<aside> 🌳

About this field guide

I built this field guide to make the UC Berkeley taxonomy of AI trustworthiness easier to visualize, explore, and learn from.

</aside>

How to read the tree

The 150 properties of trustworthiness are organized across seven AI lifecycle stages. Within those stages, properties are mapped to NIST AI RMF characteristics.

Properties marked with an asterisk are identified as potentially less relevant for AI systems that are not human facing.

graph TD
    A["AI Trustworthiness · 150 properties"]
    A --> S1["1. Plan and Design · 61 properties"]
    A --> S2["2. Collect and Process Data · 13 properties"]
    A --> S3["3. Build and Use Model · 14 properties"]
    A --> S4["4. Verify and Validate · 14 properties"]
    A --> S5["5. Deploy and Use · 12 properties"]
    A --> S6["6. Operate and Monitor · 21 properties"]
    A --> S7["7. Use or Impacted By · 15 properties"]

Lifecycle at a glance

Lifecycle stage	Purpose	Properties
Plan and Design	Articulate and document the system’s concept and objectives, underlying assumptions, context, and requirements.	61
Collect and Process Data	Collect and process data, including gathering, validating, cleaning, and documenting dataset metadata and characteristics.	13
Build and Use Model	Create, select, and train models or algorithms.	14
Verify and Validate	Verify and validate, calibrate, and interpret model output.	14
Deploy and Use	Pilot the system, check compatibility with legacy systems, verify regulatory compliance, manage organizational change, and evaluate user experience.	12
Operate and Monitor	Operate the AI system and continuously assess recommendations and impacts, both intended and unintended, against objectives and ethical considerations.	21
Use or Impacted By	Use the system or technology, monitor and assess its impacts, seek mitigation of impacts, and advocate for rights.	15

Explore the 150 Trustworthiness Properties

The taxonomy can also be explored as a structured property library. I built this view for readers who prefer working with structured data formats and want to compare attribute properties across lifecycle stage, trustworthiness characteristics, human facing relevance, relevant questions, and NIST AI RMF mappings.

Untitled

AI Trustworthiness Property Database

Full Taxonomy Tree Diagram 🔎

Expand each lifecycle stage, then select a trustworthiness characteristic to continue moving down the tree into its related subsections.