AIRS & Appropriate Reliance Research

AIRS & Appropriate Reliance Research

Domain knowledge for AI adoption measurement, psychometric instrument development, and appropriate reliance research

This skill contains knowledge about the AIRS-16 validated instrument, the proposed AIRS-18 extension with Appropriate Reliance (AR), and research methodologies for studying AI adoption and human-AI collaboration.

When to Use

• Discussing AIRS-16 or AIRS-18 instruments
• Developing or extending psychometric scales
• Analyzing AI adoption patterns
• Researching appropriate reliance / trust calibration
• Preparing academic papers or research briefs
• Meeting preparation with researchers

AIRS-16: AI Readiness Scale

Source: Correa, F. (2025). Doctoral dissertation, Touro University Worldwide.

Production: airs.correax.com | Time: 5 minutes | Built by: Alex Cognitive Architecture

Validation: N=523, CFI=.975, TLI=.960, RMSEA=.053, R²=.852

Quick Links

User Roles

8 Constructs (2 items each)

Key Finding: What Actually Predicts AI Adoption

Insight: Traditional UTAUT2 predictors (PE, EE, FC, HB) do NOT predict AI adoption. Value perception, enjoyment, and social influence matter.

Scoring & Typology

# AIRS Score = sum of 8 construct means (range: 8-40)
AIRS = PE + EE + SI + FC + HM + PV + HB + TR

# Typology (94.5% accuracy)
if AIRS <= 20: "AI Skeptic"      # 17% of sample
elif AIRS <= 30: "Moderate User"  # 67% of sample
else: "AI Enthusiast"             # 16% of sample

Appropriate Reliance (AR): Proposed AIRS-18 Extension

The Research Question

Is it not how much you trust AI that predicts adoption, but how well your trust is calibrated to actual AI capability?

Why AR ≠ Trust (TR)

Key distinction: TR asks "Do you trust AI?" — AR asks "Can you discern when trust is warranted?"

The 2×2 Independence Matrix

Proposed AR Items

CAIR/CSR Framework (Schemmer et al., 2023)

Metric: Appropriateness of Reliance (AoR) = 1 indicates optimal calibration.

Research Hypotheses for AIRS-18 Validation

Psychometric Standards

Reliability Thresholds

Model Fit Indices

Discriminant Validity

Intervention Strategies by Typology

Key References

Troubleshooting

"Is AR just measuring AI experience?"

Problem: Concern that AR conflates with general AI familiarity.

Solution:

• Include experience as covariate
• Test discriminant validity (HTMT < .85)
• AR should predict beyond experience level

"Can self-reported calibration be valid?"

Problem: People may not accurately assess their own calibration ability.

Solution:

• Self-report measures perceived calibration
• Future research: correlate with behavioral CAIR/CSR in task studies
• Perceived calibration may still predict adoption intentions

"Why was Trust marginal in AIRS-16?"

Possible explanations:

1. Trust level alone is insufficient — calibration matters more
2. Trust may be necessary but not sufficient
3. TR × AR interaction: trust only helps when calibrated
4. Sample characteristics (tech-savvy population)

Practical Application Modules

Project AI Readiness Assessment

Evaluate a project for AI integration readiness using AIRS-weighted dimensions:

Project_Readiness = (PV_score × 2.0) + (EE_score × 1.5) + (PE_score × 1.2) + (HM_score × 0.8) + (SI_score × 0.5)
Max = 30 points

Session Reliance Calibration

Over-reliance signals: Accepting all suggestions without edits, not verifying AI code, "just do it" on critical tasks. Under-reliance signals: Ignoring suggestions, manually typing generated code, rejecting help before evaluating.

Calibration interventions:

• Over-reliance: "I notice you're trusting my outputs quickly. For this critical task, would you like to review together?"
• Under-reliance: "I see you're preferring manual work. I could help with [specific subtask] — want a hybrid approach?"

Enterprise Deployment Readiness

Self-Monitoring Metrics

Activation Patterns