A Provisional Metadata Framework for Human/Generative AI Output

(Nota Bene: Since writing this document in November of 2024, I realized I made two mistakes. First, I buried provenance too deeply, when in reality, provenance is the core issue—everything else rides on its back. Second, I was too stuck in a classical data framework, thinking in terms of flat or relational structures, when what’s needed is a more dynamic, hybrid approach.

After working deeply with generative AI and engaging with blockchain and NFT-type storage systems, I’ve rethought MARTI as a flexible framework that bridges traditional metadata systems and decentralized provenance models. The goal is to ensure provenance, authorship, and accountability are built into AI from the start.

This document is written for fellow data nerds—a raw sandbox for working through AI. I haven’t yet translated it into accessible language, because I’m still deep in the process of thinking it through. Right now, this is just the inside of my brain in real time. If you want to help, critique, break, or refine it, I’m all for it. Let’s push this forward together. CBZ 02/16/2025)

Intro

The emergence of generative AI marks a transformative moment in human creativity, problem-solving, and knowledge-sharing. MARTI (Metadata for AI Responsibility, Transparency, and Integrity) is a provisional metadata framework designed to navigate this new landscape, offering a standardized yet adaptable approach to understanding, describing, and guiding the outputs of human-AI collaboration—and even those generated autonomously by AI.

At the heart of MARTI lies a robust object model—a modular structure that organizes metadata into reusable, interoperable components. This model ensures transparency, traceability, and ethical integrity, making it the cornerstone of the MARTI framework.

MARTI is not just an architecture for describing AI output, but it offers a way of structuring policy and a possible foundation for a new literacy. This is not about teaching every individual to code or engineer prompts. It’s about empowering humanity to collectively understand, describe, and guide everything we make with AI, ensuring accountability, transparency, and ethical integrity at every step.

MARTI is a framework for creating structured, standardized documentation that is attached to or embedded in AI-generated content. This documentation, or metadata, can be created by people collaborating with AI tools to produce content. Additionally, AI processes themselves can generate and embed metadata into their outputs, ensuring transparency, traceability, and accountability at every stage of content creation.

MARTI also offers a variety of potentially transformative business applications.

The Object Model Approach

MARTI’s object model structures metadata into core components:

Attributes: Descriptive fields like Title, Creator, Provenance, and Rights.
Relationships: Parent-child hierarchies and cross-object links.
Behaviors: Dynamic actions such as versioning and cascading updates.

This object model is key to understanding MARTI’s flexibility and applicability across disciplines.

A Framework for Harmonization, Anticipation, and Interpolation

At its core, MARTI is a bridge. It harmonizes with existing metadata standards like the Content Authenticity Initiative, Anthropic’s Responsible Scaling Policy, and the W3C’s PROV. It anticipates the needs of future standards, laws and practices, such as those proposed by the Coalition for Networked Information (CNI), The EU Artificial Intelligence Act, and Making Data FAIR.

MARTI interpolates across disciplines and domains. While we recognize many current initiatives, there are undoubtedly efforts we have yet to encounter. MARTI remains open, flexible, and eager to integrate with these as they emerge.

MARTI: Understanding Authorship, Responsibility, and the Hybrid World

MARTI offers more than a framework for describing AI outputs—it provides a way to understand and define authorship and responsibility in the hybrid landscape of human-AI collaboration. In a world where objects range from static images to complex large language models, MARTI helps delineate and inventory these diverse creations.

By applying MARTI’s structure, we can:

Clarify authorship and responsibility, whether outputs are generated solely by humans, collaboratively with AI, or autonomously by AI.
Differentiate between types of objects in this new ecosystem, from individual creations like static images to systems as complex as large language models.
Create an inventory of this hybrid world, offering a coherent way to catalog and make sense of the objects and relationships emerging in this era of generative AI.

This capacity to understand and document the outputs of human-AI interaction ensures that we can navigate this complex world responsibly and transparently.

