Skip to content

AIH metric for ordinal harm assessment. Experiments performed on AIAAIC dataset as a benchmark - final preprocessed data are given in the "results" folder.

License

Notifications You must be signed in to change notification settings

Datalab-AUTH/AIH

Repository files navigation

AI Harmonics

A human-centric, harm-severity-adaptive framework for AI risk assessment.
This repository contains the Streamlit apps and supporting code for the paper:

AI Harmonics: a human-centric and harms severity-adaptive AI risks assessment framework
Sofia Vei, Paolo Giudici, Pavlos Sermpezis, Athena Vakali, Adelaide Berardinelli (2025).


Overview

Contemporary AI risk assessments often rely on internal, compliance-driven checklists that overlook the lived experiences of those harmed by deployed AI systems. AI Harmonics shifts this paradigm by:

  1. Embedding human perspectives

    • Leverages 816 expert-annotated real-world AI incident reports (AIAAIC dataset)
    • Captures harms across Categories, Subcategories and Stakeholder groups
  2. Ordinal, data-driven prioritization

    • Introduces the AI Harmonics (AIH) metric: an ordinal “pseudo-Gini” measuring how concentrated harms are among the most affected
    • Benchmarks against the Criticality Index (CI) to validate ordinal consistency
  3. Robust, adaptive analysis

    • Sensitivity studies under random severity permutations and annotation removal
    • Boundary analysis for extreme “best-case”/“worst-case” harm concentration


Data

  • AIAAIC dataset

    • 816 expert-annotated AI incidents (March–April 2024)
    • Mapped to:
      • Harm Categories & Subcategories
      • Stakeholder groups (e.g., Users, Vulnerable Groups, Businesses…)
  • Processed CSVs

    • In data/results/:
      • Frequencies of each (category, stakeholder) pair

    Heatmap of incident frequencies by category and stakeholder
    Figure 1. AI incidents by harm category and stakeholder group.

- Ready for metric computation and visualization

Metrics

AI Harmonics (AIH)

An ordinal “pseudo-Gini” measuring how harms in category i concentrate among the most severely affected stakeholders.
Higher AIH ⇒ stricter concentration ⇒ higher mitigation priority.

AIH_i = ∫₀¹ ℓ_i(x) dx
    

Criticality Index (CI)

A benchmark ordinal metric (mean of cumulative ranks).
Under purely ordinal inputs, CI and AIH are provably linearly aligned, providing mutual validation.


About

AIH metric for ordinal harm assessment. Experiments performed on AIAAIC dataset as a benchmark - final preprocessed data are given in the "results" folder.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published