Taxonomy — Print - DTPR for AI

Accessibility

Ensures that everyone has equal access to a space or a service.

Arts & Culture

Enables artistic and/or cultural expression.

Border & Immigration

Supports immigration, asylum, or border-crossing decisions or processing — for example, border-control facial recognition, visa adjudication, asylum risk assessment.

Commerce

Enables the buying and selling of goods and services.

Content Moderation

Filters, ranks, or removes user-generated content based on policies — for example, comment moderation on a public-facing kiosk, image filtering, or hate-speech detection.

Dining

For providing food or meal services.

Ecology

Supports the measurement or monitoring of the natural environment.

Education & Learning

Supports teaching, learning, assessment, or educational guidance — for example, tutoring assistants, automated grading, or library and museum recommenders.

Eligibility & Public Benefits

Determines eligibility for, or distribution of, government benefits and essential public services — for example, welfare allocation, housing assistance, child-services risk scoring.

Employment & Work

Supports hiring, scheduling, performance evaluation, or workplace operations — for example, resume screening, shift assignment, productivity monitoring.

Energy Efficiency

Reduces energy use and/or helps conserve energy.

Enforcement

Used for enforcement of rules or regulations.

Entry

Supports authentication or validation in order to access a space or a service.

Financial Services

Supports lending, insurance, fraud detection, or other financial decisions about people — for example, credit scoring, insurance underwriting, transaction-fraud flagging.

Fire & Emergency

Supports services that ensure public safety and health related to emergencies.

Environmental Health

Measures or monitors aspects of the physical environment that affect human health — air quality, water quality, radiation, noise, or workplace exposures. Distinct from Healthcare, which supports clinical care for individuals.

Healthcare

Supports clinical care for individuals — for example, diagnosis assistance, treatment recommendations, patient triage, or health chatbots. Distinct from Environmental Health, which monitors the physical environment.

Inform

Supports general-purpose information assistance — concierge, FAQ, signage, or guidance about a service. Distinct from Wayfinding & Services, which covers spatial orientation and routing within a place.

Logistics

Supports the movements of goods or materials.

Marketing & Personalization

Targets advertising, personalizes recommendations, or profiles audiences for marketing — for example, in-store loyalty offers, personalized signage, behavioural ad targeting.

Mobility

Supports how people and materials move around.

Planning & Decision-making

Supports forecasting and policy-impact analysis at the population or system level — for example, urban planning, infrastructure planning, or measuring the impact of a policy decision. Distinct from Risk Assessment & Triage, which scores individuals.

Research & Development

Supports a deployed research instrument — a system that studies a place, population, or environment as part of a research program. Distinct from production AI that delivers a service to the public.

Risk Assessment & Triage

Scores or prioritizes individuals or cases for further attention or action — for example, recidivism scoring, hospital triage, child-welfare risk flags. Distinct from Planning & Decision-making, which forecasts at a population or system level.

Safety & Security

Supports preventive risk reduction in physical environments — for example, fire safety, home security, or ensuring safe passage in airports or on roads. Distinct from Enforcement (rule application) and from Fire & Emergency (incident response).

Social

For interacting with another person or a group.

Translation & Language Access

Translates speech, text, or signs to make a place or service accessible across languages — for example, real-time interpretation at a service counter, multilingual signage.

Waste Management

Supports the handling and disposal of waste, including as recyclables, compost and hazardous materials.

Water Efficiency

Reduces water use and/or helps conserve water.

Wayfinding & Services

Enables navigation of a location and its amenities and services.

Context — Role

How this organization participates in the deployment of the AI system.

Vendor	The organization that builds, supplies, or licenses the AI system.
Deployer	The organization that deploys and operates the AI system in this place.

Institution

The entity that is responsible and accountable for this data collection activity

Organization

The entity that is responsible and accountable for this data collection activity

Context — Autonomy

How much this mode decides and acts on its own. The element name says what the mode does; this label says who chooses and who carries out the result.

Human decides	This mode suggests; a person decides what to do next. The AI is always advisory — a human is in the loop on every decision. Example: a triage tool ranks cases for a clinician who chooses which to see first.
Human executes	This mode decides; a person carries out the result. Example: an optimizer plans the day’s trash-collection routes, and drivers run them.
Autonomous	This mode decides and acts on its own. No person reviews each decision or carries out the resulting action.

Acting (Agentic AI)

Acts — plans a goal into steps and carries them out, calling tools, browsing, scheduling, or composing multi-step workflows on the user's or operator's behalf. The output is an action taken in software, not a piece of content returned. Examples: booking a meeting, filing a return, escalating a ticket, running a campaign loop.

