Identify the claim
Separate factual claims, interpretations, recommendations, calculations, citations, and uncertainty statements.
Unit overview
Unit 10 begins the course's focused study of epistemic responsibility. Generative AI may produce coherent content that includes invented details, incorrect claims, unsupported conclusions, missing qualifications, contextually unsuitable advice, invalid reasoning, or unjustified precision. You will learn to divide an output into verifiable claims, classify different forms of error and uncertainty, consider possible consequences, and apply the author-approved TRACE framework to document findings, corrections, escalation, and stopping.
Classify the problem
Distinguish fabrication, factual error, unsupported claim, omission, mismatch, invalid inference, and false precision.
Correct responsibly
Determine required evidence, assess consequences, document corrections, preserve uncertainty, and stop when verification is inadequate.
Central idea: A polished response should be treated as a set of provisional claims, not as established knowledge. Verification effort should increase with consequence, uncertainty, novelty, and the difficulty of correction.
Unit learning outcomes
By the end of Unit 10, you should be able to:
- Distinguish hallucination, fabrication, factual error, unsupported claim, material omission, contextual mismatch, invalid inference, and false precision.
- Separate an AI-generated passage into independently assessable claims.
- Distinguish facts, interpretations, recommendations, predictions, calculations, citations, and uncertainty statements.
- Explain why fluency, confidence, detail, formatting, and citations do not establish reliability.
- Identify internal inconsistencies, unverifiable specifics, inappropriate certainty, and missing qualifications.
- Assess the possible consequence, reversibility, and verification burden associated with an error.
- Determine what kind of evidence or qualified review each claim requires.
- Distinguish correction from concealment, paraphrase, or unsupported replacement.
- Preserve justified uncertainty rather than forcing an unsupported answer.
- Apply the author-approved TRACE framework to a bounded output or workflow.
- Define escalation and stop conditions when verification is unavailable or consequences are unacceptable.
- Begin Assignment 5 and extend Assignment 1 with evidence from Unit 10.
Unit 10 lessons
Complete the five lessons in order. The filenames are local to the dedicated Unit 10 folder.
10.1 Hallucination and Error Types
Distinguish fabrication, factual error, unsupported claims, omissions, contextual mismatch, invalid inference, and false precision.
10.2 Claim Identification and Decomposition
Break fluent passages into factual claims, interpretations, recommendations, calculations, citations, and uncertainty statements.
10.3 Credibility Cues and False Confidence
Examine how tone, detail, structure, technical language, numbers, quotations, and references can create unwarranted confidence.
10.4 Consequence, Correction, and Uncertainty
Prioritize verification by consequence, identify appropriate corrections, preserve uncertainty, and design escalation or stopping.
10.5 TRACE Error and Uncertainty Audit
Apply TRACE to a complete output, document evidence needs and corrections, and create the initial record for Assignment 5.
Folder convention: Keep
unit-10.html and t-01.html through t-05.html in the same dedicated Unit 10 Moodle folder.Embedded Unit 10 study guide
- Decompose
Convert fluent output into assessable claims. - Classify
Identify error, omission, mismatch, inference, and uncertainty. - Prioritize
Assess consequence, evidence needs, and verification burden. - Correct
Apply TRACE, document changes, escalate, or stop.
Suggested study sessions
- Retrieval and baseline audit.
Review CREATE, read Chapter 7, and annotate one supplied AI output without external verification. - Lesson 1.
Classify error and uncertainty types using bounded examples. - Lesson 2.
Create a claim inventory for a supplied passage. - Lesson 3.
Identify credibility cues and explain why each cue is insufficient. - Lesson 4.
Rank claims by consequence and design correction, escalation, and stopping. - Lesson 5 and Assignment 5 start.
Apply TRACE, select a bounded claim set, and update Assignment 1.
Evidence to retain
- Initial output annotation
- Error-type classification
- Claim inventory
- Credibility-cue analysis
- Consequence and verification-priority matrix
- Correction and uncertainty record
- TRACE audit
- Assignment 5 claim-set proposal
- Assignment 1 reflection notes
Show the error-audit sequence
- Define the purpose and consequence of the output.
- Separate the output into independently assessable claims.
- Classify each claim and identify required evidence.
- Mark unverifiable details, citations, calculations, and quotations.
- Identify material omissions and contextual mismatches.
- Assess consequence, reversibility, and verification burden.
- Verify high-priority claims using suitable sources.
- Correct the record and preserve residual uncertainty.
- Escalate to a qualified reviewer where needed.
- Stop when adequate evidence or authority is unavailable.
Assessment milestones
Assignment 5: Verification and Fact-Checking
Weight: 10%
Start: Unit 10, Week 7.
Develop: Across Units 10 and 11, then refine during Units 12 and 13.
Recommended submission: End of Module 4, Week 8.
Official Moodle due date and time: ____________________
Assignment 1: Journal continues
Reflect on confidence, credibility cues, error types, verification burden, correction, uncertainty, and how TRACE changed your evaluation of AI output.
Recommended final submission: End of Module 6, Week 11.
Official Moodle due date and time: ____________________
Choose a bounded claim set: Assignment 5 should contain specific, independently verifiable claims and enough source access to support a documented verification process.
Show the Unit 10 Assignment 5 record
- Selected output and legitimate reason for analysis
- Bounded claim set
- Claim-type and error-type classifications
- Evidence requirements
- Consequence and verification priority
- Preliminary source plan
- Corrections and remaining uncertainty
- TRACE findings
- Required AI-use documentation
Required learning activities
| Activity | Purpose | Product | Time |
|---|---|---|---|
| Error classification | Distinguish fabrication, factual error, unsupported claims, omissions, mismatch, and inference. | Classification table | 35–50 min |
| Claim inventory | Convert a fluent passage into assessable claims. | Claim table | 35–50 min |
| Credibility-cue audit | Identify persuasive features that do not prove accuracy. | Annotated output | 30–45 min |
| Internal consistency check | Identify contradictions, unsupported specificity, and missing qualifications. | Consistency record | 30–45 min |
| Consequence matrix | Prioritize verification based on consequence, uncertainty, and reversibility. | Priority matrix | 35–50 min |
| Correction exercise | Correct supported errors and preserve unresolved uncertainty. | Before-and-after record | 40–55 min |
| TRACE audit | Integrate error, evidence, consequence, correction, escalation, and stopping. | Structured audit | 45–60 min |
Participation options
Option A: Approved AI
Generate or compare bounded, low-risk outputs using synthetic, public, supplied, or authorized information.
Option B: Supplied outputs
Use supplied passages, claims, citations, calculations, annotations, and correction records.
Option C: Non-AI analysis
Complete the same claim decomposition, error classification, consequence, and TRACE analysis without an AI platform.
Responsible-use boundaries and support
Boundaries
- Do not use health, legal, financial, safety, employment, disciplinary, or other high-consequence claims for live experimentation.
- Do not upload confidential, personal, proprietary, privileged, unpublished, or culturally governed information.
- Do not treat generated citations, quotations, calculations, or source descriptions as authentic until verified.
- Do not conceal errors by merely rewriting them more fluently.
- Do not claim certainty beyond the available evidence.
Support and accessibility
- Use supplied outputs if live AI is inaccessible, unsuitable, or unwanted.
- Contact the course coordinator for claim-set and assignment questions.
- Use library services for source planning and verification support.
- Seek qualified disciplinary or professional review when claims exceed your competence or authority.