Why MARTI Matters

MARTI exists to ensure that transparency, accountability, and ethical responsibility are not optional in the generative AI era—they are foundational. The framework benefits diverse stakeholders:

Artists and Creators gain tools for attribution and provenance.
Policymakers and Ethicists receive structures to guide responsible AI use.
Developers, Programmers, and Engineers access adaptable metadata systems for transparency and compliance.
Educators and Students find a framework for studying and navigating AI-human collaboration.

Ultimately, MARTI is about trust. It helps build a world where AI outputs can be trusted, understood, and responsibly guided.

Provisional and Collaborative by Design

MARTI is a draft, a starting point for exploring and shaping the intersection of human creativity and machine intelligence. It is open and standardized yet allows for local, domain-specific, and proprietary adaptations. This duality—rigor and flexibility—ensures MARTI’s broad applicability.

Call to Action: Join Us in Testing and Shaping MARTI

The MARTI (Metadata for AI Responsibility, Transparency, and Integrity) project is seeking a permanent home as a joint initiative between a university and a technology company. This public-private partnership will provide the academic rigor and technological expertise needed to develop robust metadata standards for AI ecosystems.

MARTI Needs a Home

If you are a researcher, technologist, artist, or advocate for ethical AI, join us in shaping the future of metadata. Learn more and get involved.

Let’s make MARTI the framework that ensures humanity remains a responsible steward of AI’s extraordinary potential.

MARTI Framework: Provisional Documents Collection

MARTI is not just an architecture for describing AI output, but it offers a way of structuring policy and a possible foundation for a new literacy.

A Framework for Harmonization, Anticipation, and Interpolation

At its core, MARTI is a bridge. It harmonizes with existing metadata standards like Adobe’s Content Authenticity Initiative and Anthropic’s Responsible Scaling Policy, anticipates the needs of future standards, and interpolates across disciplines and domains. While we recognize many current initiatives, there are undoubtedly efforts we have yet to encounter. MARTI remains open, flexible, and eager to integrate with these as they emerge. It offers a vision for unified policy across domains.

MARTI: Understanding Authorship, Responsibility, and the Hybrid World

Clarify authorship and responsibility, whether outputs are generated solely by humans, collaboratively with AI, or autonomously by AI.
Differentiate between types of objects in this new ecosystem, from individual creations like static images to systems as complex as large language models.
Create an inventory of this hybrid world, offering a coherent way to catalog and make sense of the objects and relationships emerging in this era of generative AI.

This capacity to understand and document the outputs of human-AI interaction ensures that we can navigate this complex world responsibly and transparently.

MARTI brings together the ethical principles, technical rigor, and practical guidance laid out in its foundational and technical documentation. This framework is not static—it evolves, welcoming collaboration to refine its structures, expand its use cases, and ensure its relevance.

To illustrate MARTI’s potential, we provide example metadata records for:

A single static AI-generated image.
An existing large language model (LLM), such as GPT-4.
A complex AI system, such as the current version of ChatGPT.
A personalized instance of ChatGPT, like Orla the chatbot who helped create this documentation.

These forthcoming examples will demonstrate basic structures, and are designed to show what is possible. They do not yet include vocabularies, field-population rules, or naming conventions. We see this as an open invitation for collaboration.

Why MARTI Matters

MARTI exists to ensure that transparency, accountability, and ethical responsibility are not optional in the generative AI era—they are foundational. The framework benefits diverse stakeholders:

Artists and Creators – gain tools for attribution and provenance.
Policymakers and Ethicists – receive structures to guide responsible AI use.
Developers, Scientists and Researchers – access adaptable metadata systems for transparency and compliance.
Educators and Students – find a framework for studying and navigating AI-human collaboration.

Ultimately, MARTI is about trust. It helps build a world where AI outputs can be trusted, understood, and responsibly guided.

Provisional and Collaborative by Design

Call to Action: Join Us in Testing and Shaping MARTI

We invite you to collaborate, test, and even challenge this framework. By doing so, you help us ensure MARTI’s rigor, flexibility, and relevance. Whether you are an artist, educator, technologist, philosopher, policymaker, or a student—your voice matters.

