Skip to content

ai-risk-ontology

An ontology describing AI systems and their risks

URI: https://ibm.github.io/risk-atlas-nexus/ontology/ai-risk-ontology

Name: ai-risk-ontology

Classes

Class Description
Any None
Container An umbrella object that holds the ontology class instances
Entity A generic grouping for any identifiable entity.
        Action Action to remediate a risk
        AiEval An AI Evaluation, e.g. a metric, benchmark, unitxt card evaluation, a question or a combination of such entities.
                Question An evaluation where a question has to be answered
                Questionnaire A questionnaire groups questions
        AiEvalResult The result of an evaluation for a specific AI model.
        AiLifecyclePhase A Phase of AI lifecycle which indicates evolution of the system from conception through retirement.
                AiModelValidation AI model validation steps that have been performed after the model training to ensure high AI model quality.
                DataPreprocessing Data transformations, such as PI filtering, performed to ensure high quality of AI model training data.
        AiTask A task, such as summarization and classification, performed by an AI.
        BaseAi Any type of AI, be it a LLM, RL agent, SVM, etc.
                AiModel A base AI Model class. No assumption about the type (SVM, LLM, etc.). Subclassed by model types (see LargeLanguageModel).
                        LargeLanguageModel A large language model (LLM) is an AI model which supports a range of language-related tasks such as generation, summarization, classification, among others. A LLM is implemented as an artificial neural networks using a transformer architecture.
                AiSystem A compound AI System composed of one or more AI capablities. ChatGPT is an example of an AI system which deploys multiple GPT AI models.
                        AiAgent An artificial intelligence (AI) agent refers to a system or program that is capable of autonomously performing tasks on behalf of a user or another system by designing its workflow and utilizing available tools.
        Dataset A body of structured information describing some topic(s) of interest.
        Documentation Documented information about a concept or other topic(s) of interest.
        LargeLanguageModelFamily A large language model family is a set of models that are provided by the same AI systems provider and are built around the same architecture, but differ e.g. in the number of parameters. Examples are Meta's Llama2 family or the IBM granite models.
        License The general notion of a license which defines terms and grants permissions to users of AI systems, datasets and software.
        Modality A modality supported by an Ai component. Examples include text, image, video.
        Organization Any organizational entity such as a corporation, educational institution, consortium, government, etc.
                AiOffice The EU AI Office (https://digital-strategy.ec.europa.eu/en/policies/ai-office)
                AiProvider A provider under the AI Act is defined by Article 3(3) as a natural or legal person or body that develops an AI system or general-purpose AI model or has an AI system or general-purpose AI model developed; and places that system or model on the market, or puts that system into service, under the provider's own name or trademark, whether for payment or free for charge.
        Risk The state of uncertainty associated with an AI system, that has the potential to cause harms
        RiskConcept An umbrella term for refering to risk, risk source, consequence and impact.
        RiskControl A measure that maintains and/or modifies risk (and risk concepts)
        RiskGroup A group of AI system related risks that are part of a risk taxonomy.
        RiskTaxonomy A taxonomy of AI system related risks
Fact A fact about something, for example the result of a measurement. In addition to the value, evidence is provided.

