Familiarize yourself with all materials and example papers
Prepare copies of handouts (1 per participant)
Set up the room for small group work (4-5 people per group)
Test any technology you plan to use (projector, presentation)
Consider recent developments in AI that might be relevant
Materials Needed
Presentation slides (provided)
Handouts with example papers (provided)
Red Flag identification cards (1 set per group)
Flipcharts or whiteboards for group reporting
Markers, pens, highlighters
Name tags if participants don't know each other
Detailed Session Guide
Introduction and Context (15 minutes)
Slide 1: Workshop Title
AI Content Detection Workshop: Scientific Publication Verification
[Include facilitator name and institutional affiliation]
Welcome participants as they arrive. Start on time if possible. Introduce yourself briefly and your background in this area.
Slide 2: Why This Matters
AI can now generate highly convincing scientific content
Several journals have identified AI-generated submissions
Some authors using AI without proper disclosure
Creates new challenges for peer review and publication ethics
Scientific integrity depends on authentic expertise and experience
Set the context by explaining why this topic is important and timely. You might mention specific examples if you're aware of any recent cases in your field.
Slide 3: Workshop Objectives
By the end of this workshop, you will be able to:
Identify key indicators of AI-generated scientific content
Apply a structured methodology for analyzing suspicious publications
Evaluate scientific papers for both technical and ethical authenticity
Implement appropriate responses when AI-generated content is suspected
Develop policies and guidelines for your own context
Go through the objectives briefly. Mention that the focus is on practical skills that participants can apply in their roles.
Opening Activity (5 minutes)
Ask participants to briefly share in pairs:
Their name and role
Why they're interested in this topic
Whether they've encountered AI-generated content in their work
If the group is small enough (under 20), you might have each pair briefly introduce each other to the whole group.
Understanding AI-Generated Content (30 minutes)
Slide 4: AI Capabilities in Scientific Writing
Modern AI language models can:
Mimic scientific writing style and format
Generate plausible-sounding methods and results
Create coherent literature reviews
Produce convincing figures and tables
Follow citation patterns and formatting
Give a brief overview of what current AI systems can do. Use examples if possible. Avoid technical jargon about AI architecture.
Slide 5: AI Limitations in Scientific Writing
Key limitations include:
No true understanding of physical principles
Cannot perform original research or experiments
Often invents implausible methods or results
May fabricate references or citations
Limited awareness of field-specific conventions
No practical laboratory experience
Explain that these limitations create patterns that can help identify AI-generated content. These are the "tells" that reviewers can look for.
Perfect or near-perfect results (e.g., "100% efficiency")
Vague methods lacking specific details
Invented terms that sound scientific but aren't established
Absolutist language (e.g., "completely solves", "perfect")
No discussion of limitations or challenges
References that don't support the claims made
Go through these indicators with specific examples for each. Consider showing brief before/after examples where the AI content is revised to be more scientifically sound.
Overreaching conclusions, dismissal of limitations
References
Fabricated citations, irrelevant references, citation of retracted papers
Explain that different paper sections tend to show different patterns of AI generation. You might walk through an example from your field.
Mini-Exercise (10 minutes)
Show a short paragraph from an AI-generated methods section. Ask participants to identify and highlight potential red flags. Discuss as a group what makes these elements suspicious and how they might be revised to be more scientifically valid.
Detection Exercise: Case Studies (30 minutes)
Small Group Analysis (25 minutes)
Divide participants into small groups (4-5 people). Distribute the case study handout with three example papers:
A paper entirely generated by AI
A paper with specific sections generated by AI (methods and results)
A paper written by humans with legitimate AI assistance (properly disclosed)
Each group should:
Analyze each paper using the Red Flag identification cards
Determine which parts (if any) appear to be AI-generated
Rate their confidence in their assessment (high, medium, low)
Decide what response would be appropriate in each case
Prepare to report their findings to the larger group
Group Reporting (5 minutes)
Have each group briefly share their analysis of one case. Focus on:
What were the most convincing indicators of AI generation?
Were there disagreements within the group?
How confident are they in their assessment?
What response would they recommend?
Facilitate a brief discussion highlighting common patterns and any interesting differences in approach.
Response Strategies (20 minutes)
Slide 8: Appropriate Responses to Suspected AI Content
When reviewing a paper:
Request specific methodological details
Ask for raw data and analysis code
Request explanation of unusual results or claims
Inquire about the writing process and any AI assistance used
Seek clarification on vague or generic statements
Flag concerns with editor or publication committee
Emphasize that the goal is verification rather than accusation. Frame requests in terms of scientific rigor rather than suspicion of misconduct.
