Lesson Report:
Title
Policy Memo Workshop + Democratic Legitimacy: Dialogue, Due Process, and AI in Governance
This session combined assignment coaching with substantive discussion of democratic principles under AI-mediated decision-making. Students reviewed the upcoming policy memo requirements, then analyzed when algorithmic systems undermine democratic legitimacy via two structured debates. The class closed by introducing Beckman’s “principle of publicity� to frame Thursday’s AI-governance simulation.
Attendance
– Students explicitly mentioned absent: 0
– Notes:
– Early headcount: 14; later participation: ~20–21 (some late arrivals).
– Several students reported repeated disconnections (e.g., Nirani) and re-joins.
– At least three students currently lack access to eCourse (Elijah + two others); instructor will email materials.
Topics covered (chronological)
1) Housekeeping and platform access
– eCourse access remains delayed for some students; instructor will email all critical materials (assignment page + template) to those affected.
– Confirmed submission link in eCourse: “Policy memo — due November 1.â€� Students without access should email submissions if access is not restored by the deadline.
2) Assignment workshop: Policy Memo (due Nov 1)
– Purpose and learning goals
– Demonstrate deep engagement with course content (AI + democracy).
– Move beyond diagnosing problems to proposing a concrete, evidence-based policy solution.
– Adopt a policy-memo voice: you advise a specific decision-maker.
– Topic and scope selection
– Choose one specific AI–democracy problem (examples raised: electoral transparency, algorithmic bias, AI-generated disinformation).
– Tie the problem to one jurisdiction:
– Country (e.g., Kyrgyzstan, Japan), subnational unit (e.g., California, Ontario, Uttar Pradesh), or international body (e.g., EU, UN).
– Role-play audience: address the memo to a relevant decision-maker in that jurisdiction (examples: Chair of the US Federal Trade Commission; European Commission VP for Values and Transparency).
– Required memo structure (template provided as a Word doc)
– Executive summary.
– Background of the problem in the chosen jurisdiction (brief history/current context).
– Recommendation (must be justified with comparative, real-world evidence—how others address similar problems).
– Implementation and potential challenges (steps to operationalize; foreseeable hurdles).
– The template includes section descriptions and approximate lengths; students may write directly in it.
– Evidence and citations
– Minimum 4 distinct sources total:
– At least 2 course readings (to ground arguments in class content).
– At least 2 external, credible sources demonstrating current real-world relevance (think tanks, government publications, reputable news).
– Many course readings are 5–6 years old; external sources should update the “state of the problem now.â€�
– APSA citation style required; all claims must be properly cited.
– Papers submitted without in-text citations will not be accepted; resubmission allowed within a one-week late window, with late penalty.
– Large Language Models (LLMs) such as ChatGPT/Gemini:
– Outputs do not count as sources and must not be cited.
– Copying from LLMs is not permitted.
– Academic integrity
– Zero tolerance for plagiarism (copying from peers, past assignments, or LLMs).
– Plagiarized work will not be accepted.
3) Transition: From Algorithmic Bias to Democratic Principles
– Revisited last week’s Allegheny child-welfare algorithm reading and a key quote: “You can teach people how you want to be treated… You can’t fix that number.â€�
– Guided extraction of democratic principles embedded in the quote:
– Dialogue: Democratic governance requires ongoing communication between people about processes that affect them.
– Right to be heard / free speech: Individuals must be able to voice political concerns without suppression and be meaningfully heard.
– Context-sensitivity: Human decision-makers can integrate nuanced context that rigid numerical outputs miss.
– Right to appeal / due process: When a decision is wrong or harmful, there must be a clear, accessible avenue to challenge and reverse/modify it.
4) Debate 1 (low stakes): AI auto-grading vs. human regrade
– Scenario setup
– Large, required course; first major assignment returned instantly with an AI-issued C+.
– University claims 99.9% AI grading accuracy; AI provides no feedback (only a grade and some highlighting).
– Option A: Keep the AI C+.
– Option B: Request a regrade by a (tough) professor; grade may go up or down.
– Instructions
– Groups of four; five minutes; unanimous choice required; submit rationale via poll.
– Results
– 0 groups chose to keep the AI grade; all chose professor regrade.
– Rationale discussed
– Feedback as learning: Grading should improve skills; commentary/justification matters more than the raw number.
– Human judgment can engage with ideas, arguments, and creativity; alignment with course goals and rubric is clearer.
– Even if AI accuracy is high, lack of reason-giving makes it impossible to learn or contest.
