
Why the next world connects: A mathematical test of logical form is exact within its definitions. A toy neural signal also follows a stated rule, but the model may not represent the physical system under discussion.
Life, Body & Mind · MODULE 5
5 · When is a model useful but incomplete?
Identify the boundary between correct calculations in a teaching model and credible real-world biological inference.
Read Neuron Signals and test the invented excitation/inhibition boundary. Then compare this arithmetic exercise with the broader cognition chapter.
Why this Lab is relevant — and what it cannot test: The Lab is directly supported by the published Neuron Signals guide. Its invented arithmetic is explicitly not a full account of membrane potential, ion channels, timing or a person's health.
Check your understanding
The toy system displays a signal when excitation six and inhibition two reach a boundary of four. Has a real neuron fired?
Reveal the explained answer
No. Six minus two reaches the specified numerical threshold, but all the units and boundary are invented. The output is an educational illustration, not a direct physiological recording.
Answer key: No. Six minus two reaches the specified numerical threshold, but all the units and boundary are invented. The output is an educational illustration, not a direct physiological recording.
Three original investigation tasks
Investigation 1
Define the toy model's exact inputs and decision rule before running any values. Give two numerical predictions that differ by crossing its four-unit boundary.
Compare an expected response
One invented subtraction reaches or exceeds the boundary and another does not; the results follow the rule by construction.
Example: One invented subtraction reaches or exceeds the boundary and another does not; the results follow the rule by construction.
Investigation 2
List what would need measurement to test a real biological explanation, including membrane potential and time course, and identify what the simulation omits.
Compare an expected response
The toy arithmetic provides no genuine electrophysiological measurement or context-dependent mechanism.
Example: The toy arithmetic provides no genuine electrophysiological measurement or context-dependent mechanism.
Investigation 3
Compare what a correct toy-model result and a correct experimental finding mean. State how the real question determines whether the simplification is useful.
Compare an expected response
A model can teach a defined relationship without matching every real biological mechanism or predicting an individual's activity.
Example: A model can teach a defined relationship without matching every real biological mechanism or predicting an individual's activity.
Read the complete worked synthesis
Worked example: With an invented threshold of four, six excitation units minus two inhibition units triggers the rule; five minus two does not. This is internally consistent. Actual neurons depend on membrane dynamics, channels, timing and networks. The model is useful for explaining an abstract threshold but cannot establish what occurs in a specific biological cell or cognitive process.
Worked synthesis: Worked example: With an invented threshold of four, six excitation units minus two inhibition units triggers the rule; five minus two does not. This is internally consistent. Actual neurons depend on membrane dynamics, channels, timing and networks. The model is useful for explaining an abstract threshold but cannot establish what occurs in a specific biological cell or cognitive process.
What this activity cannot establish: All units and measurements are fictional. The activity makes no biomedical prediction, diagnosis, treatment recommendation or claims about an actual person's neural activity.
Why the next world connects: The last module uses a different model: a classifier evaluated against labelled examples. Both models require clear tasks, definitions and limitations.
Technology & Futures · MODULE 6
6 · Which performance number answers the question?
Finish by auditing a fictional model's error counts, test assumptions and unsupported extrapolations.
Read Model Evaluation and operate the original classifier-tradeoffs Lab. Connect its metrics to a limited claim you can support, then return to the course's first philosophical question.
Why this Lab is relevant — and what it cannot test: The classifier Lab is directly linked to the published Model Evaluation guide. Its deliberately fictional labels demonstrate metrics rather than real-world safety, fairness or a decision about any person's status.
Check your understanding
Can a classifier that labels ninety out of one hundred fictional items correctly be assumed to find most of the positive items?
Reveal the explained answer
No. Accuracy alone aggregates correctly labelled examples. Recall depends on how many actual positives were detected, and the confusion counts are needed to understand missed cases.
