Explainable AI

What is explainable AI?

Explainable AI is a system whose outputs can be understood: which inputs were considered, which weighed the most, and why the result is what it is. NIST's AI Risk Management Framework (AI RMF 1.0) lists "explainable and interpretable" among the characteristics of trustworthy AI, and separates two questions: explainability answers "how" (the mechanism that produced the output), interpretability answers "why" (what the output means in context).

In recruiting, it means a match score is never a bare number: it comes with its components and the reasons behind them.

What is the difference between a black box and explainable AI?

Illustrative example:

• Black box: "Candidate B: 62%."
• Explainable: "Candidate B: 62%. Skills: 7 of 10 required skills (missing Kubernetes, Terraform, advanced English). Experience: 4 relevant years in cloud infrastructure, 5 requested."

The second version lets the recruiter verify, fix a resume parsing error (the candidate may list Terraform under another name), and explain a rejection to the candidate in useful terms.

Why it matters

• Verify: a recruiter cannot own a decision they do not understand.
• Correct: an explanation exposes parsing errors and badly weighted criteria.
• Detect bias: if a factor unrelated to the job weighs heavily, you can see it.
• Answer the candidate: an explanation enables honest feedback instead of an empty stock phrase.

Examples by jurisdiction

• Quebec: section 12.1 of the private sector privacy act gives a person subject to a decision based exclusively on automated processing the right to learn, on request, what information was used and the reasons and main factors and parameters behind the decision. A system that cannot produce them makes this duty hard to meet.
• Ontario: since January 1, 2026, using AI to screen, assess or select must be disclosed in public postings by employers with 25 or more employees. The law does not require explaining each score, but informed candidates will ask.
• New York City: Local Law 144 requires an annual bias audit of automated employment decision tools and publication of a summary of results.
• United States: the NIST framework is voluntary, but many organizations use it as a reference.

This summary is general information, not legal advice.

Best practices

1. Ask every vendor: "Can you explain why this candidate got this score?" If the answer is no, it is a black box.
2. Prefer job-related factors (skills, experience) and know exactly which ones enter the calculation.
3. Separate what the recruiter sees from what the candidate receives: some internal factors have no place in feedback.
4. Allow results to be challenged and corrected.
5. Review explanations regularly across many candidates, not only one by one: a pattern (for example, one factor that always drags down the same kind of profile) is often invisible in a single case.

With RecruitEasy

RecruitEasy refuses black-box mode, even for a large customer. Every match score is broken down: the recruiter sees the share of skills, experience and location, which skills are covered or missing, and which experience was judged relevant. Feedback sent to candidates, approved by a recruiter, covers only skills and experience: never location. The complete guide to AI in recruitment presents other uses.