Model Collaboration

Democratizing AI through open models, data accessibility, and collaborative development. Building a transparent, participatory ecosystem that enables diverse global contributions.

Beyond "Open-Washing"

The Problem of "Open-Washing"

Certain models described as "open" are distributed under licenses that impose restrictions on genuine use or modification, particularly when such use could result in competitive products or services.

True Openness Definition

True openness refers to enabling others to freely study, use, modify, and distribute AI models without conditional restrictions. Widely recognized open licenses—such as Apache 2.0 and MIT—are essential to maintaining this standard.

The Three Pillars of Open AI

Open Model

High Availability:

Over 1.8 million open models exist on Hugging Face

Trustworthiness:

Measurable with tools like MOF (Model Openness Framework)

Critical Gap:

Model Development Participation - R&D is currently restricted

Open Data

High-quality datasets are required to address global problems:

Global Book Dataset:

Digitize and share copyright-free academic texts

Elder Speech Dataset:

Speech data from elderly populations for accessibility

Real-World Robotics:

Multi-institutional robotics interaction data

Open Compute

Hardware Flexibility:

Deploy AI even on inexpensive systems

Global Accessibility:

Extend access to emerging economies

Challenge:

Model training is costly and compute is often inaccessible to under-resourced teams

Model Openness Framework (MOF)

AI models are categorized into three openness classes based on the availability and openness of their components. Each class enables different levels of transparency and control.

Class III - Open Model

• Model Architecture

• Final Model Parameters

• Technical Report

• Evaluation Results

• Model & Data Cards

✓ Unrestricted usage
✓ Create products/services
✓ Fine-tune and align

Class II - Open Tooling

All Class III Components +

• Training/Validation Code

• Inference Code

• Evaluation Data

• Supporting Libraries

✓ Understand training process
✓ Validate benchmark claims
✓ Inference optimizations

Class I - Open Science

All Class II & III +

• Complete Datasets

• Data Preprocessing Code

• Intermediate Parameters

• Full Research Paper

✓ End-to-end analysis
✓ Complete reproduction
✓ Full auditing capability

Use the Model Openness Tool (MOT) at isitopen.ai to determine your model's openness class.

Preserving Culture Through Language

The Challenge

Dialects represent important subcultures. If a language is not represented in AI models, its culture risks being lost in the digital world. This creates an urgent need for inclusive language preservation efforts.

Current Initiatives:

500+ African Languages:

Canadian professor training models to preserve African cultures

Chinese Dialects (BAAI):

600 participants providing speech data for different accents

Indian Languages:

AI Kosha and Sarvam AI tackling 24+ languages with different scripts

Global Collaboration Needed

Working with global organizations like the UN, ITU, and Linux Foundation could help create top-down mechanisms and best practices to encourage different countries to collect and share high-quality language data.

Proposed Solution

Establish international standards and collaborative frameworks for language data collection, ensuring cultural preservation while maintaining data quality and accessibility for AI model training.

Join the Open Model Revolution

Help democratize AI by contributing to open models, datasets, and collaborative development platforms.

Join Model Collaboration Check Model Openness