We are drowning in data, yet starving for context. Scale matters very little if we lack the context to make it useful. 

On June 4, 2026, I had the opportunity to represent Time Study at HearstLab’s inaugural AI Showcase, where my presentation focused on making the case for local language models. Given the tech industry’s obsession with scale, it’s not sexy to promote anything small. While there are great (and exciting) use cases for large language models, I believe that the future depends on culturally context-sensitive models.

Context Matters

In our work at Time Study, we classify work activities. In using unstructured data to determine these activities, we’ve learned that you often need deeper context to know how to accurately classify an activity. It is rarely just about keywords.

If accuracy matters, you may need more information about the person, their environment, their team, and their role. For example, if you want a machine to determine whether a physician is performing research or a clinical activity, in the absence of labels, you may need to understand the intention for performing that activity or the location where work occurred.

Also, language is local. Organizations have their own language, much like any other culture. Internal acronyms and shorthand can mean one thing in specific settings or roles, and something entirely different elsewhere. This friction between raw data and true meaning isn’t unique to enterprise software, it is a fundamental challenge across culture, history, and art.

Art and Context

This spring, I attended a talk at the Virginia Museum of Fine Arts titled “Yorùbá Art Objects in American Museums: Revisiting the Spiritual Presence and Absence Conundrum” with Dr. Adewale Adele from Georgia State University.

He highlighted that in the traditional Yorùbá worldview, objects like Egúngún masks, Òṣùgbó society figures, and Ifá divination objects are inseparable from ritual. When taken out of context and placed in the static, glass-case environments of museums their ritual context may be stripped away, leaving their true spiritual essence fundamentally “absent.”

Several points from Dr. Adele’s lecture clarified how much context shapes interpretation:

• Environmental Context: Some museums place these artifacts in areas of the building that are counter to the regions they originated. Q: How does the interpretation of data change when it is removed from the environment that gave it meaning?
• Intentional Context: In displaying these items, institutions often place them out of context with the artifact’s original intention. For example, some tribal masks and artifacts are meant to be disposed of after a ceremony rather than preserved indefinitely. Q: How do you display, convey, or preserve something that may only have meaning within a specific environment or moment in time?

While an object can be technically protected, displayed, and cataloged as obtained, visitors can be unintentionally misinformed without community context. They may misunderstand the purpose of the object, the environment it should be displayed in, how it should be cared for, or whether it was meant to be preserved, displayed, or seen outside of its original use.

To address this disconnect, art historians like Dr. Adewale Adele have proposed restorative frameworks such as the Indigenous Context Engagement Paradigm (ICEP), which centers the memories, voices, and living traditions of the people closest to the objects’ original meanings. This can also include more immersive practices, such as videos and stories, that can guide interpretation.

The Impact of Improper Context 

Just as a museum can technically protect, display, and catalog an artifact while losing its original context, a large language model can store, index, and process language, and still completely misinterpret it when stripped of its cultural, organizational, historical, or environmental context.

In an ocean of data, more information does not automatically lead to better understanding. A model’s accuracy and reasoning can degrade when useful context gets crowded out by excessive or irrelevant information. The issue is not simply whether a model has access to more data. The issue is whether it has access to the right context.

This might not matter when you’re using selfies to create AI-generated cartoon versions of yourself. But it matters a lot in settings where the stakes are higher and errors can lead to misinformation or cause actual harm.

Localized Intelligence

This is why smaller, community-led, fine-tuned models matter, designed not to capture the entire internet, but optimized specifically for the distinct vocabulary and environment they are meant to serve. Models that are trained to not only determine, “What is this data?” but also determine, “What gave this data meaning?”

We don’t need AI to know everything about the world to be profoundly useful. Sometimes, we just need it to more deeply understand the localized, highly specific context that makes interpretation possible.