Carrie Bickner

Have you been following Alex Reisner’s brilliant research on the creative works that are being stolen and feed to LLMs? The most recent. There’s No Longer Any Doubt That Hollywood Writing Is Powering AI. Another in his series at The Atlantic, These 183,000 Books Are Fueling the Biggest Fight in Publishing and Tech, demonstrates challenges of intellectual property protection in a world where LLM trainers will do anything to feed their children.

After reading Resiner’s articles published today, it became evident that MARTI could address this issue, at least theoretically (this is another call for help; I don’t want to do this work alone). The framework’s ability to document and enforce the relationships between training data and AI systems offers a clear solution to a complex problem.

This realization led me to an observation about MARTI itself. This is something I should have understood from the beginning: MARTI is object-oriented. This might seem self-evident, but it highlights a critical aspect of the framework’s design. MARTI inherently organizes the world into objects, each with defined attributes and relationships. This design approach is both intuitive and functional, particularly for managing metadata in a rapidly evolving technological landscape.

Object-Oriented Thinking in MARTI

The concept of object-oriented systems is rooted in modularity and reusability. MARTI applies this paradigm by treating every entity—whether it is an LLM, a dataset, a creator, or an output—as an object. Each object has defined attributes and can be linked to other objects through those attributes.

For example:

– A dataset is an object that includes attributes such as source, copyright holder, licensing agreements, and usage restrictions.

– An LLM is an object with attributes documenting its training data, version history, and use cases.

– A creator is an object whose works can be linked to the systems or outputs that depend on them.

This structure ensures that each component of the system is discrete yet interconnected, enabling traceability and accountability across the framework.

Applications of Object-Oriented Design

Object-oriented design provides practical advantages in the context of metadata management and intellectual property. Consider a scenario where an LLM is trained using television shows. Without a structured framework, the connections between the training material and the model remain opaque. This lack of transparency complicates efforts to enforce copyright or licensing agreements.

MARTI resolves this by treating each show as an object with attributes such as title, copyright holder, licensing terms, and thematic elements. The LLM itself is another object, linked to the shows it was trained on through provenance metadata. This creates an auditable relationship, enabling clearer attribution and compliance with intellectual property laws.

Now, imagine an author who wants to maintain control over whether her work is allowed to train an LLM. Using MARTI, her publications could include a rights statement specifying how her work may—or may not—be used for AI training.

That statement could include the license she wants to offer to those training an LLM. The license might:

– Specify the scope of work allowed for training (e.g., specific chapters or entire publications).

– Automatically limit the duration of the agreement.

– Include terms for royalties, specifying how and when payments should be made.

The process could be automated, invoking the license terms and executing the agreement without manual intervention. This seamless interaction is possible because MARTI treats all elements—authors, publications, training datasets, and contracts—as objects. It provides a modular, adaptable framework that can address domain-specific needs while maintaining consistency and interoperability.

Flexibility and Scalability

One of the key benefits of MARTI’s object-oriented design is its scalability. Each object operates as an independent unit, allowing the framework to adapt to new requirements without compromising its structure.

For example, as legal and regulatory standards evolve, MARTI can incorporate new metadata fields to address emerging needs:

– Jurisdictional Compliance: Attributes can specify whether a dataset complies with region-specific laws such as GDPR or CCPA.

– Licensing and Usage Terms: Smart contract integration could automate royalties or licensing fees for intellectual property holders.

– Risk Assessment: Attributes can flag potential compliance risks or ethical concerns, enabling proactive mitigation.

This modularity ensures that MARTI remains relevant and effective in managing metadata as the technological and regulatory landscape changes.

Reflection on Design

MARTI’s object-oriented nature reflects a deliberate approach to organizing complex systems. This methodology is rooted in my own experience with metadata and frameworks. From teaching cataloging at Pratt Institute to advocating for web standards with WaSP, I have consistently approached systems as collections of discrete yet interrelated objects.

This perspective shaped MARTI’s design. Each element of the framework—datasets, models, creators—is treated as an object with defined attributes and connections. This approach ensures that MARTI is not only functional but also intuitive for those who implement it.

Future Implications

The object-oriented nature of MARTI positions it as a flexible and future-proof framework. By maintaining a modular structure, MARTI can evolve alongside advancements in AI and changes in intellectual property law.

For instance, as generative AI tools become more prevalent, the need for transparent and enforceable attribution systems will grow. MARTI’s capacity to document and link training materials, creators, and outputs provides a solution that aligns with both ethical principles and practical needs.

This flexibility also makes MARTI well-suited for addressing regulatory compliance, data privacy, and other emerging challenges. By treating each entity as an object with defined attributes, the framework ensures that metadata can be adapted to meet new requirements without disrupting the overall system.

From Objecs to Outcomes

MARTI’s object-oriented design is foundational to its utility as a metadata framework. By treating datasets, creators, and systems as discrete yet interconnected objects, MARTI provides a robust method for managing relationships and responsibilities.

This approach is not only effective for current challenges but also scalable for future needs. As technology and regulation evolve, MARTI’s modularity ensures that it can adapt to new contexts while maintaining its core principles of accountability, transparency, and integrity.