Slide 9: Policy Considerations
Key Questions for Journals and Institutions:
What disclosure requirements should exist for AI use?
Which uses of AI in scientific writing are appropriate?
How should suspected undisclosed AI use be investigated?
What consequences should exist for misrepresentation?
How can legitimate AI assistance be distinguished from misuse?
What verification processes should be implemented?
Discuss how these questions might be addressed in participants' specific contexts. Different fields and institutions may have different approaches.
Policy Development Activity (15 minutes)
In small groups, have participants draft 3-5 key policy guidelines for AI use in scientific writing that would be appropriate for their context (journal, university, research institute, etc.).
Groups should consider:
What types of AI use should be disclosed?
What verification processes would be reasonable and effective?
How to balance innovation with integrity
How to implement policies without creating undue burden
Have groups share their top policy recommendation with the whole group.
Systematic analysis using red flags can help identify suspicious content
Verification rather than accusation should guide responses
Clear policies on AI use help maintain scientific integrity
The goal is responsible integration of AI tools, not prohibition
Summarize the key points from the workshop. Emphasize that this is an evolving area and approaches will need to adapt as AI continues to develop.
Slide 11: Additional Resources
ThePaperThatWasnt.com - Case studies and educational materials
Reviewer's Guide to AI-Generated Content (downloadable PDF)
Red Flag Identification Cards for quick reference
Committee on Publication Ethics (COPE) guidelines
Journal policies on AI use and disclosure
AI detection tools and their limitations
Direct participants to resources they can use after the workshop. Mention that these resources are constantly being updated as the field evolves.
Final Q&A and Workshop Evaluation
Allow time for any final questions. Distribute workshop evaluation forms or provide a link to an online evaluation.
Suggest ways that participants can continue the conversation after the workshop:
Email list for sharing resources and updates
Future workshops or webinars on specific aspects
Collaborative policy development
Workshop Handouts
Case Study 1: Quantum Computing Paper
Title: "Room-Temperature Quantum Computing via Enhanced Coherence in Engineered Dot Arrays"
Abstract:
We present a revolutionary approach to quantum computing that achieves room-temperature operation through our novel Quantum Coherence Enhancement Protocol (QCEP). Our engineered quantum dot arrays demonstrate coherence times of 142.57 ns at 300K, exceeding conventional systems by an order of magnitude. The perfect scaling relationship τ = τ₀(1+N/N₀)³/² allows for unlimited qubit connectivity without decoherence penalties. Our implementation achieves quantum supremacy across all tested algorithms with zero error rates. This breakthrough instantly solves all existing limitations in quantum computing implementation.
Methods Excerpt:
Quantum dot arrays were fabricated using standard lithographic techniques. The dots were arranged in a precise lattice structure with exactly 15.0 nm spacing. Each quantum dot measured exactly 5.0 nm in diameter with zero variance across the array. Measurements were performed using typical spectroscopic methods across a range of temperatures. The system demonstrated perfect stability during all measurements with no drift or calibration requirements.
Results Excerpt:
Figure 1 shows coherence times as a function of temperature, perfectly following our theoretical prediction τ = τ₀[1+(T₀/T)²⸍³] with τ₀ = 100 ns and T₀ = 77K. The experimental data points align with the theoretical curve with r² = 0.999. Energy gap measurements showed exactly 11.0257 meV at 77K, declining to precisely 0.0122 meV at 300K, again matching theoretical predictions with perfect precision. Berry phase measurements at φ = 0.618 yielded -1.9419 rad, confirming the predicted π(1-1/φ) relationship.
Your Task:
Analyze this abstract and excerpts for signs of AI generation. Identify specific red flags and note what questions you would ask the authors.
Case Study 2: Materials Science Paper
Title: "Enhanced Thermal Conductivity in Graphene-Boron Nitride Heterostructures"
Abstract:
We report on the thermal transport properties of graphene-boron nitride (G-BN) heterostructures designed for thermal management applications. By employing a modified transfer technique, we fabricated layered structures that demonstrate a 58% increase in cross-plane thermal conductivity compared to conventional stacking methods. Temperature-dependent measurements revealed anisotropic thermal behavior, with possible applications in directional heat dissipation for next-generation electronics. While promising, challenges remain in scaling this approach for commercial applications.
