AI Still Doesn't Understand How We Talk. That's a Problem.

Artificial intelligence is becoming part of everyday life. It answers our questions, transcribes our conversations, supports customer service, and increasingly influences decisions about people.

But there's a question we aren't asking enough:

What happens when the technology shaping our future doesn't understand the way millions of us communicate?

For many Black Americans, this isn't just a hypothetical question.

African American Vernacular English (AAVE) is a rich, systematic variety of English with its own grammatical structures, linguistic patterns, cultural history, and regional differences.

It is not broken English. It is not simply slang. And it is not something that needs to be corrected before technology can understand it.

Yet research has repeatedly identified disparities in how language technologies process African American speech and writing.

The Research Has Been Telling Us Something

In 2020, researchers examined five commercial speech-recognition systems and found that their average word error rate was 35% for Black speakers compared with 19% for white speakers in the study. [1]

Put simply, the systems made substantially more transcription errors when processing speech from Black participants.

These figures describe the systems and participants studied in 2020. They are not a benchmark of today's speech-recognition products.

Four years later, research published in Nature identified another troubling issue. The language models tested demonstrated negative racial stereotypes when evaluating African American English, even without being explicitly told the writer's race. They also made discriminatory judgments in hypothetical employment and criminal justice scenarios. These experiments do not establish how every current model behaves or measure outcomes in deployed hiring or justice systems. [2]

These are different failures, but they raise a common concern.

If an AI system misunderstands how someone speaks, or makes unfair assumptions based on how someone writes, what happens when that system is used in a job application, customer service interaction, classroom, or another consequential setting?

The Problem Isn't Our Language

We have to challenge the assumption that communities should change how they communicate to accommodate technology.

AAVE is not a technological deficiency. It is a legitimate linguistic system that deserves accurate representation in the datasets, evaluation processes, and products shaping modern AI.

The responsibility should not rest entirely on the speaker.

Developers must examine whose language is represented, who gets to define accuracy, how cultural meaning is preserved, and whether the systems perform equitably across different populations.

And that requires more than collecting additional recordings or adding slang terms to a database.

It requires cultural understanding, linguistic expertise, thoughtful annotation, rigorous testing, and continuous improvement.

But There's Another Conversation We Need to Have

Who is doing the work to improve these systems?

Who is labeling the data, explaining the cultural context, identifying mistakes, and helping developers understand what their models missed?

And perhaps most importantly, who benefits financially when that work makes AI products more valuable?

Our communities have contributed enormously to American language, music, entertainment, and culture.

As artificial intelligence expands, we have an opportunity to rethink how that cultural knowledge is valued.

We should not be satisfied with simply being better represented in the technology.

We should be discussing ownership, compensation, access, and long-term economic opportunity.

What would it look like for skilled community contributors to be paid fairly for their expertise? For Black-owned companies to manage language annotation projects? For thoughtfully governed datasets to generate commercial opportunities without surrendering control of contributors' recordings or identities?

These aren't questions technology companies should answer on our behalf.

Building Something Better

At The TableAI, we believe improving AI should involve the people whose language and experiences are being represented.

Our vision connects three priorities: more accurate language technology, responsible stewardship of cultural knowledge, and meaningful economic participation.

That means approaching AAVE annotation as skilled work, rather than simply collecting language samples and assuming the work is complete.

It means building a framework where contributors are compensated, quality is measurable, and questions about data ownership and future use are addressed from the beginning.

No company owns AAVE as a language. Any rights in recordings, datasets, tools, or annotation methods must be addressed through clear contributor agreements and consent. The economic models discussed in this series are possibilities we aim to explore, not benefits we can promise today.

And it means recognizing that the future of AI shouldn't be something our communities merely adapt to.

We should have a meaningful role in building it—and sharing in the value it creates.

This is the first article in Our Language. Our Data. Our Future., a series exploring the failures, research, economics, and opportunities surrounding AAVE and artificial intelligence.

Because the future of technology should understand us.

And we deserve a seat at the table.