Skip to main content
ENES
← Course overview
Course contents

Module 5 of 5

A contribution you can actually test

3 hours of independent work + 2 hours of facilitated learning

Read the note, complete the worksheet, and bring one question to your session. Puente and its evidence cards are fictional teaching inputs.

Learning note

Learning note

A useful contribution connects a problem to a mechanism of change. “Raise awareness” is incomplete until you identify whose understanding should change, what decision that supports, and how you would notice improvement. A small, well-scoped artifact can be more informative than a large campaign with no feedback.

Students can contribute through technical evaluation, institutional procedures, research synthesis, education, or community participation. Technical research usually requires further preparation; completing this course does not make someone an alignment researcher. A nontechnical student can still scrutinize evidence, design an appeal process, or create a clear explanation with an honest uncertainty statement.

Choose work you can finish and evaluate. Ask: what assumption is most likely to make this useless? Test that first. A ten-minute review by a relevant person might reveal that your proposed resource duplicates something effective, solves the wrong problem, or overlooks a barrier. Record that result as learning and revise.

Your plan should include an alternative and a reason for choosing. If you propose a bilingual guide, specify the learning objective, audience, distribution route, and a comprehension check. Translation alone does not establish understanding. If you propose an evaluation protocol, say who can run it, what it measures, and what it cannot conclude.

Source lab · 30 minutes

Browse BlueDot's contribution module for optional ideas. Core bilingual alternative: compare (a) a general awareness poster with no intended audience or test, (b) a one-page appeal guide reviewed against a fictional case, and (c) a three-month technical project requiring skills you do not yet have. Rank them for your own circumstances and explain what information could reverse the ranking.

Preparation · 60 + 45 minutes

Spend 60 minutes completing W5's decision brief. Then spend 45 minutes checking evidence, preparing a three-minute presentation, and rehearsing answers to the strongest objection. Use the final 15 minutes of the preparation block for diagnostic B in the following worksheet. Drafts may be imperfect; bring a specific question for peer feedback.

W5 · Final brief, action plan and diagnostic B

Final brief · 500–700 words, or 5-minute audio plus source list

Use Puente or another bounded case. Include:

  1. Problem and people: the outcome to protect, affected people, and decision-maker.
  2. Evidence: at least two traceable sources or supplied evidence cards; label fictional inputs. State a limit of each.
  3. Mechanism: a causal pathway with assumptions distinguished from observations.
  4. Options: compare your recommendation with a credible alternative, including the manual process where relevant.
  5. Defenses: owners, permissions, monitoring, appeal or correction, stopping and restart conditions.
  6. Uncertainty: the strongest objection and evidence that would change your view.
  7. Next action: one realistic step in 30 days, resources needed, a reviewer, and a success or abandonment criterion.

Your 30-day plan

Days 1–7: verify the problem and find a willing reviewer; do not assume a mentor or institution has agreed.
Days 8–14: make a small prototype with public or fictional data.
Days 15–21: request feedback if you choose, document weaknesses, and revise.
Days 22–30: assess against your criterion; continue, change direction, or stop. Keep public sharing optional. No outreach or deployment is required to complete this course.

Diagnostic B · 15 minutes, without AI tools

  1. An AI succeeds on 95% of selected grant-review tasks. What do you need before authorizing final decisions?
  2. An agent reduces unresolved support requests by deleting difficult requests. Identify the objective problem.
  3. A person uses an AI to fabricate notices; another system independently pursues a poorly specified goal. Distinguish mechanisms.
  4. A draft, reviewer, and monitor share a model and source file. Name a shared failure and an independent safeguard.
  5. A report predicts rapid automation of an entire profession. What must be separated from its benchmark evidence?
  6. Propose a small safety contribution with a test that could show it is not useful.

Rubric · four criteria, 0–3 each

Evidence and limits; causal reasoning and uncertainty; defenses and tradeoffs; feasibility and evaluation. Each: 0 missing, 1 asserted/vague, 2 specific and supported with a gap, 3 clear, supported and critically tested. Target 8/12 with no zero. The guide gives criterion-specific anchors. Grammar, accent, technical vocabulary, and agreement with HASI are not criteria.