Community-Centered Language Models

In the Studio – project

The Memory & Language Models project examines how AI & language models might help people preserve and protect what they know, make connections across their collective memory, and create new ways to access knowledge that might otherwise be fragmented, overlooked, or lost. This is not simply a smaller version of a large language model. It is an exploration of what becomes possible when a model is created for a specific context, shaped by the people closest to that context, and designed around a clear purpose.

What We’re Exploring

Our goal is to create tools that help people and communities work with the knowledge they already hold.

That knowledge may live in stories and memories, local language and cultural references, personal archives and oral histories, organizational records, or the shared questions and experiences that connect a group of people.

Rather than building systems intended to know a little about everything, we are exploring models designed to understand a particular body of knowledge more deeply. These tools could help communities preserve and project what matters, uncover connections across people and generations, interpret information within its original context, and create new ways to learn from what they collectively know.

The goal is not to reduce lived experience to data or allow technology to speak on behalf of a community. It is to explore how language models can support remembering, connecting, interpreting, and creating while the people closest to the knowledge retain meaningful agency over how it is represented, accessed, and used.

Our Approach

Each project begins with people, not technology.

We begin by asking who the model is intended to serve and what it should help people understand, remember, explore, or create.

Participants should understand what they are contributing, how it may be used, and what should remain private, protected, or outside the project.

Stories, language, and records are not treated as interchangeable pieces of information. We consider their source, meaning, relationships, history, and cultural context.

Each model is designed around a defined purpose rather than unlimited use. Technical choices may include smaller models, local processing, private knowledge systems, or other approaches appropriate to the people, information, and goals involved.

Participants are not merely sources of data. Their perspectives can help us interpret what emerges, identify what is missing, challenge what is inaccurate, and shape how the tools and the project evolve.

more info