Answer key: No. Accuracy alone aggregates correctly labelled examples. Recall depends on how many actual positives were detected, and the confusion counts are needed to understand missed cases.
Three original investigation tasks
Investigation 1
Choose two classifier thresholds in the Lab and record the four confusion-matrix counts for each. Explain how the error balance changed.
Compare an expected response
A higher or lower threshold can alter the pattern of true positives, false positives, true negatives and false negatives in the invented dataset.
Example: A higher or lower threshold can alter the pattern of true positives, false positives, true negatives and false negatives in the invented dataset.
Investigation 2
Compare accuracy, precision and recall, and identify the case in which precision is undefined because no positive predictions were made.
Compare an expected response
Different metrics describe different aspects of labelled fictional outcomes; an undefined denominator is not a verified zero.
Example: Different metrics describe different aspects of labelled fictional outcomes; an undefined denominator is not a verified zero.
Investigation 3
Write a final claim-evidence audit for one course example: question, method, observed result, rival explanation, scope and what would require independent specialist review.
Compare an expected response
A cautious conclusion uses only the method's demonstrated output, identifies missing evidence and does not infer consciousness or global safety.
Example: A cautious conclusion uses only the method's demonstrated output, identifies missing evidence and does not infer consciousness or global safety.
Read the complete worked synthesis
Worked example: A classifier can label ninety of one hundred invented cases correctly and still miss a large share of the minority positive class. Reporting TP, FP, TN and FN permits targeted accuracy, precision and recall calculations; when TP and FP are both zero, precision is undefined. Such results describe one labelled test and do not establish how an untested system will behave, whether it is safe in every setting or whether it has experience.
Worked synthesis: Worked example: A classifier can label ninety of one hundred invented cases correctly and still miss a large share of the minority positive class. Reporting TP, FP, TN and FN permits targeted accuracy, precision and recall calculations; when TP and FP are both zero, precision is undefined. Such results describe one labelled test and do not establish how an untested system will behave, whether it is safe in every setting or whether it has experience.
What this activity cannot establish: All labels, thresholds and outcomes are educational. No benchmark results, fairness audit, safety certification or estimate of real-world deployment performance is reported.
Why the next world connects: Close the loop: different fields can each use rigorous tools, but the strength of a conclusion depends on the exact question, observational scope and interpretation of the evidence.
Bring the worlds together
Produce one two-page claim-evidence audit using a case from any of the six modules. Identify the exact question, distinguish directly observable information from assumptions, name the model or method, propose one competing explanation, record an uncertainty or limitation and explain which different published guide could help investigate what remains. Revisit a claim if a cited source or modelling assumption changes.
No grading or profile: reflect privately, revisit earlier steps and compare your answers with the source-linked explanations.
Continue: the consciousness trail ↗
Other complete Academy courses
- How Models Help Us Think — Work through three already-published guides and labs to learn how different fields test a model, interpret results and recognize limits.
Research references
These are the references already listed by the linked published guides and labs; review the original resources for context.
- Stanford Encyclopedia of Philosophy — Consciousness
- Stanford Encyclopedia of Philosophy — Neuroscience of Consciousness
- Stanford Encyclopedia of Philosophy — Argument and Argumentation
- Stanford Encyclopedia of Philosophy — Classical Logic
- OpenStax — Psychology 2e: How Memory Functions
- OpenStax — Biology 2e: How Neurons Communicate
- OpenStax — Anatomy and Physiology: The Action Potential
- NIST — Artificial Intelligence Risk Management Framework
- Google Machine Learning Crash Course — Classification Metrics
- Google Machine Learning Crash Course — Overfitting and Generalization
Original classroom investigations and invented scenarios built from six previously published source-linked Nexus Humano guides. Candidate references in those guides are research starting points, not independent specialist endorsements or verification of these invented examples. This guided sequence was compiled with AI assistance from published Nexus Humano materials; no external academic endorsement or course accreditation is implied.