– Acknowledged counterpoint: Human graders can be biased; however, reasoned feedback enables challenge and growth.
5) Debate 2 (high stakes): Plagiarism appeal — AI verdict vs. faculty committee
– Scenario setup
– Student receives a plagiarism accusation; policy dictates failure on the assignment and the course.
– Student insists they did not plagiarize.
– Option A: New AI appeals system (99.9% accurate), complete black box; outputs only “guilty/not guilty.â€�
– Option B: Committee of three overworked professors; live hearing with reason-giving, student explanation, and on-the-spot questioning/appeal.
– Instructions
– Groups deliberate; unanimous choice required; rationale submitted via poll.
– Results
– Near-even split across groups between AI and committee options.
– Rationale for choosing AI
– Perceived speed and consistency; reduced risk of human bias or pre-judgment.
– Anxiety management: Live hearings can disadvantage nervous students; “awkwardnessâ€� and performance pressure may skew outcomes.
– Belief in institutional accuracy claim (99.9%), especially when innocent.
– Rationale for choosing committee
– Due process: Ability to present evidence, narrate writing process, and directly challenge the decision.
– Reason-giving: Transparent justification builds trust and enables correction of error.
– Authenticity showing: Students can demonstrate their own work in ways a black-box system cannot evaluate.
– Acknowledged risks: Human error/bias and lack of uniform standards; nonetheless, the hearing provides recourse.
6) Synthesis: Legitimacy and the Principle of Publicity (setup for Thursday)
– Defined legitimacy: The degree to which people accept that rulers should rule (trust in the process even without liking outcomes).
– Introduced Beckman’s argument: Many AI governance systems are inherently undemocratic because they violate the “principle of publicity,â€� which requires:
– Reason-giving: Authorities must explain the reasons and evidence for decisions affecting citizens.
– Accessibility: Citizens must be able to access, understand, and challenge those reasons (i.e., have a meaningful right to appeal).
– Link to debates: High “accuracyâ€� without transparency, reason-giving, and accessible appeal undermines democratic legitimacy.
– Preview of Thursday
– Reading: Beckman article (≈7–8 pages).
– Activity: Students will be “governed by an AIâ€� and will evaluate where it meets/fails publicity and legitimacy standards.
Actionable items
Urgent (before Thursday)
– Email materials to students without eCourse
– Send the policy-memo template and assignment page to Elijah and the two other students lacking access; confirm receipt.
– Ensure reading access
– Post/share Beckman reading; verify all students (including those without eCourse) can access it.
– Prepare Thursday’s AI governance simulation
– Finalize scenario design, decision rules, and how “reason-givingâ€� and appeals will (or won’t) be provided so students can assess publicity.
– Coordinate with IT on eCourse accounts
– Request expedited account creation; communicate a clear contingency submission path (email) if access is still delayed by Nov 1.
– Fix polling/logistics
– Investigate why students couldn’t see poll results; prepare a reliable backup (link) if Zoom polls fail.
This week (to support the memo)
– Share APSA quick guide and citation examples
– Provide a one-page APSA cheat sheet; model in-text citations for memos; reiterate minimum four sources (2 course + 2 current, credible externals).
– Clarify AI and integrity policy
– Reiterate that LLM outputs are not acceptable sources and text must be student-written; restate consequence and late resubmission policy for missing citations.
– Provide scaffolded exemplars
– Post an annotated sample memo using the template (executive summary, background, evidence-based recommendation, implementation/challenges).
– Offer office hours/workshop
– Hold a short session for topic selection, audience matching, and sourcing credible current evidence.
Before Nov 1
– Encourage early check-ins
– Invite optional 1-paragraph proposals (issue + jurisdiction + audience + initial sources) for quick feedback.
– Resource curation
– Consider posting a short list of credible outlets (think tanks, government agencies, reputable news) relevant to AI-and-democracy topics to streamline students’ search.
Homework Instructions:
” ASSIGNMENT #1: Policy Memo — Propose a Solution to a Specific AI-and-Democracy Problem
You will write a concise, evidence-based policy memo that moves beyond diagnosing problems to proposing a concrete solution for a specific AI-and-democracy issue in a specific jurisdiction. This builds directly on today’s focus on democratic principles (dialogue, the right to be heard, due process/appeal, legitimacy, and the principle of publicity), asking you to recommend an actionable policy for a real decision-maker.
Instructions:
1) Choose your topic and scope
– Select one concrete AI-and-democracy issue (e.g., algorithmic bias in public services, electoral transparency, regulation of AI-generated disinformation).