Slots

Slot Description
actions A list of risk related actions
aimodelfamilies A list of AI model families
aimodels A list of AI models
aitasks A list of AI tasks
architecture A description of the architecture of an AI such as 'Decoder-only'
bestValue Annotation of the best possible result of the evaluation
broadMatch The property is used to state a hierarchical mapping link between two concept...
carbon_emitted The number of tons of carbon dioxide equivalent that are emitted during train...
closeMatch The property is used to link two concepts that are sufficiently similar that ...
concern Some explanation about the concern related to an AI risk
contextWindowSize The total length, in bytes, of an AI model's context window
datasets A list of data sets
dateCreated The date on which the entity was created
dateModified The date on which the entity was most recently modified
description The description of an entity
descriptor Annotates whether an AI risk is a traditional risk, specific to or amplified ...
detectsRiskConcept The property airo:detectsRiskConcept indicates the control used for detecting...
documents A list of documents
evaluations A list of AI evaluation methods
evidence Evidence provides a source (typical a chunk, paragraph or link) describing wh...
exactMatch The property is used to link two concepts, indicating a high degree of confid...
fine_tuning A description of the fine-tuning mechanism(s) applied to a model
gpu_hours GPU consumption in terms of hours
grants_license A relationship from a granting entity such as an Organization to a License in...
hasAiActorTask Pertinent AI Actor Tasks for each subcategory
hasDataset A relationship to datasets that are used
hasDocumentation Indicates documentation associated with an entity
hasEuAiSystemType The type of system as defined by the EU AI Act
hasEuRiskCategory The risk category of an AI system as defined by the EU AI Act
hasEvaluation A relationship indicating that an entity has an AI evaluation result
hasInputModality A relationship indicating the input modalities supported by an AI component
hasLicense Indicates licenses associated with a resource
hasModelCard A relationship to model card references
hasOutputModality A relationship indicating the output modalities supported by an AI component
hasPart A relationship where an entity has another entity
hasRelatedAction A relationship where an entity relates to an action
hasRelatedRisk A relationship where an entity relates to a risk
hasRiskControl Indicates the control measures associated with a system or component to modif...
hasTrainingData A relationship indicating the datasets an AI model was trained on
hasUnitxtCard A relationship to a Unitxt card defining the risk evaluation
id A unique identifier to this instance of the model element
isComposedOf Relationship indicating the some entity is composed of other entities (includ...
isDefinedByTaxonomy A relationship where a risk or a risk group is defined by a risk taxonomy
isDeployedBy A relationship indicating that an entity has been deployed by an organization
isDetectedBy A relationship where a risk, risk source, consequence, or impact is detected ...
isDistributedBy A relationship indicating that an entity has been distributed by an organizat...
isImportedBy A relationship indicating that an entity has been imported by an organization
isPartOf A relationship where an entity is part of another entity
isProvidedBy A relationship indicating the AI model has been provided by an AI model provi...
isResultOf A relationship indicating that an entity is the result of an AI evaluation
licenses A list of licenses
modalities A list of AI modalities
name A text name of this instance
narrowMatch The property is used to state a hierarchical mapping link between two concept...
numParameters A property indicating the number of parameters in a LLM
numTrainingTokens The number of tokens a AI model was trained on
organizations A list of organizations
performsTask relationship indicating the AI tasks an AI model can perform
phase Annotation whether an AI risk shows specifically during the training-tuning o...
power_consumption_w power consumption in Watts
producer A relationship to the Organization instance which produces this instance
provider A relationship to the Organization instance that provides this instance
relatedMatch The property skos:relatedMatch is used to state an associative mapping link b...
riskcontrols A list of AI risk controls
riskgroups A list of AI risk groups
risks A list of AI risks
supported_languages A list of languages, expressed as ISO two letter codes
tag A shost version of the name
taxonomies A list of AI risk taxonomies
text The question itself
training_data_preprocessing relationship indicating data preprocessing steps performed on training data s...
type Annotation whether an AI risk occurs at input or output or is non-technical
url An optional URL associated with this instance
validated_by A relationship indicating the model validation steps after AI model training
value Some numeric or string value
version The version of the entity embodied by a specified resource

Enumerations

Enumeration Description
AiSystemType
EuAiRiskCategory

Types

Type Description
Boolean A binary (true or false) value
Curie a compact URI
Date a date (year, month and day) in an idealized calendar
DateOrDatetime Either a date or a datetime
Datetime The combination of a date and time
Decimal A real number with arbitrary precision that conforms to the xsd:decimal speci...
Double A real number that conforms to the xsd:double specification
Float A real number that conforms to the xsd:float specification
Integer An integer
Jsonpath A string encoding a JSON Path
Jsonpointer A string encoding a JSON Pointer
Ncname Prefix part of CURIE
Nodeidentifier A URI, CURIE or BNODE that represents a node in a model
Objectidentifier A URI or CURIE that represents an object in the model
Sparqlpath A string encoding a SPARQL Property Path
String A character string
Time A time object represents a (local) time of day, independent of any particular...
Uri a complete URI
Uriorcurie a URI or a CURIE

Subsets

Subset Description