Humanitarian AI Wiki

Definitions, notes and sources for humanitarian AI terms.

Browse entries

78 entries

Definitions

Humanitarian AI

application of AI systems to tasks undertaken as part of humanitarian action

Definitions

Humanitarian action

organised action that supports people and communities affected by crisis while seeking to protect life, alleviate suffering and uphold dignity

Definitions

AI system

engineered system that generates outputs such as content, forecasts, recommendations or decisions for a given set of human-defined objectives

Definitions

AI model

machine-readable representation of knowledge

Definitions

AI output

any data or information returned by the AI system when processing a given input

Definitions

Generative AI

the class of AI models that emulate the structure and characteristics of input data in order to generate derived synthetic content

Definitions

Traditional AI

informal label for established AI approaches such as rule-based AI, classification and prediction models

Definitions

Rule-based automation

software process that applies predefined rules to choose an action when specified conditions occur

Definitions

AI judgement call

context-dependent choice or interpretation made by an AI system within a task when the response is not completely specified in advance

Definitions

Justified reliance on an AI output

decision to use an AI output for a defined purpose after its relevant claims, limits and permissions have been assessed by a responsible person

Definitions

Humanitarian task

defined unit of work undertaken within humanitarian action, with an intended result and a responsible actor

Definitions

Humanitarian AI use

particular application of an AI system to a defined task in humanitarian action, with an identifiable purpose, setting and responsible actor

Definitions

AI capability

kind of processing an AI system can perform, identified by the result it produces rather than the humanitarian purpose it serves

Definitions

Humanitarian function

purpose served by a task within humanitarian action, regardless of the AI capability used to perform it

Definitions · C-PRED

Prediction

AI capability that estimates a numeric value or probability for an outcome using available information

Definitions · C-CLASS

Classification

AI capability that assigns an item to one or more predefined categories

Definitions · C-GROUP

Clustering

AI capability that groups items according to patterns of similarity without relying on a fixed set of category labels

Definitions · C-ANOM

Anomaly and change detection

AI capability that flags an unusual item or a change relative to an expected pattern

Definitions · C-EXTRACT

Perception and extraction

AI capability that identifies information in unstructured text, images, audio or video and makes it available for further use

Definitions · C-SYNTH

Synthetic data generation

AI capability that produces artificial records intended to resemble selected features of other data for a defined purpose

Definitions · H-ASSESS

Understanding needs

humanitarian function concerned with identifying needs, priorities and information gaps among people affected by crisis

Definitions · H-ANTICIPATE

Preparing for change

humanitarian function concerned with using information about possible future conditions to prepare a response

Definitions · H-PLAN

Planning a response

humanitarian function concerned with comparing courses of action and preparing resources for a proposed response

Definitions · H-DELIVER

Delivering services

humanitarian function concerned with providing or following up assistance or services for people affected by crisis

Definitions · H-SUPPLY

Moving goods and services

humanitarian function concerned with arranging the supply, routing or scheduling needed to provide assistance

Definitions · H-LEARN

Monitoring and learning

humanitarian function concerned with examining implementation, feedback and results to inform changes in a response

Definitions · H-SUPPORT

Organisational support

humanitarian function concerned with internal drafting, coordination or record preparation that supports humanitarian work

Definitions

Person affected by a humanitarian AI use

person whose rights, safety, access to assistance or experience of a humanitarian response could be influenced by a particular humanitarian AI use

Definitions

Affected population

group of people affected by a crisis or its consequences who may need assistance and may participate in humanitarian action

Definitions

Participation

involvement of people in an affected population in humanitarian action, with opportunities to influence decisions that concern them

Definitions

Do No Harm

approach to humanitarian action that seeks to anticipate and reduce avoidable harm arising from an intervention

Definitions

Community engagement

ongoing process of listening to, sharing information with and involving an affected population in humanitarian action

Definitions

Needs assessment

systematic process of identifying and analysing the needs, priorities and capacities of an affected population to inform humanitarian action

Definitions

Accountability to affected people

responsibility of humanitarian actors to use power responsibly and be answerable to an affected population for decisions and actions that concern them

Definitions

AI system permission

scope of information an AI system is authorised to access and actions it is authorised to perform in a defined humanitarian AI use

Definitions

Human review

examination by a person of an AI output or process against stated criteria before it is relied on

Definitions

Verification of an AI output

check of a consequential AI claim or output against evidence independent of that AI output

Definitions

Human approval

recorded authorisation by a person with the required authority for a specified humanitarian AI use or consequential action to proceed

