What we make with each other

Start here

A showcase, not a finish line

The group used the final session to make their developing work visible: a redesigned AI-for-social-impact course; a schoolwide AI-literacy framework; a project model rooted in community interviews; a constrained, AI-assisted visual tool; an artist's investigation of a model's defaults; and a presentation for school leaders about generative AI. Each share was practical, but none reduced the question to whether a tool can produce something quickly. The repeated question was what students and educators need to understand, decide, and remain responsible for.

Who was in the room

A small group for the last gathering, joining from the US, the UK, and Brazil. Six people shared work. In keeping with the recaps for Sessions 1–3, participants would normally be identified by first name or chosen display name only — but because this session was not recorded and consent responses are still coming in, contributors are described here by role until they opt in.

What participants shared

Redesigning an AI for Social Impact course

One participant is reshaping an upper-school course for grades 9–12 around AI and machine learning foundations, bias and intersectionality, environmental consequences, policy and law, and an entry-level coding component using creative-coding resources and ml5.js. The proposed capstone asks students to make and peer-review arguments about what policy, regulation, or preconditions should govern an AI system — not merely to build one.

The course also reserves its longer periods for student-led, seminar-style discussion of news and journal articles, leaning on the source-evaluation habits students bring from their humanities classes. That structure foregrounds a basic responsibility cycle: people make systems; systems produce benefits and harms; people must respond. The group discussed rewriting a constitutional amendment to cover digital rights, and speculative fiction, as ways to give students room to imagine what protections could exist. A practical constraint sat underneath all of it: the school's content filter currently blocks most of the tools the course would need, so access has to be negotiated before the spring.

An AI-literacy framework led by pedagogy

A technology integrator shared a schoolwide framework informed by OECD, UNESCO, and Brazilian Ministry of Education guidance, presented to the whole faculty across two trainings — one on how LLMs work and an acceptable-use policy, one on classroom application. Its four strands — engaging with AI, creating with AI, managing AI, and designing AI — gave teachers a way to distinguish understanding a system's influence from using it, directing it, or building with it. The stated principles: pedagogy leads and technology follows; don't outsource learning to AI; teach AI literacy; keep humans at the center.

Examples made the framework concrete: comparing generated images of “an athlete” across sports to examine body bias; younger students co-writing a story with a teacher-configured chatbot, then discussing who the author is; using a chatbot for pseudocode and troubleshooting in an Arduino project while keeping computational thinking central; and eighth graders training micro:bit movement classifiers and wearing them on a field trip. A teacher activity on acceptable and unacceptable use produced the useful friction of contextual judgment: most teachers rejected students using AI to translate an essay into English — and then had to sit with the fact that they translate their own staff messages the same way.

Community as the critical-thinking method

A proposed project model begins with a community issue or value. Students use AI to brainstorm and prototype, interview people who are actually connected to the issue, compare the system's answers with those responses, revise, and present what AI did and did not help with. The design draws on constructivist and constructionist learning: knowledge is built through making, but also through sustained relationships and feedback.

This model sharpened a limit of asking an LLM to “be skeptical.” A system can imitate the language of critique or apply a supplied framework — one participant described using exactly that setup to productively challenge teachers' lesson plans — but it does not bring lived stakes, local knowledge, or a relationship to consequences. The room's proposed extension: don't just ask the model to critique its own plan, go find the people in the community who disagree with it. Community input is not an extra validation step; it is part of the thinking.

Making a small tool, not a replacement for a practice

A design-semiotics instructor shared a constrained collage tool made in Processing through many iterations of human–AI collaboration. The tool randomly loads images, public-domain words drawn from old advertisements, and simple graphic elements onto a rule-of-thirds grid, with randomizer, color-shape, scaling, and save-to-PNG controls. Its intentionally narrow scope was part of the point: it gives students material to compose with without presenting itself as a rival to a full design environment. Students still collage by hand with paper and scissors first.