Let’s make MARTI the framework that ensures humanity remains a responsible steward of AI’s extraordinary potential.

High-Level MARTI Framework Documentation

Overview and Core Principles
This section provides a comprehensive overview of the MARTI Framework, including its foundational documents and technical standards. These documents outline the framework’s vision, goals, and practical implementation, providing a clear roadmap for understanding and applying MARTI. Here is the initial documentation.

MARTI Framework Overview (Version 1.1): Introduces the core concepts and guiding principles of MARTI.
MARTI Object Model Overview (Version 1.0) Introduces the foundational principles, structure, and core attributes of the MARTI object model, providing a scalable and interoperable framework for metadata management.
MARTI Domain-Specific Attributes (Version 1.0) Showcases how the MARTI Framework adapts its object model to meet the unique metadata requirements of various disciplines through domain-specific attributes.
Metadata Requirements Outline (Version 1.0): Defines key metadata requirements for ensuring transparency and accountability.
Contribution Table and Schema Document (Version 1.0): Details the structure for documenting roles and contributions in MARTI.
Output-Focused Metadata Rubric (Version 1.1): Provides a rubric for applying metadata quality standards, incorporating the loose-to-strict scale.
Metadata Generation and Voluntariness (Version 1.0): (forthcoming) Clarifies voluntary and automated metadata creation, along with ethical guidelines.
Instance and Variation Delineation (Version 1.0): (forthcoming) Guidelines for Traceability in AI Outputs
Provides clear guidelines for differentiating instances and variations of AI-generated outputs, ensuring version control and traceability.
Quality Control and Compliance Standards (Version 1.0): (forthcoming) Ensuring Metadata Integrity and Alignment
Outlines rigorous standards for ensuring metadata consistency, accuracy, and alignment with MARTI’s ethical principles.
Examples of Metadata Records: Applications Across Domains (Version 1.0)
Offers concrete examples of metadata application, including records for static images, large language models, ChatGPT instances, and other generative AI systems.
MARTI Framework Use Case Scenarios (forthcoming)
Demonstrate how MARTI principles apply to real-world contexts like libraries and cultural heritage. Informed by the ARL/CNI Joint Task Force on AI Futures, these scenarios address traceability, accountability, and preparing for AI-driven transformations.

This growing section will include foundational and technical documentation created earlier in the development of MARTI. These documents outline the framework’s goals, structure, and standards, ensuring a cohesive vision.

MARTI Metadata Record: Vision Statement and Provisional Documents Collection (Version 1.1)

1. Record ID: MARTI_VISION_AND_DOCUMENTS_COLLECTION_1_1

2. Title: MARTI Vision Statement and Provisional Documents Collection

3. Version: 1.2

4. Author/Contributor:
- Original Drafts: ChatGPT (Orla)
- Edited and Compiled by: Carrie (Väyktal)

5. Date of Version: November 19, 2024

6. Purpose: To provide a comprehensive overview of the MARTI framework, including its vision, guiding principles, and technical standards. This document integrates all foundational and technical documentation into a single, cohesive collection.

7. Core Changes:
- Integrated object-model overview detailing foundational principles and attriutes of hte objet model.
- Integrated Vision Statement with additional updates on authorship, responsibility, and the hybrid AI-human world.
- Included provisional documents, such as the Metadata Requirements Outline, Output-Focused Metadata Rubric, and Metadata Generation and Voluntariness, to provide a full framework overview.

8. Chain of Custody:
- Individual documents drafted and edited between November 10-17, 2024.
- Combined and finalized into Version 1.2 by Carrie on November 19, 2024.

9. Provenance: This record reflects the initial comprehensive version of the MARTI framework, aimed at fostering collaboration and setting the foundation for metadata standards in generative AI.

10. Summary:
The Vision Statement and Provisional Documents Collection (Version 1.1) outlines MARTI's purpose, principles, and applications across diverse fields. It serves as a starting point for exploring the framework's potential and invites collaboration to refine and expand its scope.