R0044/2026-03-29/Q004/SRC01/E01¶
CaTE Guidebook provides recommendations on discussing, designing, and integrating trust, trustworthiness, calibrated trust, and ethics.
URL: https://www.sei.cmu.edu/documents/6226/CaTE_Guidebook_8gwzU7B.pdf
Extract¶
The CaTE Guidebook "provides recommendations and observations on how to discuss, design, and integrate trust, trustworthiness, calibrated trust, and ethics based on emerging practices, frameworks, metrics, measures, formats, and tools."
Companion guides include: - "Human-Centric, Teaming-Focused Approach for Design and Development of Non-Deterministic Systems: A Human Machine Teaming Design Framework" - "Human Systems Integration Test and Evaluation of Artificial Intelligence Enabled Capabilities: What to Consider in a Test & Evaluation Strategy"
The guidebook's scope includes TEVV (Test, Evaluation, Verification, and Validation) of LAWS (Lethal Autonomous Weapon Systems), placing it in the highest-stakes military AI context.
CaTE defines calibrated trust as "a process by which human interactions with machine-automation strive to achieve an ideal state in which the human places an appropriate amount of trust in machine intelligence based on its strengths and weaknesses." This definition is explicitly human-focused — the "calibration" is of the human's trust, not of the system's behavior.
JUDGMENT: The CaTE Guidebook addresses system design properties (trustworthiness dimensions) as inputs to human trust calibration, but does not address AI systems adjusting their own output behavior. The framework is measurement-oriented: "providing evidence for assurance and developing measures to determine calibrated levels of trust."
Relevance to Hypotheses¶
| Hypothesis | Relationship | Strength |
|---|---|---|
| H1 | Contradicts | No evidence of system output behavior addressed |
| H2 | Supports | Strong human-focused framing |
| H3 | Supports | Addresses system design properties (trustworthiness) without addressing system output behavior |
Context¶
The CaTE Guidebook could not be fully extracted due to PDF encoding issues. This analysis relies on publication metadata and descriptions from SEI and CMU press materials.