Deciding (Analytical AI)

Decides — predicts, classifies, scores, or ranks from structured data. The system reads numeric or categorical inputs and returns a label, a score, or a ranking. It does not generate new content. Examples: a credit-risk score, a fraud flag, a queue-length forecast, a recommendation rank.

Creating (Generative AI)

Creates — produces new text, images, audio, video, or code that did not exist before this system ran. The output is content authored by the system, not a label or score about existing data. Examples: drafting a paragraph in response to a prompt, generating an image from a description, composing audio, completing a snippet of code.

Sensing (Perceptive AI)

Senses — turns raw camera, microphone, or scanned-document signals into structured detections, transcriptions, or extracted fields. The system reads pixels, audio, or document images and returns things the rest of the pipeline can use. Examples: detecting a vehicle in a camera frame, transcribing a spoken sentence, reading the fields off a paper form.

Moving (Physical AI)

Moves — acts on the physical environment. The system drives motors, valves, signals, or other actuators and changes something in the world. Examples: steering a robot, changing a traffic-signal phase, opening a gate, adjusting an HVAC valve, releasing a parking-garage barrier.

Understanding (Semantic AI)

Understands and remembers — pulls meaning from text or speech, finds related ideas, and grounds the system in saved context. The system reads language and returns matches, intents, summaries, or links to relevant prior knowledge. It does not generate new content. Examples: matching a customer's question to the right help article, retrieving the policy clause that applies to a case, recognizing the topic of a recorded sentence.

Loss of autonomy

The risk that the AI system limits a person's ability to make their own decisions, control their identity, or choose freely. This includes risks of restricted access to alternatives, deceptive design that nudges decisions toward a particular outcome, and lack of meaningful consent. Mitigations may include opt-out mechanisms, transparent recommendations, defaults that protect agency, and human review of consequential decisions.

Civil liberties harm

The risk that the AI system compromises fundamental human rights and civil liberties — speech, assembly, movement, due process, or freedom from arbitrary surveillance. This includes risks from mass facial recognition, predictive policing, censorship of lawful expression, and chilling effects on protest. Mitigations may include strict purpose limits, judicial or independent oversight, narrow data retention, and human-rights impact assessments before deployment.

Environmental harm

The risk that the AI system causes environmental degradation through energy consumption, water use, e-waste, carbon emissions, or resource extraction for hardware. This includes risks from training-time compute, always-on inference, cooling demands of data centers, and rapid hardware churn. Mitigations may include energy and water efficiency targets, transparent reporting of compute footprint, hardware-lifecycle planning, and renewable-energy procurement for data-centre operations.

Financial & business harm

The risk that the AI system causes financial losses to individuals or organizational damage through erroneous pricing, denied services, fraud, or market manipulation. This includes risks from automated credit decisions, dynamic-pricing inequities, AI-enabled scams, and supply-chain disruption from misuse. Mitigations may include affordability guards, fairness audits of pricing models, fraud-detection layers, and clear redress procedures.

Physical harm

The risk that the AI system's outputs lead to physical injury to individuals or damage to property. This includes risks from autonomous vehicles, robotic actuation, faulty navigation guidance, or critical-infrastructure errors that put bodies or property in harm's way. Mitigations may include rigorous safety testing, fail-safe defaults, geofencing of hazardous behavior, and human oversight of high-stakes physical actions.

Political & economic harm

The risk that the AI system manipulates political discourse, interferes with elections, concentrates market power, or damages public institutions. This includes risks of synthetic media targeting voters, opaque ad-targeting, competitive distortion from monopoly access to data and compute, and erosion of institutional trust. Mitigations may include disclosure of synthetic content, antitrust attention to AI markets, audit access for regulators, and political-ad transparency requirements.

Psychological harm

The risk that the AI system causes emotional or mental-health impairment, directly or indirectly. This includes risks from addictive feedback loops, distress caused by surveillance, harassment via generated content, and anxiety from automated denials. Mitigations may include content warnings, well-being safeguards, age-appropriate design, accessible support channels, and rate limits on engagement-maximizing behavior.

Reputational harm

The risk that the AI system damages the reputation of individuals, groups, or organizations through misidentification, false categorization, or stigmatizing labels. This includes risks of inaccurate face recognition, defamatory generated content, and unfair public scoring that follow people across contexts. Mitigations may include accuracy thresholds before public-facing labels are applied, human review of high-impact identifications, takedown procedures, and clear rights to correction.

Societal & cultural harm

The risk that the AI system harms communities or culture through erosion of trust, loss of linguistic and cultural diversity, or unhealthy dependency on opaque systems. This includes risks of misinformation at scale, homogenization of cultural expression, replacement of local knowledge, and degradation of public-information ecosystems. Mitigations may include multilingual support, partnerships with affected communities, provenance signals on generated content, and investment in local-language and local-context models.