Definitions

Risk assessment of a humanitarian AI use

structured examination of possible harm from a particular humanitarian AI use, its context and the capacity to prevent or respond to that harm

Definitions

Human oversight

arrangement through which responsible people can observe, challenge, intervene in or stop a humanitarian AI use during its operation

Definitions

Human fallback

arrangement for people to continue or restore a humanitarian task when an AI system cannot be used reliably

Definitions

Incident involving a humanitarian AI use

event in which a humanitarian AI use causes, contributes to or reveals a failure or potential harm requiring investigation or response

Definitions

Prompt

input to a generative AI system that supplies a task, instructions or information for processing

Definitions

CHEF brief

prompt structured by the AidGPT CHEF method to state a role, method, task context and desired result

Definitions

Standing instructions

instructions retained in a saved setup to guide an AI assistant across related tasks

Definitions

Project

saved setup of instructions and reference material for a recurring family of tasks with an AI assistant

Definitions

Skill

saved method that an AI assistant can apply to a task when its application makes the method available

Definitions

AI assistant

AI system presented through an application to help a person carry out tasks through interaction

Definitions

Connector

route through which an AI assistant can reach data or actions in another system

Definitions

Tool

action or function made available to an AI assistant by its application or a connector

Definitions

Prompt injection

An attack which exploits the concatenation of untrusted input with a prompt constructed by a higher-trust party such as the application designer

Definitions

Hallucination

AI output that presents inaccurate factual material or an unsupported claim as established

Definitions

Unsupported claim

factual claim in an AI output for which the evidence needed to justify it has not been established

Definitions

Self-review

checking pass in which an AI assistant examines its own earlier output against stated criteria

Definitions

Reviewable handover

record accompanying an AI output that identifies work done, checks performed, unresolved issues and the next human decision

Definitions

Workflow

sequence of actions, checks and decisions through which a person and an AI assistant complete a task

Definitions

Short-term memory

information available to an AI assistant within its current chat or working context

Definitions

Project note

checked record of agreements, current files, progress and open issues kept for a continuing workflow

Definitions

Responsible refusal

decision to pause, redesign or stop a humanitarian AI use when conditions for responsible use cannot be met

Definitions

Decision gate

defined point at which authorised people assess whether a proposed humanitarian AI use may proceed, continue, change or stop

Definitions

Community co-design of AI

collaborative process in which people affected by a humanitarian AI use have defined influence over decisions about its purpose, design, deployment and evaluation

Definitions

Right to know about AI use

expectation that people affected by a humanitarian AI use can know when and how AI shapes decisions concerning them and how to question or contest those decisions

Definitions

AI transparency card

cumulative record of a humanitarian AI use, its assessments, decisions, community engagement and monitoring

Definitions

Technical assurance of an AI use

repeated, documented examination of a humanitarian AI use against agreed performance, safety and risk criteria before and during deployment

Guides

How the definitions are written

Read the ISO-based entry method and follow the concept map.

Definitions · C-GEN

Content generation and transformation

Producing new or revised text, speech, images or translations.

Definitions · C-RANK

Ranking and decision support

Putting items in order or suggesting a choice.

Definitions · H-ACCESS

Access to assistance

Registration support, referrals and assistance-related case review.

Definitions · H-COMMUNICATE

Communicating with people

Translation, information services and accessible communication.

Definitions · H-LISTEN

Feedback and remedy

Sorting feedback and supporting complaints handling.

Guides

Record a humanitarian AI task

Use an editable record to connect the task, evidence, separate assessments and human decision.

Guides

Humanitarian AI: start here

A practical route from describing an AI use to examining its consequences.

Research

How this work continues the earlier research

The connection between the humanitarian reports, DCI/GIZ work and this collection.

Guides

Classify a humanitarian AI use

Describe capabilities, humanitarian purpose and permissions separately.

Guides

Assess risk and record human decisions

Keep DCI and SAFE AI assessments distinct and connect them to accountable decisions.

Guides

Connect the record to SAFE AI

Link the taxonomy to existing governance records throughout a system’s life.

Examples · C-GEN / H-COMMUNICATE

Example: translate a service notice

An invented example for testing the taxonomy and discussing human checks.

Examples · C-EXTRACT / C-CLASS / C-RANK / H-LISTEN

Example: sort incoming complaints

An invented example for testing the taxonomy and discussing human checks.

These are working entries for review. How definitions are written and research history and authorship.