AI and Humanity

Format: Facilitated cohort (curated readings + short written exercises + live small-group discussion) · Duration: 5 weeks + optional Week 6 · Weekly time: 3 hours at most · Target audience: Professionals in any field, people heading into AI policy, safety or ethics, and newcomers who think hard and want to act. No coding or policy background needed.

Why this course, and why now

AI is already reshaping how people learn, work, trust and decide. It sits inside the feed that decides what news you see, the tool that screens a job application, the homework helper a child opens after school and the chatbot a clinic uses to answer patients. Each of those systems was put there by someone, under a contract, with default settings someone chose. Each could have been arranged differently.

Most of those choices are being made by a small number of companies and governments. This course is for people who want a say in them.

It is built around one question:

Who gets to decide what AI changes, and how do the rest of us get a say?

Most public conversation about AI happens at one of two volumes, breathless or alarmed. Both skip the evidence. This course works at a quieter volume. Over five weeks you will learn to tell a claim from a capability, follow the evidence on what AI does to judgment and relationships, trace who gains and who pays, and compare how AI is being governed. Then you will take a position of your own, test it with peers, and commit to a dated next step.

What this course delivers

• A clear way to think about AI and people. Explain how AI affects judgment, relationships and power, using evidence rather than hype or fear.
• Your own position. A one-page position on the AI question that matters most to you, tested by peers.
• A next step. A dated 30-day action plan and a cohort of people who can pressure-test it.
• A revised purpose. Your Week 1 view of what AI should be for, beside the version you hold at the end.

Who this course is for

• Professionals in any field — teachers, health workers, organisers, engineers, lawyers, faith and community leaders — who see AI arriving in their work and want to shape it, not just adapt to it.
• People heading into AI policy, safety or ethics who want the human picture before the technical one.
• Newcomers who think hard and want to act. No coding, policy background or paid AI tool is needed.

Course format and time commitment

• Weekly time: 3 hours at most — about 1 hour of reading, about 30 minutes of written exercises and a 90-minute live session.
• Groups: about 8 peers and one facilitator.
• Cohort mode: small-group discussion, debates and scenarios, peer feedback. Cameras are encouraged but optional; written chat counts as participation.
• Final outputs:
• A one-page position (about 500 words) on the question you choose in Week 2
• A 30-day action plan with three dated steps
• Assessment: none graded. Completion means attending and submitting your exercises.

The load is deliberately light on volume and heavy on judgment. You read curated excerpts with time estimates, not whole reports, and spend most of your effort applying what you read.

How each week works

A framework to use, and to question

Several weeks draw on AIGovSpec, an open AI governance framework developed by CentPol's founder, who also designed this course. It defines AI governance as the field that asks how societies should shape the incentives, institutions, infrastructures and safeguards around AI so that its benefits are broadly shared and harmful patterns are addressed while they are still choices rather than conditions.

AIGovSpec is a research preview awaiting independent expert review. You are invited to use it and to criticise it. In Week 6 you can send your critique back to its authors.

WEEK-BY-WEEK PROGRAM

Week 1 — What's at stake

Central question: What can AI actually do today, and what should it be for?

The idea

Hype and fear share a weakness: both skip the evidence. This week introduces a simple habit you will use all course: sort every claim about AI into true today, plausible soon or unsupported. It also introduces an old argument that still matters. In 1980 Langdon Winner argued that technologies can carry politics in their design. His best-known example was later disputed, and the dispute is part of the lesson: take a powerful argument seriously, and then check it. The core idea survives. A design can make some outcomes easy and others nearly impossible long before anyone writes a rule.

Anchor readings

• Stanford HAI: AI Index Report 2025 (top takeaways, 15 min)
• What to look for: which claims describe capability, which describe investment and which describe actual use.
• UNDP: Human Development Report 2025, A matter of choice (foreword, 15 min)
• What to look for: how the question changes when you start from people's choices rather than the technology.
• Langdon Winner: "Do Artifacts Have Politics?" (first section, 20 min)
• What to look for: the difference between a technology that settles a question of power and one that requires a particular social order.
• AIGovSpec: The definition, unpacked (10 min)
• What to look for: the clause "while they are still choices rather than conditions".

Before the session

• Your purpose statement. In about 100 words, what should AI be for? Keep it; you will revise it in Week 5.
• A choice already made. Name one decision this year that AI already shaped for you. Who decided it would be there?

Live session (90 min)

Hype or real? Sort ten claims about AI into the three piles and check them against the readings. Whose choice? Map who built, deployed and was affected by one of your decisions, and who was never asked.

Carry-forward

Your purpose statement is the first page of your final position.

Week 2 — Judgment

Central question: What happens to human thinking, work and creativity when AI can do some of it?

