[NEW]: AI model open-ness

[bact]

Specs

AI

New Concept(s)

Transferred from #82 and referred to some of the proposed concepts there (like new:OpenWeights).

Term: (used as unique ID) can be a classification like ai:MOFClass1Model, ai:MOFClass2Model, MOFClass3Model; or a set of components as criteria for classification - as proposed in #82 (new:OpenArchitecture, new:ClosedArchitecture, etc)

Alternatively, this can be modelled as a property ai:hasOpennessLevel for ai:Model.

Label: (human label) "Model Openness"

Definition: (definition or description of term)

Parent term: (existing term in DPV) ai:Model

Source: (url or reference)

• Model Openness Framework
  • paper - https://arxiv.org/abs/2403.13784
  • assessment tool - https://isitopen.ai/
  • blog - https://lfaidata.foundation/blog/2024/04/17/introducing-the-model-openness-framework-promoting-completeness-and-openness-for-reproducibility-transparency-and-usability-in-ai/

Usage note: (considerations when using the term e.g. scope, caution)

• A data subject may agree to allow the use of their data for AI training on a condition that the AI model must be Class I (Open Science) according to Model Openness Framework.

--

below is a text copied from #82

On "open-ness", as discussed in https://harshp.com/dev/dpv/ai-training, the "Model Openness Framework" (see ref below) provides a classification of model openness:

• Class III - Open Model
• Class II - Open Tooling
• Class I - Open Science

Class III is least open, Class I is the most open.

DPV already provides some concepts that can be used to describe components within each Class.

Class III will be related to these concepts:

• new:OpenArchitecture / new:ClosedArchitecture (Model Architecture)
• new:OpenWeights / new:ClosedWeights / new:PartiallyOpenWeights (Model Parameters - final checkpoint)
• tech:Documentation (Technical Report, Evaluation Results, Model Card, Data Card)
  • also eu-aiact:TechnicalDocumentation

Class II will be related to these concepts:

• new:OpenSource / new:ClosedSource
• ai:ValidationData / ai:TestingData (Evaluation Data)

Class I will be related to these concepts:

• tech:Documentation (Research Paper)
• new:OpenWeights / new:ClosedWeights / new:PartiallyOpenWeights (Model Parameters - intermediate checkpoints)