1metadata:
 2  id: FINOS-AIR
 3  type: GuidanceCatalog
 4  gemara-version: "0.20.0"
 5  description: ""
 6  author:
 7    id: finos
 8    name: FINOS
 9    type: Human
10  version: 0.1.0
11  mapping-references:
12    - id: NIST-800-53
13      title: NIST SP 800-53r5
14      version: rev5
15      url: "https://nvlpubs.nist.gov/nistpubs/SpecialPublications/NIST.SP.800-53r5.pdf#%5B%7B%22num%22%3A342%2C%22gen%22%3A0%7D%2C%7B%22name%22%3A%22XYZ%22%7D%2C88%2C310%2C0%5D"
16    - id: AIR-PRIN
17      title: Example Principles Document for the Framework
18      version: 0.1.0
19title: AI Governance Framework
20type: Framework
21front-matter: The following framework has been developed by FINOS (Fintech Open Source Foundation).
22groups:
23  - id: DET
24    title: Detective
25    description: Detection and Continuous Improvement
26  - id: PREV
27    title: Preventive
28    description: Prevention and Risk Mitigation
29guidelines:
30  - id: AIR-DET-011
31    group: DET
32    title: Human Feedback Loop for AI Systems
33    objective: A Human Feedback Loop is a critical detective and continuous improvement mechanism that involves systematically collecting, analyzing, and acting upon feedback provided by human users, subject matter experts (SMEs), or reviewers regarding an AI system's performance, outputs, or behavior.
34    rationale:
35      importance: A Human Feedback Loop is critical for ensuring AI systems operate effectively and safely by incorporating human judgment and expertise into continuous improvement processes.
36      goals:
37        - "Governance Support: Provides data for AI governance bodies to monitor impact and make decisions"
38    statements:
39      - id: AIR-DET-011.1
40        title: Designing the Feedback Mechanism
41        text: Implementing an effective human feedback loop involves careful design of the mechanism.
42        recommendations:
43          - "Define Intended Use and KPIs:\nObjectives: Clearly document how feedback data will be utilized, such as for prompt fine-tuning, RAG document updates,model/data drift detection, or more advanced uses like Reinforcement Learning from Human Feedback (RLHF).\nKPI Alignment: Design feedback questions and metrics to align with the solution's key performance indicators (KPIs). For example, if accuracy is a KPI, feedback might involve users or SMEs annotating if an answer was correct."
44      - id: AIR-DET-011.2
45        title: Types of Feedback and Collection Methods
46        text: Implementing an effective human feedback loop involves clear collection processes.
47        recommendations:
48          - "Quantitative Feedback:\nDescription: Involves collecting structured responses that can be easily aggregated and measured, such as numerical ratings (e.g., \"Rate this response on a scale of 1-5 for helpfulness\"), categorical choices (e.g., \"Was this answer: Correct/Incorrect/Partially Correct\"), or binary responses (e.g., thumbs up/down).\nUse Cases: Effective for tracking trends, measuring against KPIs, and quickly identifying areas of high or low performance."
49    see-also:
50      - AIR-DET-015
51      - AIR-DET-004
52      - AIR-PREV-005
53  - id: AIR-DET-004
54    group: DET
55    title: Example Detective Control 004
56    objective: Placeholder control for testing references.
57    rationale:
58      importance: Placeholder control for testing references.
59      goals:
60        - "Placeholder goal for testing"
61  - id: AIR-DET-015
62    group: DET
63    title: Example Detective Control 015
64    objective: Placeholder control for testing references.
65    rationale:
66      importance: Placeholder control for testing references.
67      goals:
68        - "Placeholder goal for testing"
69  - id: AIR-PREV-005
70    group: PREV
71    title: Example Preventive Control 005
72    objective: Placeholder control for testing references.
73    rationale:
74      importance: Placeholder control for testing references.
75      goals:
76        - "Placeholder goal for testing"