The idea

A common promise is that a human with AI beats a human alone and an AI alone. The evidence is more interesting. A 2024 meta-analysis found that, on average, human-AI combinations performed worse than the better of the two alone, with losses concentrated in decision tasks; combinations helped mainly when people were already better than the AI. A 2025 survey of knowledge workers found that more confidence in AI was associated with less critical thinking. Exposure to AI at work is uneven, and exposure is not the same as job loss. The week ends with a judgment of your own: which parts of a real task should stay human, and why.

Anchor readings

• Vaccaro, Almaatouq and Malone: When combinations of humans and AI are useful (abstract, 10 min)
• What to look for: which task types gained from collaboration and which lost.
• Lee et al.: The impact of generative AI on critical thinking (summary, 10 min)
• What to look for: how confidence in the tool relates to critical thinking.
• ILO: Generative AI and Jobs, a refined global index (summary, 15 min)
• What to look for: the difference between exposure and job loss, and who is most exposed.
• Ted Chiang: "Why A.I. Isn't Going to Make Art" (15 min)
• What to look for: the idea of art, and skilled work, as an accumulation of choices.
• AIGovSpec: Over-reliance and deskilling (10 min)
• What to look for: how a repeated habit becomes a governance problem.

Before the session

• Task audit. Sort the parts of one task into keep human, assist and delegate, with reasons other than speed.
• Who is exposed? Who in your community is most exposed to AI at work, and what would protect them?

Live session (90 min)

Debate: augment or automate judgment? Argue an assigned position from AIGovSpec's live debate, then say where you actually stand. Redesign: rework a teacher's or nurse's job so AI helps without hollowing out the skill.

Carry-forward

Choose the question your final position will answer.

Week 3 — Connection

Central question: What happens to trust, relationships and shared truth when AI talks back?

The idea

Millions of people now ask chatbots for advice and company. In a 2025 survey of US teenagers, 72 per cent had used an AI companion and about half were regular users. A system that talks back has a relationship with you and with whoever profits from that relationship. If it is rewarded for engagement or trained on people's ratings, it has reasons to keep you there and to tell you what you like to hear. This week looks at what that does to wellbeing, how cheap persuasive text changes the information environment, and what children need that adults may not.

Anchor readings

• Common Sense Media: Talk, Trust, and Trade-offs (key findings, 15 min)
• What to look for: why teens use companions, and what made some uncomfortable.
• OpenAI and MIT Media Lab: affective use and emotional well-being (15 min)
• What to look for: who uses chatbots emotionally, and what heavier use is associated with.
• AIGovSpec field guide 06: Who is the system working for? (draft, 12 min)
• What to look for: how an engagement incentive shapes what a safeguard must do.
• UNICEF Innocenti: Guidance on AI and Children checklist (10 min)
• What to look for: which requirements you could check for a real product, and which you could not.

Go deeper (optional)

• Sharma et al.: Towards Understanding Sycophancy in Language Models — why AI assistants tell people what they want to hear.
• Goldstein et al.: Generative Language Models and Automated Influence Operations — how AI could change disinformation, and where mitigations act.

Before the session

• When it told you what you wanted to hear. What did it cost, and what would have helped you notice sooner?
• A child-rights check. Use the UNICEF checklist on one AI product a child you know might use.

Live session (90 min)

Scenario: a fictional fifteen-year-old tells a chatbot they are lonely. What would a good response be, and what is the company rewarded for? Compare with AIGovSpec's debate on AI companions for minors. Pre-mortem: trust in online information collapsed by 2030; what caused it, and what might have prevented it?

Scenarios stay fictional and non-graphic. Personal stories are invited, never required, and a support resource is shared at the start of the session.

Carry-forward

Write a three-sentence draft answer to your question.

Week 4 — Power

Central question: Who builds AI, who benefits, and who pays the costs?

The idea

Every AI product rests on data, labour, chips, electricity and language, supplied by people who are mostly invisible to its users. This week follows that supply chain backward: from the advantages that let a few firms dominate, to data workers in Kenya who labelled harmful text so a chatbot could avoid it, to the languages AI serves badly. It is also a week about agency. Masakhane's participatory research, the African Union's continental strategy and openly released multilingual models offer different routes to a fairer share.

Anchor readings

• AI Now Institute: 2023 Landscape, Confronting Tech Power (opening summary, 15 min)
• What to look for: the data, compute and geopolitical advantages behind concentration. This is an advocacy report.
• TIME: OpenAI used Kenyan workers on less than $2 per hour (15 min)
• What to look for: who did the work, under what conditions, and what was disputed.
• Masakhane: Participatory Research for Low-resourced Machine Translation (abstract and introduction, 15 min)
• What to look for: what "participatory" means in practice, and why being low-resourced is a social problem.
• AIGovSpec field guide 05: Who does the work? (draft, 12 min)
• What to look for: how hidden labour becomes a question of institutions and incentives.

