Reliability
The ability of an AI system to perform consistently and correctly under specified conditions over time.
How the laws define this
Both instruments agree that robust operation is a core element of reliability. They diverge, however, in how they scope and evaluate system performance. The Artificial Intelligence (AI) Ethics Guidelines for Digital Thailand view reliability as a degree of accurate and predictable performance sustained over time and across contexts. This framework mandates active operational safeguards, requiring testing, validation, monitoring, and mitigation measures to reduce failures and maintain safe operation. Conversely, the NATO Principles for Responsible Use of AI in Defence define reliability through clear deployment boundaries and broader operational qualities. Rather than detailing continuous testing or cross-context accuracy, NATO requires applications to have explicit, well-defined use cases, and mandates that systems be developed and used securely and effectively. While Digital Thailand focuses on ongoing failure reduction and safety, NATO anchors reliability to defined usage, security, and effectiveness.
Synthesised from the 2 statutory definitions below. Each one is quoted in full, with its source.
How jurisdictions differ
What is distinctive in each definition, binding law in force first.
| Jurisdiction | Instrument | Status | What is distinctive |
|---|---|---|---|
| NATO | NATO Principles for Responsible Use of AI in Defence | In Force | Limits reliability to having explicit, well-defined use cases and being developed and used in a robust, secure, and effective manner, without addressing performance accuracy or testing. |
| Thailand | Artificial Intelligence (AI) Ethics Guidelines for Digital Thailand | Adopted | Adds assessment through testing, validation, monitoring, and mitigation measures to ensure accurate, robust, and predictable performance over time and contexts. |
What it triggers
- NATO Principles for Responsible Use of AI in Defence set out that AI applications will have explicit, well-defined use cases and be developed and used in a robust, secure, and effective manner.source
- Thailand Artificial Intelligence (AI) Ethics Guidelines for Digital Thailand call for testing, validation, monitoring and mitigation measures to reduce failures and maintain safe operation.source
Definitions (2)
The degree to which an AI system performs accurately, robustly and predictably over time and across contexts, assessed through testing, validation, monitoring and mitigation measures to reduce failures and maintain safe operation.
AI applications will have explicit, well-defined use cases and be developed and used in a robust, secure, and effective manner.
Not to be confused with
Robustness focuses on maintaining performance and resisting failures or attacks across the lifecycle, while reliability emphasizes consistent and correct performance over time under specified conditions.
Resilience stresses recovery from failures and adaptation under stress, whereas reliability centers on consistent correct performance over time.
Questions and answers
What does NATO add to AI reliability in defense?
NATO describes reliability as explicit, well-defined use cases and robust, secure, effective development and use. It guides ethical and safe AI in defense.
How does Thailand assess AI system reliability?
Thailand describes reliability as accurate, robust, predictable performance over time and contexts, assessed via testing, validation, monitoring, and mitigation to reduce failures.
How does reliability differ from robustness in AI?
Reliability is about consistent correct performance over time under specified conditions. Robustness is about resisting failures, attacks, or unexpected conditions across the lifecycle.
Related Terms
Robustness
The ability of an AI system to maintain acceptable performance and resist failures, attacks, or unexpected conditions across its lifecycle....
Accuracy
Measure of how close an AI system’s outputs are to true or accepted values; usually expressed by statistical metrics (e.g., TP/TN/FP/FN-based rates)....
Resilience
An AI system's ability to recover from failures, adapt to changing conditions, and maintain acceptable operation under stress....