HRA Methods Compared for AI-Assisted Operations¶
Cross-method comparison of SPAR-H, IDHEAS-ECA, and ATHEANA for modelling AI-assisted operator actions.
Comparison¶
| Dimension | SPAR-H | IDHEAS-ECA | ATHEANA |
|---|---|---|---|
| Approach | PSF multipliers | Macrocognitive functions + CFMs | Error-forcing contexts |
| Quantification | Numerical (HEP) | Numerical (HEP) | Qualitative to semi-quantitative |
| AI extension path | Add PSF multipliers | Add AI-specific CFMs to teamwork function | Add AI-related EFCs |
| Dependency model | 5 discrete levels | PIF-based | Scenario-based |
| Handles 3-way conditional | No (needs structural extension) | Partially (through CFM framework) | Yes (through EFC search) |
| Commission errors | Limited | Moderate | Primary strength |
| Best suited for | Screening, quantification | Detailed cognitive analysis | Identifying dangerous scenarios |
Recommended Workflow (Report 5 §7)¶
- ATHEANA first: Identify which AI-related scenarios produce error-forcing contexts (which scenarios are dangerous?)
- IDHEAS-ECA second: Detailed cognitive failure mode analysis for the identified scenarios (how does the failure occur?)
- SPAR-H third: Quantification for PSA (what is the HEP?)
Key Finding¶
IDHEAS-ECA provides the most natural extension path because its "teamwork" macrocognitive function already models crew interaction — extending it to human-AI teaming maps onto existing structure. The distinction between cognitive failure modes and performance-influencing factors is cleaner than SPAR-H's PSF model, making it easier to add AI-specific failure modes.
All quantitative values in the walkthroughs are illustrative — no empirical nuclear AI data exists as of April 2026.