Our language is not a barrier. It is infrastructure.
For generations, speaking Spanish was treated as a problem to be corrected. It was always the opposite: a skill that calms a frightened patient, closes a misdiagnosis, and carries a culture forward. As machines are built to stand between people and their doctors, courts, and classrooms, one question decides whether that skill survives the next decade, or is quietly written out of it.
Artificial intelligence will learn to speak for us, or it will learn from us. Only one of those keeps our language and our culture alive.
When the language matches, the outcome changes. This is measured, not felt.
Language concordance, a patient and a professional who share a language and the culture inside it, is one of the most consistent and underused levers in American care. The research is not ambiguous.
Percentage-point drop in A1C for Latino diabetes patients matched to a Spanish-speaking Latino primary-care physician. A clinically meaningful change, equal to adding a second medication.
Fewer emergency-department visits for limited-English-proficient patients who received language-concordant primary care, with 12% lower 30-day readmissions, versus care through an interpreter.
Patients are roughly twice as likely to understand their diagnosis and follow a treatment plan when it is explained in their own language by someone who understands the culture around the kitchen table, not only the clinic.
The same technology can erase a culture or carry it.
AI is now being trained to take histories, translate consent forms, draft legal filings, and tutor children. The models learn from whoever helps build them. Right now, we are rarely in that room.
Erasure by default.
A model trained mostly on English, on textbook Spanish, on one dialect, treats everything it has not seen as noise to be cleaned up.
- Regional words, indigenous-rooted terms, and family idioms get “corrected” into something no abuela would recognize.
- Cultural context, who speaks for the patient, what fear or faith they bring, gets flattened out of the record.
- The bilingual worker who has always interpreted is replaced by a tool that does it worse, and no longer paid for the skill.
- A generation grows up hearing a machine speak a thinner version of their own language back to them.
Culturally intelligent by design.
Technology trained with the community instead of on it can do the opposite: hold a language steady and hand it forward.
- Dialects, heritage vocabulary, and cultural context are treated as ground truth, not error.
- The people who carry the language help build, label, and check the tools, and are credited and paid for it.
- AI extends a bilingual professional’s reach instead of replacing the human judgment around the meaning.
- Every interaction becomes a record that strengthens the language for the next child who speaks it.
Millions already hold the skill. Almost no one measures it.
The country treats bilingualism as a private quirk instead of the professional competency it is. The asset is already here, in clinics and break rooms and front seats of cars on the way to appointments. It has simply never been counted, certified, or paid.
Four commitments, so the language survives the machines.
Treat language as a credential, not a favor.
We work to measure and certify the bilingual skill people already carry, so the aide who interprets, the nurse who explains, and the heritage speaker who never “studied” Spanish can be recognized and paid for a professional competency, not asked to provide it for free.
Put our people on the teams that build AI.
Tools that mediate medicine, law, and learning are being trained right now. We push for Latino clinicians, linguists, and community members to help design, label, and audit them, because a model only respects what its makers respected.
Hold technology to a cultural standard.
We name a simple test for any AI that touches our communities: does it preserve dialect and context, or correct them away? Does it extend a human relationship, or quietly sever it? Tools that fail the test do not deserve our trust, and we will say so publicly.
Turn every interaction into preservation.
Used well, the same systems that could flatten our language can instead archive and strengthen it, capturing how it is really spoken across regions and generations, so the next child inherits a fuller language, not a thinner one.
A people that keeps its language keeps the key that unlocks its chains.Frantz Fanon · on language and freedom
The machines are being taught this decade. Let us teach them well.
Our language and our culture are worth measuring, certifying, defending, and building into the tools that will shape the next generation. That is not nostalgia. It is infrastructure, and it is ours to protect.