Part 1
AI4RA Introduction
10 min
The intersection between AI and data science
Foundations Module Preview
Part 2
Data Science & AI Intersection
10 min
Part 3
Sandbox Time
20 min
Part 4
Decision Frameworks
10 min
AI4RA Objectives
Core Pillars for AI4RA
Accuracy
Evidence-based answers grounded in validated data sources.
Reproducibility
Repeatable workflows with transparent prompts, data, and methods.
Flexibility
Support for multiple institutional contexts, teams, and use cases.
Security
Policy-aligned controls for access, privacy, and operational risk.
Community of Practice Ecosystem
Vandalizer
AI workflow interface for structured analysis and reproducible outputs.
Data Lakehouse
Shared institutional data foundation to support trusted analytics at scale.
AI Literacy Course
Frameworks and companions for building institutional AI fluency across research administration.
Other Projects
Additional initiatives that extend AI4RA methods across research administration.
AI4RA: The Bridge to Research Analytics
FAIR Principles as Operational Guardrails
Use persistent identifiers and searchable metadata.
Provide clear retrieval pathways with governed permissions.
Adopt shared formats and vocabularies across systems.
Document provenance, context, and usage constraints.
The Intersection of Data Science and Artificial Intelligence
Shift 1
Proactive Data Science
A Shift in Our Reach
Agentic coding and development tools make building new systems attainable in a short period of time.
Traditional data science backgrounds are well-aligned with advances in artificial intelligence.
Endless opportunities and development trees can lead to “Brain Fry.”
The Intersection of Data Science and Artificial Intelligence
Shift 2
Reactive Data Science
A Shift in Stakeholder Values
Stakeholders now expect faster, more effective solutions powered by AI.
Perceived competition and fear of getting left behind drives initiatives.
Less appeal to dashboards, more appeal to chatbot interfaces.
Build a sentence one token at a time
Click the next word. Each choice shifts what comes next, just like an LLM.
Simplify the Approach
Step 1
Identify a black hole
Spot well-defined tasks that are creating unnecessarily large bottlenecks.
Step 2
Classify the task
Can data science or artificial intelligence be used to automate or collaborate?
Step 3
Build a simple solution
The solution must be less intimidating and complex than the original task.
Stepping Into the Vandalizer
Jump straight into the University of Idaho’s AI prompt workspace.
Launch Application
Open Vandalizer
Prompt Playground
UI Research Support
Fast Iteration
Exercise 1: Simple Extraction
Challenges
Exercise 2: Repeated Extractions
Sample NSF Award 1
NSF award to PI Radagast Brownleaf
Sample NSF Award 2
NSF award to PI Juniper Quillstone
Sample NSF Award 3
NSF award to PI RJ MacReady
Challenges
Exercise 3: Multi-Document Interactions
Sample Budget
Grant proposal budget for Frodo B. Underhill
Sample Budget Justification
Grant proposal budget justification for Frodo B. Underhill
Sample Research Strategy
Grant proposal Research Strategy for Frodo B. Underhill
Challenges
Who owns AI-generated content?
Research Analytic Accountability
AI has no integrity, no professional ethics, no accountability. Someone with professional judgment must always own the final call.
Does it matter how the work got done?
- Human leads — Every step must be auditable. AI may assist, but a human owns the reasoning.
- Human in the loop — AI prepares, a qualified person reviews and decides.
- AI leads — Only the result matters. AI can lead; validate with spot-checks.
Classify these analytic tasks
Drag each slider. What do you think?
Faculty Workload Modeling
ROI Analysis
Financial Projections
Compliance Cycle-Time
Impact Reporting
Competition Intelligence Dashboards
Human Leads
Human in the Loop
AI Leads
Data Science Tasks vs. AI Tasks: Then and Now
BEFORE
Careful distinction between tasks that are well-suited for “data science work” or “AI work.”
AFTER
Agentic workflows call specialized tools that are tailored to the target task
TRUE PRINCIPLE
Task fit matters for optimized results from artificial intelligence
From Prompt Engineering to Intent Engineering
BEFORE
Significant emphasis placed on crafting effective prompts to optimize the output
AFTER
Better processes reduce the need for perfecting prompts with large, mainstream models
TRUE PRINCIPLE
Prompt design helps users get output that aligns with their intent
Four disciplines for AI interaction
Your AI has a page limit — the context window
Intent
Telosa
Information
Mnemos
Instructions
Promptulus
Conversation
Dialogos
The AI Literacy Companions
Sequita
Auditability
Modulus
Decomposition
Telosa
Intent
Promptulus
Prompts
Mnemos
Context
Dialogos
Conversation
Vitrea
Transparency
Veridex
Evaluation
Clarion
Reporting
Amid Constant Change, Data Foundations Stay Essential
✅ Quality
Validated, timely, and complete records.
⚖️ Governance
Clear ownership, policy alignment, and accountability.
🔐 Security
Access controls, monitoring, and incident response plans.
📝 Documentation
Lineage, assumptions, and known limitations.