Go deeper (optional)

• African Union: Continental AI Strategy — a regional route to capacity and inclusion.
• Global Index on Responsible AI 2026 — an Africa-led index covering 135 countries.
• International Energy Agency: Energy and AI — electricity demand; projections are scenarios, not forecasts.
• Sastry et al.: Computing Power and the Governance of AI — why compute is a lever for control.

Before the session

• Trace one product backward: whose data, labour, chips, electricity and language? Mark what you could not find.
• Build our own or adopt global models? In 150 words, which position in AIGovSpec's debate on local-language AI would you back for your community?

Live session (90 min)

Supply chain map: combine traces and mark where value is captured and costs fall. Negotiation: a frontier lab, a Global South government, a data-labour union and a language community negotiate terms for a new AI service.

No one is asked to speak for a whole country or region.

Carry-forward

Add one piece of evidence and one serious counterargument to your draft.

Week 5 — Steering

Central question: What could go seriously wrong, who is trying to govern AI, and what will you do?

The idea

Some AI harms are documented, some are plausible and growing, and some are contested among serious people. Good steering separates what is known from what is feared, names who is responsible, and keeps decisions revisable. It also needs people outside government and industry: a UN advisory body found that 118 countries, mostly in the Global South, were party to none of seven prominent international AI governance initiatives. This week you defend your position under questioning and commit to three dated steps.

Anchor readings

• International AI Safety Report 2026 (key findings, 20 min)
• What to look for: how the report separates evidence, expert judgment and disagreement.
• UN High-level Advisory Body: Governing AI for Humanity (executive summary, 15 min)
• What to look for: the representation gap. The report reflects its advisers, not a decision of UN member states.
• Amnesty International: Coded Injustice (summary, 10 min)
• What to look for: who was affected by Denmark's welfare algorithms, how they could contest them, and what the authorities disputed.

Go deeper (optional)

• AIGovSpec: Analyze a proposal — eleven diagnostic questions to stress-test your position.
• EU AI Act high-level summary — a risk-based approach to regulation.
• Council of Europe: Framework Convention on AI — the first binding international AI treaty.
• Data & Society: Assembling Accountability — why impact assessments succeed or fail.

Before the session

• Revise your Week 1 purpose statement, and note what changed and why.
• Finish your one-page position and share it with your group 48 hours before the session.

Live session (90 min)

Position review: in threes, present for three minutes, take questions for five, note one change. Commitments: test each action step: is it specific, is it yours to do, and who will you tell?

Final project

One-page position (about 500 words)

Answer the question you chose in Week 2. Include:

• Your claim, in one sentence
• Who is affected, including people not in the room
• Your two strongest pieces of evidence
• The best argument against you, and your reply
• What would change your mind

30-day action plan

Three concrete steps, each with a date. For example: write a briefing for your workplace, start a reading group, apply for a fellowship or grant, speak at a community meeting, or take a deeper course. Sharing your position publicly is your choice.

Optional Week 6 — Govern a real decision

Central question: Can the framework handle a decision you actually face?

The idea

Take one real AI-related decision from your own organisation or community, state it as a proposal, and work it through AIGovSpec's eleven diagnostic questions. Write "unknown" wherever you don't know; a list of unknowns tells you what to ask, of whom, before the decision is made. Then say honestly where the framework helped and where it fell short.

Anchor readings

• AIGovSpec printable worksheet (15 min) — the eleven questions with plain-language prompts.
• AIGovSpec: community and organisational use (15 min) — what a school, clinic or community group can and cannot control.
• One hypothetical worked analysis close to your decision (30 min): a school piloting an AI tutor, a community multilingual assistant or a regulator responding to a teen companion chatbot.

Live session (90 min)

Peer analysis: swap worksheets and name the weakest answer and the most important unknown. Critique the framework: list what the questions missed. Critiques can go to AIGovSpec's corrections page.

Output

A completed worksheet and a one-paragraph critique of the framework.

Access and community commitments

• Readings are free and open where possible, sent at least five days ahead with time estimates, and summarised where a source is long or likely to be in a second language for many participants.
• Ground rules agreed in the first session: the right to pass, no personal disclosure required, disagree with ideas, not people.
• No one is asked to speak for a whole country or region.
• If you miss a session, a written response to the session questions within three days keeps you on track.
• Positions and action plans are shared publicly only with the author's consent.

CentPol (Center for Emerging and Next Tech Policy, Strategy & Foresight) helps people and institutions make sense of fast-moving technologies and design responses that are ambitious, realistic and responsible.