Asked how he decides a tool is finished, he offered a useful answer for anyone caught in an open-ended loop with a model: stop when it is good enough to put in front of students, then let watching them use it decide the next feature. The share also prompted a question about assessment in an era of AI-written papers. Mind maps, iterative artifact trails, and conversation can make student thinking more visible than a polished final essay alone. The personal objects at the center of the course — one student's object was a camera her grandfather took from a dead soldier in the Second World War — create a productive contrast with AI's pattern-based, impersonal generalities.

Using a model's defaults as artistic material

An artist and educator has been working with Stable Diffusion 1.5, generating images from a prompt for a doctor photographed on a twin-lens reflex camera with Kodachrome film. The outputs varied widely, and the interesting failure was that the model kept losing track of whether the doctor or the camera was the subject — a confusion the artist treated as material rather than error. The mid-century look nobody asked for came along with the camera reference, since that is the era the model has seen those photographs from.

The work connects to a longer practice of making art with a grandfather's tools — ink pens he used as a shipbuilder, a house-painting brush — for a man the artist never met. The lost twin-lens reflex camera is the one tool that cannot be inherited, so it is being simulated instead. The method is accumulative rather than subtractive: gather a large body of outputs first, like collecting post-consumer material, and shape it later. The share ended on the value of elders and of studying the past, and on a point the group returned to more than once: nothing about the present was inevitable.

Talking to school leaders about generative AI

The facilitator shared a presentation for administrators that used short activities — a sentence-compression word game, an AI familiarity-and-experimentation spectrum, a model of what an LLM can and cannot know from lived experience, A/B comparison of AI-generated messages, and image-prompt comparison — to open conversation rather than sell a product. Two framings did the most work: experimenting with a tool is not the same as endorsing it, and comfort is not expertise. A major concern was the addition of generative AI to school productivity suites for all students this year, with no meaningful opt-out for the school.

The presentation also estimated the energy and water associated with producing its own materials, and compared that with average daily household electricity use in Bangladesh. The figures were presented as a rough estimate and a classroom prompt, not a measurement: if routine educational use depends on a high material cost, “use it everywhere” is not a serious sustainability plan. The deck is on GitHub; a version without school-specific references is planned.

What connected the shares

• Pedagogy leads; technology follows. The strongest examples began with a learning goal, a community question, or a discipline's practice — not a tool's feature list.
• Human responsibility survives automation. Students need language for bias, authorship, environmental costs, policy, and the people affected by a system.
• Fluency is not situated knowledge. A plausible answer can be helpful for brainstorming, but it cannot substitute for an interview, a stakeholder, or a student's own reasoning.
• Constraints can be a teaching choice. A limited tool, a Socratic assistant, or a structured discussion can preserve room for judgment rather than optimize it away.
• Uniformity is a risk worth naming. The group noted the tendency of AI-generated text, images, and “vibe-coded” products to converge on the aesthetics and assumptions of large technology companies.
• Skepticism is a form of optimism. Criticism only makes sense if you believe the system can still respond to it. Several people made the same argument from different directions: if you want AI to get better, the skeptics are the ones who have to be given room.

Announcements

Teaching notes to carry forward

Action items

• Complete the consent form. Use the consent page to choose whether and how your name, words, or described work may appear in the written recaps. If you do not respond, you will be described by role or removed — this recap stays anonymous until you opt in.
• Complete the Session 3 feedback form if you have not already: Session 3 feedback.
• Register interest in the AP Java / Processing course — or pass it to a teacher who would want it — via the interest form.
• Consider proposing a CC Fest workshop when the call opens. Presenting work in progress is a way to find collaborators, not just an obligation to have something finished.
• Reach out to collaborate. Several participants offered to work together on classroom projects; contact details are on the course classroom.

Links & resources from the session

This summary was drafted with AI assistance and reviewed by a human facilitator. This session was not recorded; quotations and claims were checked against the session transcript and facilitator notes.