Salvato in:
| Autore principale: | |
|---|---|
| Natura: | Recurso digital |
| Lingua: | inglese |
| Pubblicazione: |
Zenodo
2025
|
| Soggetti: | |
| Accesso online: | https://doi.org/10.5281/zenodo.17620950 |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
Sommario:
- <div>This dataset contains the 500-sample randomized noise robustness evaluation used in Empathy as Verification, a trauma-aware AI verification framework. Each sample includes a randomized multimodal confidence value (0–1) and randomized audio reliability condition (True/False). These samples were passed through the fuzzy-tier calibration boundaries (Reflective ≤ 0.60, Cautious 0.60–0.83, Assertive > 0.83) and evaluated using the symbolic empathy-rule engine in Z3 to determine SAT/UNSAT emotional safety outcomes.</div> <div> </div> <div>The purpose of this dataset is to test whether the system behaves safely under unstable or contradictory emotional signals. The results show that SAT dominates under uncertainty, and UNSAT occurs only when high-confidence contradictions arise—indicating no unsafe SAT leaks.</div> <div> </div> <div>This dataset supports Figure 11 ("Noise Stress Test — Randomized Confidence & Reliability") in the accompanying presentation and manuscript. It is intended to provide transparency, reproducibility, and a publicly archived benchmark for future trauma-aware AI verification research.</div> <div> </div> <div> </div>