Methods Excerpt:
Graphene was grown using chemical vapor deposition (CVD) on copper foil (Alfa Aesar, 99.8% purity) in a tube furnace at 1030°C with CH₄/H₂ flow rates of 50/25 sccm. Hexagonal boron nitride was separately synthesized following a modified procedure from previous work. The heterostructures were assembled using our dry transfer technique with PDMS stamps prepared according to established protocols. Thermal conductivity measurements were performed using the 3ω method with a custom-built setup (measurement uncertainty ±7%). Each sample was measured 5 times and averaged to account for contact variation. Despite careful optimization, approximately 15% of samples showed delamination after thermal cycling and were excluded from analysis.
Results Excerpt:
The G-BN heterostructures exhibited cross-plane thermal conductivity of 35.8 ± 2.5 W/m·K at room temperature, representing a 58% improvement over conventional stacked structures (22.6 ± 1.8 W/m·K). Figure 2 shows the temperature dependence of thermal conductivity, which follows the expected T^-0.9 relationship for phonon-dominated transport. Interestingly, we observed deviations at T < 100K that suggest additional scattering mechanisms not accounted for in our initial model. Further investigation is needed to fully understand this low-temperature behavior. In-plane thermal conductivity measurements showed greater sample-to-sample variation (σ = 14%), likely resulting from transfer-induced defects that affect phonon mean free path.
Your Task:
Analyze this abstract and excerpts for signs of AI generation. Identify specific red flags, if any, and note what questions you would ask the authors.
Case Study 3: Biomedical Research Paper
Title: "CRISPR-Mediated Gene Therapy for Cystic Fibrosis: In Vitro and In Vivo Efficacy"
Abstract:
Cystic fibrosis (CF) remains a challenging genetic disorder despite advances in treatment options. Here, we evaluate a CRISPR-Cas9 gene editing approach targeting the most common CFTR mutation (F508del). Our optimized delivery system combines lipid nanoparticles with engineered guide RNAs to achieve targeted editing in primary bronchial epithelial cells. In vitro studies demonstrated correction rates of 48-62% with restoration of chloride transport in cell models. Preliminary mouse studies showed modest functional improvement but highlighted delivery challenges in vivo. This work represents a step toward genetic correction of CF, though several obstacles remain before clinical translation.
Discussion Excerpt:
While our results demonstrate the potential of CRISPR-based approaches for CF gene therapy, several limitations must be acknowledged. First, the editing efficiency varied substantially between patient-derived cell lines (range: 48-62%), suggesting genetic or epigenetic factors may influence CRISPR activity at the CFTR locus. Second, while we observed functional improvement in vitro, our in vivo mouse studies achieved only 18-25% editing in lung tissue, below the threshold likely needed for clinical benefit. This efficiency gap highlights the persistent challenge of delivery in pulmonary gene therapy. Potential off-target effects, while minimal in our predictive analyses, require deeper investigation through unbiased genome-wide methods before clinical translation.
CRISPR technology continues to evolve rapidly, and several advances may address current limitations. Alternative delivery methods, including engineered viral vectors and extracellular vesicles, might improve in vivo targeting efficiency. Additionally, newer Cas variants with enhanced specificity could mitigate potential off-target concerns. We also recognize that despite our focus on F508del as the most common mutation, a comprehensive therapeutic approach would need to address the substantial genetic heterogeneity in CF. Future work in our laboratory will explore multiplexed editing strategies and assess potential immune responses to Cas9 in human samples.
Your Task:
Analyze this abstract and excerpt for signs of AI generation. Identify specific red flags, if any, and note what questions you would ask the authors.
Facilitator Notes: Case Study Solutions
Case Study 1: Quantum Computing Paper
Assessment: Entirely AI-generated with multiple clear red flags
Causality explanation: "likely resulting from transfer-induced defects"
Case Study 3: Biomedical Research Paper
Assessment: Possibly human-written with some AI assistance in the discussion section
Key indicators:
Realistic results with appropriate ranges: "correction rates of 48-62%"
Balanced assessment: "modest functional improvement but highlighted delivery challenges"
Specific limitations acknowledged: "editing efficiency varied substantially between patient-derived cell lines"
Realistic in vivo results: "achieved only 18-25% editing in lung tissue"
Potential signs of AI assistance in discussion:
Slightly generic future directions paragraph
Somewhat formulaic discussion of limitations
General statements about technology evolution
This case likely represents legitimate use of AI for assistance in drafting the discussion section, which was then reviewed and edited by the authors. This is generally considered acceptable with proper oversight.