The Rise of the Humanist: The Inheritance

While LLMs train on what can be scraped from the internet, millions of primary sources remain in darkness—archives holding cultural heritage yet inaccessible for training the very systems shaping our future. As this gap persists, AI's version of human culture will continue to drift further from reality. The solution isn't just technical; it requires the very skills humanists have cultivated for centuries: discernment, cultural context, the ability to distinguish authentic voices from colonial ones, to detect bias, and much more.

More than 12 years ago, when my paternal grandmother, Olga Iglesias, passed away, I became the unwitting steward of a primary source historical archive. I just thought I was inheriting family photos, but instead it soon dawned on me that what I truly inherited was a research archive, that of my beloved grandmother. I had opened a forgotten vault, revealing a hidden treasure trove of approximately 3,000 cultural heritage items currently being inventoried documenting mid-20th century Puerto Rican classical music history—all connected to the legacy of Pablo Casals, a 20th-century humanitarian, peace crusader, and master musician known for transforming how cello was played. Through Olga Iglesias, my grandmother, he found his soprano of choice for his own compositions as documented in the archive. Growing up I had known about her connection to Casals and her incredible career through family lore and old photographs. Yet my grandmother was just abuela—or as my brother and I would call her, Ía (EE-ah)—not this artist figure, the soprano Olga Iglesias.

It was staggering to realize the depth of the archive: programs, personal and professional correspondence, photography, contracts, newspapers, magazines, recordings, video—a true hidden archive that chronicles Puerto Rico's development through its classical music history at a time when the island was defining itself. Puerto Rico was emerging as a commonwealth under its first freely elected governor, Luis Muñoz Marín, who understood that promoting the arts and courting cultural prestige was essential to having Puerto Rico be seen as a world player, through a form of soft power. The Muñoz Marín administration with Casals, who lived in Puerto Rico from 1956 til his death in 1973, built cultural institutions such as the Puerto Rico Symphony Orchestra, the Conservatory of Music, and the Casals Festival; institutions that rooted the island's cultural identity and still stand today. This cultural investment was part of Muñoz Marín's larger vision, Operación Serenidad, a purposeful counterbalance to the rapid industrialization Puerto Rico was experiencing. Therefore, the archive reveals how Puerto Rican artists like my grandmother weren't simply performers but insiders in these institutions and in these circles, cultural ambassadors in a deliberate project of nation-building through the arts.

This archive has been supported by the National Endowment for the Humanities and the Puerto Rican Foundation for the Humanities. As its steward, I've spent 12 years intimately familiarizing myself with every object in the archive - photographs, letters, postcards, programs, newspapers, audio recordings and more - digitizing most items while researching its context. Through this hands-on practice, I've learned preservation, cataloging techniques, and archival standards. I feel an immense privilege to shepherd its beautiful contents into the world, bringing to light what was once hidden, for the enlightenment of curious minds. It's meticulous work requiring care and precision, yet it transformed how I understood my grandmother. For example, while I knew intellectually that Casals attracted international attention to Puerto Rico, I never grasped how centrally my grandmother participated in these cultural exchanges. Her archive revealed correspondence with representatives from cultural organizations from Spain to Mexico, from Greece to Germany, from Israel to France, among many others. She wasn't just performing with Casals, she was an active node in his international network. 

And hers is just one archive among thousands, ranging from niche family archives like my own to documents of world-historical importance in GLAM (galleries, libraries, archives, museums) institutions. My grandmother's archive, historically significant enough to be recognized by major humanities institutions, remained basically invisible to researchers and communities who could benefit from it. What began as my family discovery revealed itself as a microcosm of a systemic issue, this pattern repeating time and again: world-historical documents and hidden archives, small and large alike, sitting in the dark, their discoveries waiting.

Collectively, much cultural heritage sits as ‘dark data’—a Library of Alexandria that never burned but remains unread, dispersed across institutions worldwide in both analog and digital formats. Here's the problem: as archival scholars have long noted, digitization alone changes little. That letter inviting my grandmother to perform with Casals for the first time? It’s trapped in a TIFF file, as invisible as if still in its envelope. The photo of her with Queen Elisabeth of Belgium? Just another unlabeled JPG. Even digitally-born PDFs remain equally hidden. Without cataloging—extracting entities, applying controlled vocabularies, structuring relationships—digitized materials yield little. And GLAM institutions face massive backlogs of both analog and digital materials. The challenge isn't just volume; it's also the specialized work required: applying institution-specific standards across multiple languages and cultural contexts.