Right to Access

The right to request and receive information about what personal data an AI system has collected about you, how this data is being used, and what decisions have been made using this information. This includes the right to obtain a copy of your data in a readable format.

Right to Algorithmic Transparency

The right to be informed in plain language about how an AI system works in general — the logic involved, the significance of the processing, and the likely consequences for the people it affects. This is the right to understand the system itself; the right to an explanation of a specific decision made about you is a separate right.

Right to Be Forgotten

The right to have your personal data deleted from an AI system when the data is no longer needed for its original purpose, when you withdraw consent, or when there is no legitimate interest in continuing to process it. Where technically feasible, this also includes having the system "unlearn" information derived from your data. This is a right of deletion; the right to stop processing without deletion is a separate right.

Right to Contest

The right to challenge a decision an AI system has made about you — to provide additional information, express your point of view, and have the decision reconsidered based on your input. This is a remedy after a decision has been made; the right to a human review before a decision is acted on is a separate right.

Right to Correct Your Data

The right to have inaccurate or incomplete data that an AI system holds about you corrected or completed. This is separate from the right to access your data (which is the right to read it) and the right to be forgotten (which is the right to delete it). When the original data is corrected, downstream decisions made from the faulty data should be revisited.

Right to an Explanation of a Decision

The right to a clear and meaningful explanation of how an AI system contributed to a specific decision affecting you — what role the system played, the main factors that led to the outcome, and what it means for you. This right covers the decision actually made about you; the right to be informed about how the system works in general is a separate right.

Right to Non-discrimination

The right to be free from discriminatory treatment by AI systems based on protected characteristics such as race, gender, age, religion, disability, or sexual orientation. Organizations must implement and demonstrate appropriate technical and organizational measures to prevent discriminatory outcomes from their AI systems.

Right to Stop Processing

The right to ask an organization to stop using your personal data, even when the data itself is not deleted. This is separate from the right to be forgotten (which deletes the data) and from the right to purpose limitation (which constrains the original stated purposes). It is the user-invoked stop-now affordance — your data may stay on file, but no further processing should happen against it.

Right to Purpose Limitation

The right that your data is only used for the specific purposes that were stated when it was collected. The organization must declare and document those purposes before collection begins, and is barred from later using your data for new, unrelated purposes without your knowledge or consent. This is an upstream limit on what use is ever allowed; the right to stop processing already underway is a separate right.

Right to a Human Review

The right to have a person — not an automated system — make or meaningfully reconsider a decision affecting you. This is a path to a non-automated decision, separate from the right to challenge a decision that has already been made. The reviewing person must have the authority and information to actually change the outcome.

Right to Be Informed of AI Use

The right to know that an AI system is in operation before any decision affecting you is made. This includes being told when you are interacting with an automated system rather than a person, when content is AI-generated, and when biometric or emotion-recognition technology is in use. This is the precondition for every other right in this category — you cannot exercise rights you do not know apply to you.

Context — Personal Information

How identifiable the data is when this system processes it. The element name says what the data is about; this colour says how directly it identifies a person.

Anonymized data	Data about people with the link to who is broken. Stripped of identifiers, blurred, aggregated, or noised so this system can’t reasonably tie a record back to an individual.
Pseudonymous data	Each person’s data is tied to a token (hash, ID, template) that lets this system recognise the same person across events, but the token itself doesn’t reveal a name. Reidentification is possible with extra information.
Identifiable data	The data either contains a direct identifier (name, address, account name, recognisable face or voice, plate number) or carries a token this system uses to look up legal identity during processing.

About a measurement

Sensor readings — air quality, temperature, sound level, energy, water flow. The AI takes in numeric measurements from physical sensors, usually with no person attached. Examples: PM2.5 readings from an air-quality monitor, decibel levels from a sound sensor, kilowatt-hours from a meter.

About a place

Locations, routes, surroundings, or floor plans. The AI takes in data describing where something is. Examples: GPS coordinates from a phone, foot-traffic counts on a sidewalk, a map of building corridors.

About behaviour

Taps, choices, paths walked, dwell time, purchases, or queries. The AI takes in records of what people did. Examples: tapped destinations on a wayfinding kiosk, items added to a cart, a search query typed into a public terminal.

Biometric

Faces, fingerprints, voice prints, gait, gestures, gaze, posture. The AI takes in biological signals from a person's body. Examples: a face captured at a turnstile, a voice recorded by a kiosk microphone, gait analyzed by an overhead camera.

A decision about you

An eligibility, classification, ranking, or yes/no that another system already made about a person. The AI takes in this prior decision and uses it as input. Examples: a credit score from another model, an access-allow flag from upstream, a triage class assigned earlier in the pipeline.