– Tie the problem to one clearly defined jurisdiction:
• A country (e.g., Japan, Kyrgyzstan)
• A subnational unit (e.g., California, Ontario, Uttar Pradesh)
• An international body (e.g., EU, UN)
– Keep the problem specific to that place’s context and institutions.
2) Choose your audience (role-play element)
– Address your memo to a named, relevant decision-maker for your chosen jurisdiction (e.g., an agency chair, minister, or EU Commissioner). This should be a real person with plausible authority over the issue.
3) Use the provided template
– Download the Word template linked on the assignment page. It is fully formatted and describes each section’s purpose and suggested length. Write directly in that template.
4) Structure your memo around the four required parts
– Executive Summary: In one short, clear paragraph, state the problem, your recommendation, and the expected impact.
– Background of the Problem (in your jurisdiction): Explain the history, status, and stakes of the issue where you’ve located it. Use context that matters for democratic governance (e.g., transparency norms, oversight bodies, legal environment).
– Recommendation (evidence-based): Propose a specific, feasible policy. Justify it with real-world examples and comparative evidence (how other places tackled similar issues; what worked/failed and why).
– Implementation and Potential Challenges: Lay out practical steps (who must act, what legal/administrative changes are required, timelines, resources), and anticipate obstacles (political, legal, technical), including how to address them.
5) Ground your analysis in course themes
– Connect your proposal to democratic principles highlighted in class (e.g., reason-giving and accessibility/the principle of publicity; the right to appeal/due process; dialogue and the right to be heard). Show how your policy preserves or strengthens democratic legitimacy.
6) Conduct targeted research
– Identify how this problem is currently unfolding in your chosen jurisdiction (recent developments, regulatory debates, public controversy).
– Find comparative cases that inform your solution design.
7) Meet the source requirements (citations are mandatory)
– Use a minimum of four distinct sources:
• At least 2 course readings
• At least 2 external, credible sources (e.g., think-tank reports, government publications, reputable news)
– Outputs from LLMs (e.g., ChatGPT, Gemini) do not count as sources and should not be cited as evidence.
– All claims must be supported with in-text citations.
8) Follow the citation style
– Use APSA style for in-text citations and the reference list, as specified on the assignment page.
– Papers without in-text citations will not be accepted; if this happens, you may use the one-week late window to fix and resubmit (late penalty applies).
9) Length and formatting
– Aim for a concise 4-page memo (as noted in class), following the template’s guidance for section lengths and headings.
10) Draft and revise
– Outline each section using the template prompts before drafting.
– Check that your recommendation directly addresses the specific problem in the specific jurisdiction and is actionable.
11) Final checks
– Verify every claim has a supporting citation.
– Ensure your recommendation is explicitly tied to the evidence you present.
– Confirm your memo is addressed to a named decision-maker and that the Executive Summary clearly states the “ask.â€�
12) Academic integrity
– Do not plagiarize. Do not submit text generated by LLMs as your own writing. All work must be your own, with proper citations of human-authored sources.
13) Submission and deadline
– Due date: November 1.
– Submit via the “Policy Memo — Due November 1â€� submission link.
– If you do not yet have platform access, email your memo to the instructor by the deadline.
14) Start early
– The instructor strongly recommends starting by selecting your issue and jurisdiction and doing targeted reading now so you have ample time to draft and revise.
” ASSIGNMENT #2: Read Beckman on Publicity, Reason-Giving, and Accessibility (for Thursday)
You will read the Beckman article (approx. 7–8 pages) to prepare for Thursday’s activity, where you will evaluate an AI “governing� scenario. This reading connects directly to today’s discussion of democratic legitimacy and the principle of publicity (reason-giving and accessibility).
Instructions:
1) Read the assigned Beckman article carefully before Thursday’s class.
2) As you read, track the core claim: why many AI-based governance systems are inherently undemocratic when they cannot provide adequate publicity (reason-giving and accessibility).
3) Annotate with today’s principles in mind:
– Dialogue and the right to be heard
– Context and judgment
– Due process and the right to appeal
– Legitimacy and public acceptance of decisions
4) Prepare to apply the reading:
– Identify one example (from class scenarios or real life) where an AI system lacked reason-giving or accessibility, and note how that affected perceived legitimacy.
– Be ready to explain how Beckman’s argument would evaluate that example.
5) Come prepared to participate in Thursday’s activity, where you will critique an AI’s decisions using Beckman’s criteria (publicity: reason-giving + accessibility) and the democratic principles we surfaced today.