A practical module on using AI to strengthen humanitarian preparedness, coordination and decision-making in conflict and disaster settings — without replacing human judgement.
Four non-negotiable conditions for any AI use in civil-military engagement. They frame every tool, every workflow and every output in this module.
Safety of staff, partners and the people we serve takes priority over speed, efficiency or analytical depth.
Humanity, neutrality, impartiality and independence are encoded into how we choose, configure and deploy AI tools.
AI accelerates analysis and amplifies expertise. Decisions, accountability and context remain firmly with the operator.
Bias, hallucination, data leakage and adversarial misuse are surfaced, named and managed throughout the workflow.
These four conditions are cumulative, not selective. An AI workflow that protects personnel but erodes neutrality is not acceptable. An output that is operationally effective but cannot be explained to an affected population is not acceptable. The principles are the gate before every step that follows.
A short conceptual map before we begin. Where a system sits on these two axes tells you what it is for — and where the first principles will press hardest. Click any node to see how it fits the workflow.
The practical session walks each participant through a complete AI-assisted workflow on the fictional Shapeland scenario. Each step builds on the previous one.
Download the Shapeland situation briefing as raw source material.
Use NotebookLM to extract insights from the briefing.
Turn insights into deployable tools with agentic systems.
Every participant works from the same source material: the Shapeland Situation Briefing. It is a self-contained, fictional briefing pack covering political context, humanitarian needs, security posture and the actors in play.
Download the PDF. We will load it into NotebookLM in the next step.
Participants load the Shapeland briefing into NotebookLM and use it as a grounded research workspace. Outputs stay anchored to the source document — every claim links back to a citation in the briefing.
What this gives you in the training:
The NotebookLM output becomes the input for agentic coding and reasoning systems. Participants prompt them to build small, single-purpose tools that an operator can actually use in the field — not slide decks about the scenario, but working artefacts.
Generate a clickable network of civilian, military and non-state actors with their relationships, mandates and last-known positions — refreshable as the briefing evolves.
Using the attached Shapeland situation briefing as the only source, extract every named actor (state, military, non-state armed, civilian authority, humanitarian agency). For each actor return: - name, short alias, type - area of operation - stated mandate - relationships to other actors (ally / adversary / neutral / mediator) - source paragraph Output as valid JSON, then generate a self-contained HTML file using d3-force that renders the network. Colour nodes by type. Flag any relationship that is inferred rather than explicit.
A single web app that reshapes the briefing for the user's role — medical lead, security advisor, liaison officer — surfacing only the parts they need, with the rest one click away.
Build a single-page web app from the attached Shapeland briefing. No external network calls; everything runs offline. Top of page: role selector with options - Medical Lead - Security Advisor - Liaison Officer - Logistics Lead For each role, surface the three sections of the briefing most relevant to that role and collapse the rest behind "Show more". Next to every section, cite the source paragraph. Add a "Reset role" link and a print stylesheet. Output: one HTML file plus one CSS file. No build step.
For a specific access dilemma, generate counterpart profiles, interests, BATNA estimates and a sequenced talking-points playbook — always grounded in the source briefing.
You are preparing me for a negotiation with the northern administration over medical-convoy access through the corridor. The Shapeland briefing is your only source. Draft, in this order: 1. Counterpart profile — mandate, decision authority, public posture 2. Likely interests and red lines, with reasoning 3. My estimated BATNA and theirs 4. Three opening positions ordered by escalation 5. Talking-points playbook, with anticipated objections and humanitarian-principle-aligned responses Stay neutral, impartial and grounded in the briefing. Flag any claim that is inferred rather than sourced.
A dashboard that pulls timeline events, access incidents and humanitarian indicators into one view, with the ability to ask follow-up questions and export shift handover notes.
From the attached Shapeland briefing, build a self-contained HTML situational dashboard. Required views: - timeline of events from D-90 to today, colour-coded by category - access-status heat-grid by district (open / conditional / closed) - one card per humanitarian indicator (food, water, medical, protection) with trend arrow and last-update timestamp - a follow-up question field that answers only from the briefing Add an "Export shift handover" button that produces a printable summary of the last 24 hours. Accessible, keyboard navigable, no external network calls.
Before any AI-generated tool leaves the practicum, participants run it through a short, repeatable review — anchored back to the four first principles.
Every operational claim can be traced to a specific paragraph of the briefing or a named source. No untraceable assertions reach an operator.
The output is reviewed against humanity, neutrality, impartiality and independence — and against the duty of care toward affected populations.
Personal data, location data and sensitive identifiers are stripped or minimised before any data leaves a controlled environment.
A named human operator validates the tool's fitness for purpose. AI proposes; a person decides.
The AICME module can be delivered as a stand-alone half-day practicum or as part of the full SHAPE programme. We tailor the scenario and tool stack to your operating context.