Generated content

Text, images, audio, or video that another AI made. The AI takes in this synthetic content and processes it further. Examples: an LLM-written summary fed into a translator, a generated photo passed to a moderation model, a text-to-speech clip routed to an audio system.

Operational data

Schedules, routes, budgets, occupancy counts, public records, or other administrative data — not about any one person. The AI takes in data describing how a place or service runs. Examples: school occupancy by hour, trash-collection routes, a transit timetable, a department budget.

A physical action

A signal that something in the physical world changed — a door opening, a light turning on, an alert sounding. The AI takes in evidence of these changes from upstream actuators or sensors. Examples: a turnstile-unlock event, an HVAC adjustment recorded by a sensor, a public-address alert played upstream.

A recommendation or prediction

A suggestion, forecast, risk score, or ranking produced earlier by another system. The AI takes in this advisory output and uses it. Distinct from a binding decision because it advises rather than determines. Examples: a forecasted demand curve, a risk score from a prior model, a recommended route from a navigation service.

Sensitive personal information

Health, finances, beliefs, sexuality, immigration status, or other personal information that carries legal and social risk. The AI takes in this kind of protected data. Examples: insurance eligibility at a clinic kiosk, a benefits-application status, a self-declared pronoun.

Affect & Emotion Analysis

AI systems that infer emotional state, mood, or sentiment from text, speech, facial expressions, or physiological signals. Used to gauge crowd sentiment, evaluate service experience, or screen for distress. The accuracy and cultural fairness of affect inference is contested, and some jurisdictions restrict its use in workplaces and public services.

Anomaly Detection

AI systems that flag unusual events, patterns, or values that depart from a learned baseline of normal behavior. Used for equipment-failure prediction, fraud screening, intrusion detection, water-leak monitoring, and unusual crowd-flow alerts. Outputs are typically thresholded scores rather than direct decisions.

Biometric Recognition

AI systems that identify or verify a specific person from biological signals — face, fingerprint, voice, iris, or gait. Distinct from generic computer vision because the output is tied to a named individual. Many jurisdictions restrict or prohibit biometric recognition in publicly accessible spaces.

Classification & Prediction

AI systems that assign a label to an input or estimate a numeric outcome — including classifiers (spam vs. not spam, risk tier), regressors (predicted demand, expected wait), and time-series forecasters (next-hour ridership, energy load). Built from labeled historical data using statistical or machine-learning models.

Clustering & Segmentation

AI systems that group people, places, or events into clusters based on similarity, without using pre-defined labels. Used to segment audiences, mobility patterns, energy-use profiles, or service-demand zones. Outputs are cluster assignments that downstream systems may treat as categories — even though clusters are inferred, not authored.

Computer Vision

AI systems that interpret images or video — counting people, detecting objects, reading text in pictures, tracking motion, or segmenting scenes. Common in occupancy sensing, traffic and crowd monitoring, accessibility assistance, and signage inspection. Distinct from biometric recognition, which identifies specific individuals.

Language Models

AI systems that read, write, summarize, translate, or hold a conversation in human language. Includes large language models (LLMs), chatbots, and generative-text systems. They work by predicting the most likely next tokens given the input and any prior context.

Optimization

AI systems that search for the best assignment, route, schedule, or allocation under constraints. Includes vehicle routing, transit signal timing, energy and HVAC scheduling, staff rostering, and resource allocation. Works by exploring possibilities to maximize or minimize a stated objective such as cost, time, or emissions.

Privacy-Preserving Transformation

AI systems whose primary job is to transform input data so identifying details are removed, blurred, or replaced before any downstream use. Includes face blurring, license-plate masking, tokenization, k-anonymization, differential privacy, and synthetic-data generation. Acts as a barrier between raw collected data and the data used for analysis or decisions.

Recommendation & Ranking

AI systems that suggest or order items for a person — services, listings, news, wayfinding options — based on preferences, behavior, or similarity to other people. Includes collaborative filtering, content-based recommenders, and learned ranking. Personalizes what each person sees from the same underlying catalog.

Search & Retrieval

AI systems that find and surface relevant content from a body of documents, records, or media in response to a query. Includes semantic search, vector retrieval, and retrieval-augmented generation (RAG) pipelines that feed relevant passages into a language model. Used in document QA, public-records lookup, and library or archive search.

Speech & Audio

AI systems that convert between spoken words and other formats, or that classify non-speech sound. Includes speech-to-text (transcription), text-to-speech (synthetic voice), keyword spotting, and audio event detection. Voice-as-identity (recognizing who is speaking) belongs in Biometric Recognition.