SOURCE READING
Original and translation
The excerpt and translation use the primary source below. Further references inform SHIO’s analysis.
Primary source
- Original title
- Four Principles of Explainable Artificial Intelligence
- Original author / institution
- P. Jonathon Phillips, Carina Hahn, Peter Fontana, Amy Yates, Kristen K. Greene, David A. Broniatowski, Mark A. Przybocki / National Institute of Standards and Technology
- Primary source
- National Institute of Standards and Technology
- Source language
- English
- Original publication date
This is an excerpt from the primary source abstract, not the complete report. Chinese and Japanese translations were prepared by SHIO on October20, 2026; SHIO analysis is presented separately. The original appears once when it matches the page language. Republished courtesy of the National Institute of Standards and Technology.
This edition uses the source language, so the original appears once.
We introduce four principles for explainable artificial intelligence (AI) that comprise fundamental properties for explainable AI systems.We propose that explainable AI systems deliver accompanying evidence or reasons for outcomes and processes; provide explanations that are understandable to individual users; provide explanations that correctly reflect the system’s process for generating the output; and that a system only operates under conditions for which it was designed and when it reaches sufficient confidence in its output.We have termed these four principles as explanation, meaningful, explanation accuracy, and knowledge limits, respectively.SHIO analysis
An AI explanation should help someone ask a better question
The following is SHIO’s independent judgment and discussion of methods. It is separate from the source text and its translation.
SHIO reading note
NIST’s 2021 principles and DARPA’s completed XAI program provide historical context for designing explanations that a user can examine and challenge.
The central idea
SHIO view: explanation quality appears in what a person can do with the explanation.
What the sources establish
NIST published Four Principles of Explainable Artificial Intelligence on September 29, 2021. The report addresses explanations that users can understand, that reflect the system’s process, and that recognize limits. These are properties to examine, rather than a guarantee that any text presented as an explanation satisfies them.
DARPA’s XAI program page describes combining explainable models with interfaces that support useful explanation dialogues. It explicitly states that the program is complete and the page is no longer maintained. No original publication date is supplied. Its objectives are historical research context, not evidence of a currently available universal explanation tool.
SHIO method: start with the recipient
Before adding an explanation panel, name the person receiving it and the decision they face. A reviewer checking a source needs different information from an operator deciding whether to pause a task. Choose a question that matters in that situation, then identify the evidence the system can actually provide. Avoid suggesting that a fluent narrative reveals internal processing when that has not been established.
For a proposed document assistant, offer a link to the passage supporting a statement, identify missing evidence, and let the reviewer correct the interpretation. Ask the reviewer to use this information to find a deliberately introduced, harmless error in a test example. Observe what they can detect and where the explanation leaves them uncertain. This is a proposed local test, not a reported result.
SHIO judgment and limits
An explanation can be easy to read while failing to support a useful challenge. SHIO would therefore assess comprehension alongside the ability to identify a limit, request another source, or decline an action. The design should also let the user say that the available explanation is inadequate. Agreement with the output is not the only successful outcome.
Providing more text does not necessarily resolve uncertainty, and an explanation does not establish that the underlying answer is correct. Keep that distinction visible. Giving an idea form means building a relationship between evidence and decision in which a person can ask a sharper question and change the next step.
A question for practice
Choose one user question, then test whether the explanation helps that person challenge the result.
REFERENCES & EDITORIAL NOTE
Sources & further reading
Source excerpts, translations, and SHIO’s independent analysis are presented separately. The resources below offer additional context and further reading.
National Institute of Standards and Technology · Primary source
Four Principles of Explainable Artificial Intelligence
Original source date：Accessed：The official NIST Technical Series Publications policy states that NIST employee works are not copyright-protected in the United States and grants a worldwide, royalty-free right to reprint and create derivative works to the extent NIST asserts foreign rights. The selected original report text is attributed to NIST, not to an external quotation; no third-party tables, figures, logos or images are reproduced. The policy warns that third-party-authored works may remain protected. Attribution is retained and the Chinese/Japanese text is marked as SHIO translation dated2026-10-20. Policy/source: https://www.nist.gov/open/copyright-fair-use-and-licensing-statements-srd-data-software-and-technical-series-publications
Defense Advanced Research Projects Agency
Explainable Artificial Intelligence (XAI)
Original source date：No publication date providedAccessed：Original editorial analysis, no full translation. Official page says program complete and page no longer maintained; no original publication date or blanket article-text reuse license verified. No illustrations reproduced.