Authenticity of student work in the age of genertive AI

1. Use Process-Based Evidence (Most Reliable)

These methods verify how the work was produced, not just the final text.

1. Version history

   • Draftback -  Place in your 610 folder on your Google drive and make Dr. Herr (csun.science.education@gmail.com) an editor so he can observe your revision history in real time using Draftback.  

   • Use Google Docs / Word Track Changes to review writing history.

   • Look for continuous writing versus a single large paste.

2. Draft submissions

   • Require outlines, rough drafts, and revisions.

   • Compare evolution of ideas, errors, and voice across versions.

3. Writing logs or reflection memos

   • Ask students to briefly explain:

     • Their argument

     • Key sources

     • Why they made specific revisions

4. In-class writing checkpoints

   • Short handwritten or timed writing connected to the larger assignment.

   • Compare style, vocabulary, and reasoning to the final paper.

2. Oral Verification (Highly Effective)

1. Brief oral defense

   • Ask students to explain their thesis, sources, or reasoning.

   • AI-generated work is often poorly understood by the student.
     
     

2. Targeted follow-up questions

   • Ask about:

     • Why a specific example was chosen

     • How a conclusion was reached

     • What alternative arguments were rejected
       
       

3. Mini-conferences

   • 5–10 minute meetings can quickly reveal authorship.
     
     

3. Assignment Design That Discourages AI Pasting

1. Personalized or localized prompts

   • Require connections to:

     • Class discussions

     • Specific lectures

     • Local data or experiences

   • AI struggles with non-public, course-specific content

2. Source-based constraints

   • Require engagement with:

     • Assigned readings

     • Page-number citations

     • Course-only materials

3. Multi-stage assignments

   • Proposal → annotated bibliography → draft → final

   • Makes outsourcing or pasting far more difficult.

4. Creative or analytical transformations

   • Ask students to:

     • Critique an argument

     • Apply a theory to a novel case

     • Compare multiple interpretations
       
       

4. Stylistic & Consistency Checks (Human Judgment)

1. Sudden shifts in voice

   • Unexplained jumps in sophistication, tone, or vocabulary.

2. Generic structure

   • Overly polished but vague prose

   • Predictable paragraph patterns and safe conclusions

3. Lack of idiosyncrasies

   • No minor errors, personal phrasing, or risk-taking

   • AI text is often “too smooth”

4. Inconsistent citation behavior

   • Perfect formatting paired with shallow engagement

   • Citations that don’t clearly support claims
     
     

5. AI Detection Tools (Use with Caution)

1. AI detection software

   • Examples: Turnitin AI, GPTZero , Copyleaks

     1. Samples: 

• Chapter 5 - Scientific Methods - Sourcebook for Teaching Science - written by Norm Herr

• James Webb Space Telescope - Written by ChatGPT

1. • Treat results as signals, not proof.

2. Cross-check with plagiarism tools

   • AI text may include lightly paraphrased or fabricated content.

3. False-positive awareness

   • Non-native speakers, formulaic writing, and strong students can be flagged incorrectly.

6. Clear Policy & Transparency (Preventive)

1. Explicit AI-use policy

   • Define what is allowed (e.g., brainstorming vs. drafting).

   • Require disclosure of AI assistance.

2. Honor statements

   • Ask students to affirm compliance with course policy.

3. Explain why authenticity matters

   • Emphasize skill development, not punishment.