Lesson 12 / 26

Transparency and Explainability

Tell people when AI is used and give explanations that are honest and useful.

By · 15 years in web development

Say it, and explain it

Transparency means telling people when they are interacting with AI or when AI contributed to a decision about them, what data it uses, and what its limits are. Explainability means being able to give a reason for an outcome that a person can understand and challenge: which factors mattered, and what could change the result. For decisions that affect people significantly, offer a way to contest the outcome and reach a human. Over-claiming ("our AI is unbiased") is itself an ethical problem; say what you tested and what you do not know.

A clear notice and a good explanation

Compare the vague version with the useful one. The useful explanation names factors and a route to appeal.

Notice:       "You are chatting with an AI assistant. It can make mistakes.
               A human agent is available at any time: type AGENT."

Vague:        "Your application was declined by our system."

Useful:       "Your application was declined mainly because reported income is
               below the minimum for this product (about 40% weight) and the
               credit history is under 12 months (about 25%). You can submit
               updated documents or request a human review within 30 days."

Do not fake human-ness

Giving a bot a human name and photo without disclosure misleads people. A short, friendly disclosure builds more trust than a hidden machine that gets caught.

Quick check: What makes an explanation of a decision useful?

  • It says "the system decided"
  • It names the main factors and how to contest the outcome
  • It is as long as possible
  • It hides the data used
Answer

It names the main factors and how to contest the outcome — People need reasons they can understand and a path to challenge or correct them.