I discovered the power and jaggedness of LLMs when we lost our cataloging grant mid-project for my grandmother's archive. The Puerto Rican Foundation for the Humanities had been funding a consultant to help catalog the archive for the Digital Library of the Caribbean, at the University of Florida and Florida International University. My heart sank. How was I going to juggle cataloging work with running a nonprofit at the same time? This was why I applied for the grant to begin with—to get help!

This is where necessity became the mother of invention. With my background in software product management and database systems, I'd always fantasized about structuring every item of her archive into a SQL database—a clean relational dataset of each performance, each program, each letter, each photo, each person, etc. All the metadata, perfectly structured. Before AI, this would have taken years of tedious manual entry with little ROI, but recent breakthroughs in AI and data processing made me ask myself, why not build an application to help me instead?

My application started as a simple LLM workflow to extract entities from typewritten correspondence through named entity recognition (NER)— classifying people, organizations, dates, locations, etc. This alone was transformative: I could review the LLM's output for accuracy instead of manually extracting entities and metadata, speeding up processing time significantly. One feature led to another, such as matching entities with their controlled vocabularies, and it quickly hit home how cheap and easy it was to build software nowadays, shipping production code faster than ever before. It truly is a brave new world! 

However, as elegant and modern as that sounds, this is what we must guard against: LLMs’ nondeterministic nature. An LLM that brilliantly extracts metadata from a letter about my grandmother's first performance with Casals might, in the next query, confidently misidentify her as a mezzo-soprano from elsewhere. Wrong country, wrong voice type!

Another example - imagine you are part of a marginalized community where much of your history isn't well documented, let alone used by AI companies to train their models. What documentation exists is often rife with bias and colonialism. As a member of your community, you prompt an LLM and receive information that may be wrong, biased, or incomplete presented with confidence. This same flawed information is being provided to everyone else asking the same or similar questions. Even when accurate, the LLM can't distinguish between authentic community knowledge and external interpretations, making it difficult to trace the origin of the information or credit its true creators. This compounds erasure with misinformation and even appropriation. 

Furthermore, LLMs are deployed globally despite being trained primarily on Western sources that reflect existing biases rather than on diverse primary sources or community-authored materials. Hence, these models too often prioritize technological capabilities over cultural sensitivity. When viewed through the lens of local cultures, AI tools should amplify community voices and preserve cultural knowledge—not homogenize them. For instance, LLMs could identify and elevate community-authored materials over external interpretations, ensuring that when someone searches for the cultural history of Puerto Rico through classical music, they find authentic Puerto Rican voices, historical and contemporary, rather than outdated 20th century colonial perspectives about Puerto Rico. These models risk converting outdated and biased history into AI-generated content that appears authoritative but draws from internet secondary or tertiary sources like Reddit and Wikipedia rather than verified community-authored or primary source documentation. There is much to be desired.

All that said, instead of being purely skeptical and wary of AI in cultural heritage, we must think of new ways to harness AI - to promote archives, empower those who catalog and use them, and serve the communities from which the documentation originates. Humanists are essential to this vision. They serve as sentries who discern AI's output, ensuring we promote rather than demote human and collective knowledge through incomplete, biased, or inaccurate information. In the AI Age, discernment becomes an ascendant skill, a skill central in the humanities and liberal arts. Interestingly, in an era labeled "post-truth," we've had the answer all along: armies of humanists ready to navigate our brave, new world. They abound in society, though often unrecognized: teachers, archivists, librarians, family genealogists, curators, etc. Instead of drowning in the floods of an AI tidal wave, we must learn to work with its currents, embracing not just discernment but the full humanist toolkit: reason, ethics rooted in human needs, empathy, critical thinking, and intercultural competence to just name a few.

As a Fellow at the Virginia Tech Institute of Leadership in Technology, I focus on this new frontier: the intersection of AI technology with cultural heritage - for all its glorious bits and also for all its troubling dimensions. I'm exploring how to reposition GLAM archives as actual profit centers through AI, while highlighting the increasingly crucial work humanists already do as our sentries ensuring authenticity and accuracy in the AI Age. It is my hope to live in a day in which communities can empower themselves with their own history, through access to their own archives. I want to live in a day in which GLAM institutions are making discovery after discovery pushing their respective research forward. Lastly, I want to live in a day in which my grandmother's voice, and many others, is no longer trapped in digital suspended animation, but alive in the communities that need to hear their song. The technology now exists, the archives wait, and the humanists stand ready. What we need now is the wisdom to bring them together ethically and responsibly.