Decency

Podcast 2 — In conversation

Speaker A: I was reading a piece about online moderation and it kept using the phrase basic decency — as in, the platform should enforce basic decency in comments. I thought that was an interesting standard to try to codify.

Speaker B: It's a real challenge, isn't it — because decency is partly conventional and partly moral. What counts as basic decency in one community might be different in another.

Speaker A: Right. Though there are some things that seem pretty universal — not threatening people, not humiliating them publicly. Those feel like decency wherever you are.

Speaker B: That's the core meaning — treating people with dignity. And that's slightly different from propriety, which is more about following social norms. You can be proper without being decent, and decent without being particularly proper.

Speaker A: That's a good distinction. Propriety is about form; decency is about substance — genuine respect versus merely observed convention.

Speaker B: Exactly. And a synonym worth knowing: civility — which is decency enacted in speech and conduct. Civility is the visible behaviour; decency is the underlying quality. You can have decency without perfect civility, and you can perform civility with no real decency behind it.

Speaker A: The phrase common decency is one I find interesting — it implies decency is shared and ordinary, not a high standard.

Speaker B: Yes — common decency means the minimum expected of anyone. It's the baseline. When you say someone lacked common decency, you're saying they fell below the floor, not just short of some ideal.

Speaker A: Common mistake: confusing decency with prudishness. Decency isn't about being old-fashioned about bodies or language — it's about genuine respect and fair treatment.

Speaker B: Well put. Decency is outward-facing — it's about how you treat others. Prudishness is about personal discomfort with certain content. Very different things.

Podcast 3 — Prompt engineering

Speaker B: Now let's look at how decency works in AI prompts. In content moderation, platform policy, and ethics tooling, decency is a practical concept that needs to be operationalised.

Speaker A: Turning an abstract moral concept into something a system can actually enforce — that's the challenge.

Speaker B: Exactly. First prompt.

Design a content moderation scoring system for a community platform that evaluates user posts on a decency scale from 1 to 10, where 1 is grossly indecent and 10 is exemplary. Define the rubric criteria for each score level — covering personal attacks, dignity, factual honesty, and civil tone — and output the score with a short explanation.

Speaker B: A decency scale as the primary metric frames the system around human dignity rather than just keyword matching.

Speaker A: What about generating platform policies?

Write a community standards document for a developer forum that defines minimum decency requirements for posts and comments. Include specific examples of what constitutes a decency violation — such as personal insults, deliberate misrepresentation, and exclusionary language — and outline the escalating consequences for repeated violations.

Speaker B: Minimum decency requirements is clear and actionable — it tells the AI you want a floor, not an ideal.

Speaker A: How about HR systems?

Build a TypeScript service for an HR platform that tracks employee conduct reports and categorises each incident by severity: basic decency violation, conduct policy breach, or legal threshold breach. Generate a monthly report grouped by category and department, and flag any employees with three or more decency violations in a quarter.

Speaker B: Separating decency violations from policy breaches in the schema reflects real HR practice — they have different processes and consequences.

Speaker A: What about AI assistant behaviour?

Create an evaluation dataset for testing whether an AI assistant maintains decency standards when responding to hostile or provocative user inputs. Include 50 test cases with expected responses that are firm, honest, and dignified. Score each model response on a decency rubric covering tone, honesty, and respect.

Speaker B: A decency rubric for AI evaluation is a genuinely useful framing — it captures what you want from the model beyond just accuracy.

Speaker A: One more — for a moderation dashboard?

Build a React dashboard for content moderators that displays pending reviews sorted by estimated decency score. Include a side-by-side view of the post, the decency score breakdown by category, the user's previous decency violation history, and a one-click moderate button that triggers the appropriate action.

Speaker B: Decency score breakdown by category makes the moderator's decision traceable and consistent — not just a gut call.

Speaker A: Making decency measurable and traceable in a system — that's surprisingly elegant.

Speaker B: The clearer your vocabulary in the prompt, the clearer the system it produces.

Speaker B: Prompt example: