Responsible AI

Moderate · Governance, Ethics & Equity track · ~35 min hands-on + readings and quiz

← Back to the Governance, Ethics & Equity track

Pull the ethics thread together: who’s accountable, what to disclose, and how to govern AI use.

What you’ll be able to do

  • Apply the accountability principle: AI drafts, a human decides
  • Decide when to disclose AI use
  • Use a governance framework (WHO, ASPPH) to guide decisions

Overview

Responsible AI isn’t a separate task — it’s how you do every task. The governing principle from Workshop 1 holds: the person who acts on AI output owns the result. Disclosure, verification, and human oversight are the operational habits.

For grants and manuscripts, this principle now has real regulatory teeth: NIH will not review applications substantially developed by AI, NSF requires disclosure in the project description, and ICMJE/COPE prohibit AI authorship — see Grant Proposal Development and Manuscript Peer Review for worked examples.

Public-health bodies (WHO, ASPPH) offer frameworks for AI ethics and governance. They translate values — accountability, transparency, equity, safety — into checkable practices.

Practice activities

Activity 1 · Moderate — Write a disclosure norm

Time ~15 min · Tools ChatGPT Edu

Goal. Decide where and how to disclose AI use.

Setup. Below are three role-specific lists of places AI commonly shows up at DSPH. Find the list closest to your role:

Admin staff:

  • Drafting email reminders about deadlines (reaccreditation, HR, events)
  • Summarizing long policy documents or meeting minutes
  • Reformatting data from spreadsheets into reports or slide decks
  • Writing first drafts of job postings or committee charges

Research faculty/staff:

  • Summarizing literature or generating search strategies
  • Brainstorming and critiquing grant aims (not drafting them — see NIH/NSF policy), drafting abstracts or progress reports
  • Generating or cleaning code (R, Stata, Python)
  • Producing tables, figures, or lay summaries from results

Teaching faculty:

  • Drafting syllabi, rubrics, or assignment descriptions
  • Creating quiz questions or case studies
  • Writing recommendation letters or advising notes
  • Generating lecture outlines or discussion prompts

Pick the three uses from your list that feel most relevant. If none fits, substitute your own.

Steps.

  1. For each of your three uses, decide: Should you disclose AI assistance? To whom? How? (A footnote? A verbal mention? Not at all?)

  2. Draft a norm:

    Draft a one-paragraph AI-use disclosure norm for a public-health school covering these three cases. Be balanced and practical — neither requiring disclosure for every spell-check nor allowing silence on consequential uses. Cases: [your three uses].

  3. Refine to fit your context. A good norm should pass the “new colleague” test — could someone who just joined your team follow it without ambiguity?

Expected result. A short, shareable disclosure norm that distinguishes trivial use (no disclosure needed) from consequential use (disclose).

Check your work. Is it specific enough to act on, or just aspirational? “Use AI responsibly” is not a norm. “Disclose AI assistance in any document submitted externally — grants, publications, accreditation reports — by noting the tool and the human verification step” is.

Common pitfalls. Over-disclosing trivial use (auto-complete, grammar check) is noise; under-disclosing consequential use (drafting a grant aim, writing a student recommendation) is a problem. Aim for “where it matters to the reader.”

Stretch (optional). Add a line on who’s accountable when AI informs a decision, and circulate the draft to one colleague for feedback.

Activity 2 · Moderate — Apply a governance framework

Time ~15 min · Tools WHO or ASPPH framework + ChatGPT Edu

Goal. Turn principles into practice.

Setup. Skim one framework (e.g., WHO ethics and governance of AI for health).

Steps.

  1. Pick two principles (e.g., transparency, equity).

  2. Make them concrete:

    For each principle, give two concrete actions a public-health team could take in everyday AI use. [principles]

  3. Adapt the actions to a real task you do.

Expected result. Two principles turned into actions you’d actually take.

Check your work. Could you tell whether you’re following the action, or is it vague?

Common pitfalls. Frameworks become wallpaper if not operationalized. Make them checkable.

Stretch (optional). Identify one current practice that violates a principle and fix it.

Check your readiness

Answer these, then check — your score suggests whether to dive in or skim the readings first.

Useful resources