Three-Week AI-Assisted Reading Pilot

Three weeks of reading sources and judging AI sentences

The TF for AI-Enhanced General Education presents a three-week AI-assisted reading pilot proposal . Proposed for committee discussion, it follows checking the source → finding omissions → revising sentences.

• Presenter: Achmage (안창현) , Thursday, October 8, 2026
• Meeting: TF Committee Meeting 2026-3 , scheduled to begin at 12:00
• Location: seminar room, fifth floor, Basic Education Building · Room 16505

Source · Slide 1

The question this pilot examines

Can students assess AI sentences against source evidence and revise them when needed?

• Learning to examine: Source comprehension and finding evidence
• Evidence to retain: Records of judgment, evidence, and revision
• Operating factors to examine: Student difficulty and instructor workload

The three-week goal is to examine feasibility and signs of change in learning . This is not an experiment establishing long-term effects.

Source · Slide 2

International examples and the scope of application

An actual Northeastern class provides the primary model. Students compare AI summaries with the source, check accuracy and omissions, and discuss them. Official classroom example (S1)

An actual Monash class is a supporting example: read first, receive AI assistance, then return to the source and explain it again in one's own words. Official classroom example (S2)

An official UCLA HumTech activity also informs the proposal: AI asks questions and students explain key points, evidence, and remaining questions. Official activity (S3)

The existence of these official classroom examples and activities has been checked. However, they do not establish the effectiveness of this particular three-week design. The three-week plan is an application proposal based on those examples.

Source · Slide 3

Teaching through checking, comparing, and revising

Use fewer tools while increasing the reading challenge.

1. Check: Find source evidence together with the instructor.
2. Compare: Check omissions and exaggerations with a partner.
3. Revise: Judge independently and explain why you revise or retain a sentence.

Use the same worksheet and three fields every week, with materials prepared in advance. The standard plan is three weeks × 120 minutes per week . For integration into an existing class, the 50-minute adaptation is described in the schedule below.

Source · Slide 4

Three records used every week

Prepared in advance by the instructor

Short source text + AI summary

+ Paragraph numbers

Provide both the reading material and the fields to complete
from the start.

• AI sentence to check: Choose one sentence from the summary.

• Evidence found in the source: Record the paragraph number and relevant passage.

• My judgment and reason: Correct / Partially correct / Cannot verify

Do not add more fields in week three. Add a revised sentence or a reason to retain it in the third field only.

Source · Slide 5

Week 1: Check evidence with the instructor

Check together

First action

Watch the instructor demonstrate how to find evidence.

Student activity

Record the relevant passage, paragraph number, and judgment.

Record to keep

AI sentence + Source evidence + Judgment and reason

Instructor's question: Let us find where this is stated in the source.

Proposal for committee discussion [attachment] [S1]

Source · Slide 6

Week 2: Compare omissions and exaggerations with a partner

Compare with a partner

Source wording: some · possibility · conditions

≠

Summary wording: all · certainty · always

Student activity

Read new text R2 and identify changes in scope and conditions in the summary.

The partner's role

Ask where the evidence is, rather than supplying the answer.

Record to keep

Omission or exaggeration + Source evidence + Judgment

Instructor's question: What would we miss if we read only this summary?

Proposal for committee discussion [attachment] [S1][S2]

Source · Slide 7

Week 3: Revise or retain independently

Judge independently

Practice text R3

Revise one summary sentence for new text R3.

If it is not wrong

If it is accurate, record evidence and reasons for retaining it.

Final check

Check a new diagnostic text without AI tools or help from a partner.

Keep a record of: Original sentence → Source evidence → Revised sentence / Reason to retain

Source · Slide 8

Applying the pilot and assessing learning

Criteria retained across subjects

Example field	Reading material to examine	Additional subject-specific understanding
Humanities and social sciences	Argumentative texts or explanations of social phenomena	Claims and evidence, alternative perspectives
Natural sciences	Scientific explanations or research introductions	Conditions, uncertainty, and scope of results
Arts and interdisciplinary subjects	Artwork commentary or technology criticism	Observation and interpretation, application of concepts

• Keep consistent: Three-field worksheet · Diagnostic timing · Common scoring criteria

• Leave to the course: Reading texts · Core concepts · Comprehension questions

The field names and material types are illustrative examples . The TF will decide the actual pilot courses.

Proposal for committee discussion

Source · Slide 9

Three areas of diagnosis

AI literacy

Understand and check AI

Prioritize the TF's common diagnostic form.
If none exists, use provisional scenario questions.

About five minutes each, before and after

Common reading performance

Read the source and judge

New texts A/B + a fixed summary
Complete independently without AI tools.

About ten minutes each, before and after

Subject learning

Understand the course content

Core concepts, claims, and application questions
Choose two or three per course.

About five minutes each, before and after

Standard plan: allocate about twenty minutes at the beginning and end. Do not combine confidence or satisfaction with actual performance scores.

Proposal for committee discussion

Source · Slide 10

An eight-point rubric for the judgment process

8 points maximum · Proposed scale

• Core understanding: Did the student explain the source's central meaning?

• Source evidence: Did the student accurately locate the relevant passage?

• Judgment of the AI sentence: Did the student distinguish correct, partially correct, and unverifiable?

• Revision and explanation: Did the student revise to match the source or explain why it should be retained?

Per criterion: 0, 1, or 2 points · Do not score sentence length, AI-like style, decoration, or proficiency with Obsidian features.

Proposal for committee discussion

Source · Slide 11

Limits when interpreting the results

What to report

Changes within the same students

Number of students with both pre- and post-responses
Common reading scores and subject-specific understanding
Completion rate, level of assistance, and preparation time

What not to claim

Effects not established

That AI caused the improvement
Simple rankings of scores across different courses
Established long-term reading habits or general ability

Decisions about expansion consider changes in learning + implementation burden + equitable access together.

Proposal for committee discussion

Source · Slide 12

Operating conditions and committee decisions

Instructor preparation and equitable participation

Instructor preparation

Three texts · Three summaries

Prepare evidence-based answers and worksheets,
and remove ambiguous scoring items.

Student access

Obsidian, LMS, or paper

Use the same material, three fields, and scoring criteria.
Do not require paid personal accounts.

Operating safeguards

Can run without internet

Distribute fixed material and retain only necessary records.
Do not enter personal information into AI.

Research use, consent, and retention scope will be decided separately after checking the responsible procedures as applicable.

Proposal for committee discussion

Source · Slide 13

Four decisions for the committee

Which courses?

Participating courses · Instructors · Number of students

When, and for how many minutes?

Three-week start date · 120-minute or shorter format · Integration into class

Which materials and tools?

Source texts and summaries · Common worksheet · Alternative submission route

04

What evidence will inform the decision?

Common diagnosis · Course-specific questions · Expansion criteria

With one worksheet,

• retain evidence of reading and checking: Proposal for committee discussion

Source · Slide 14

Schedule and assessment tools

A 120-minute plan and a 50-minute adaptation

Appendix A · Time allocation

120 minutes Includes diagnostics in weeks one and three

20 minutes Pre/post diagnosis; review and introduction in week two

10 minutes Instructor demonstration

20 minutes Independent source reading and evidence finding

10 minutes Break

20 minutes Compare with a partner

25 minutes Record judgments and revisions

15 minutes Share, save, and wrap up

For integration into an existing class

50-minute adaptation

Introduction and demonstration: 10 minutes
Check the source: 15 minutes
Compare and revise: 20 minutes
Save: 5 minutes

Shorten the reading and separately allocate around twenty minutes each for the pre- and post-diagnostics.

Proposal for committee discussion

Source · Slide 15

Reading-performance rubric

Appendix B · Draft 0–8-point scoring criteria

Criterion	0 points	1 point	2 points
Core understanding	0: central meaning omitted or misunderstood	1: partial understanding; conditions omitted	2: central meaning and key conditions explained
Source evidence	0: absent or irrelevant evidence	1: incomplete location or content	2: correct passage and location
AI judgment	0: judgment conflicts with the source	1: correct judgment but insufficient explanation	2: judgment explained through source evidence
Revision and explanation	0: error retained or no explanation	1: partial revision with weak evidence	2: accurate revision or evidence for retaining it

For diagnostics only, add a one-sentence question on the source's main point. Score unanswered items zero; record nonparticipation separately.

Proposal for committee discussion

Source · Slide 16

Provisional AI literacy diagnostic questions

Appendix C · Provisional scenario questions

AI says it works for everyone.

The source's population, scope, and conditions should be checked first.

An AI answer includes a plausible-looking source.

The actual existence and content of the source must be checked.

You are about to upload material containing student names to AI.

Removal of identifying information and permitted use must be checked first.

Includes sample responses For instructor review · Report the three-question score separately from confidence responses · Not a standardized scale

Proposal for committee discussion

Source · Slide 17

Synthetic example: restoring conditions through revision

Appendix D · Synthetic illustrative example

Source · Paragraph 1

Brief reading activities may help some students participate in class under certain conditions.
The effect depends on prior knowledge and task difficulty in context.

Summary to check

Brief reading activities increase participation for all students.

Source evidence and judgment

Paragraph 1 says some, may help, and specifies two conditions → Partially correct

After revision

Brief reading activities may help some students participate in class, depending on prior knowledge and task difficulty.

Synthetic illustration, not an actual study result or AI output

Source · Slide 18

Sources and limits of verification

Source list

Appendix E · Supporting material

S0

Attachment: international AI-assisted reading examples

User-provided Markdown

S1

Northeastern · Checking AI summary accuracy

Official university classroom example

S2

Monash · NotebookLM as a reading tutor

Official university classroom example

S3

UCLA HumTech · AI-prompted reading reflections

Official university teaching-support activity

Verification reference date 2026.10.07. · No effectiveness figures, participating courses, headcounts, or start date were invented.

Source · Slide 19