More Than a Place in the Data

A harmful assumption becomes a measurement. The measurement becomes a classification. The classification helps determine what happens to a person.

That is one way prejudice gains authority. Put it inside a system that looks scientific, and people may stop questioning the judgment underneath it.

For people of color who want a say in how AI affects our lives, learning the technology is a practical place to begin. We need to understand it well enough to recognize when an old assumption shows up in a new answer, document the problem, and help change what the system learns.

That means becoming proficient users, participating in training and testing, and bringing our knowledge into the rooms where decisions are made. The work is already underway. There is a place for more of us in it.

The better we understand AI, the better prepared we are to recognize what it gets wrong about us and help change it.

Old Assumptions Can Enter New Systems

Eugenics was built on the false belief that society could be improved by controlling who reproduced. It ranked people through claims about heredity and used those claims to justify exclusion and coercion.

Its history includes forced sterilization and other abuses against marginalized communities. The National Human Genome Research Institute explains that history in Eugenics and Scientific Racism.

People were judged through categories they had little power to challenge. Institutions treated those categories as knowledge, and the people placed inside them lived with the consequences.

AI creates another place where we need to watch that process carefully. A biased model is not automatically a eugenics program. But systems that classify people, rank their potential, or influence their access to resources deserve scrutiny, especially when they learn from records produced by unequal institutions.

A newer tool can carry an older worldview. We need people who know enough about both to recognize the connection.

Bias Can Look Like a Reasonable Answer

We often imagine bias arriving as something obvious: a slur, an insult, or a statement anyone can recognize as racist.

With AI, it can arrive as a polished answer that sounds professional.

In Hofmann et al. (Nature, 2024), “AI generates covertly racist decisions about people based on their dialect,” researchers found that language models made prejudiced judgments about people based on African American English. In experimental scenarios, that prejudice affected recommendations about employment and judgments in fictional criminal cases.

The models did not need to be told a person's race. The way someone spoke was enough to activate the prejudice.

That should matter to anyone who has been told to change how they speak to be taken seriously. A system can recognize our words and still misunderstand our intelligence, professionalism, or character.

This is where cultural knowledge matters. Someone familiar with a community may notice an assumption that another person accepts as ordinary. AI proficiency helps turn that recognition into a clear test, a documented problem, and a case for changing the system.

What Gets Measured Shapes Who Gets Help

The report I reviewed also describes a healthcare algorithm that used future medical spending to predict who needed additional care.

That choice sounds practical. But spending and need are not the same thing.

When Black patients have received less care, a model can learn to predict lower spending even when they are seriously ill. In Obermeyer et al. (Science, 2019), “Dissecting racial bias in an algorithm used to manage the health of populations,” Black patients were sicker than White patients assigned the same risk score. Correcting the disparity would have increased the Black share of patients selected for additional help from 17.7% to 46.5%.

That was an estimate of what a fairer selection process would produce. The lesson is still direct: choosing the wrong measure can make people who need help harder to see.

This part became personal for me. I was recently hospitalized for malignant hypertension. The doctors could not identify a root cause, but they kept referring to the fact that I am Black and resistant to treatment. I was left questioning how much explanation my race was being asked to carry.

I do not know whether AI influenced my care. My experience reminded me how much it matters to question an assumption about a group when we are trying to understand an individual. My Blackness is not a complete explanation for what is happening inside my body.

That same habit of questioning belongs in AI development. Who chose the measure? What does it leave out? Whose experience would reveal the problem?

We Need to Help Train and Test the Systems

Being involved in AI training means more than adding diverse faces to a company photograph.

People make decisions about which information a model learns from, how examples are labeled, which answers count as helpful, and what problems must be corrected before a system is used.

People of color should be part of those decisions.

That can include reviewing training material, helping label examples accurately, comparing model responses, designing tests, contributing expertise, and studying how a system performs in a real community.

A teacher may notice that a model confuses dialect with poor understanding. A healthcare professional may question a measure that hides unmet need. An artist may recognize when a model reduces a culture to a handful of stereotypes.

Those observations become more useful when the person making them understands how to test the tool and explain what needs to change.

We also need clarity about what participation actually means. Using a chatbot does not necessarily mean our feedback will be used to train it. Meaningful involvement requires a defined role, a way to report problems, and evidence that the people responsible will act.

Our lived experience is valuable expertise. It deserves compensation, respect, and influence.

Proficiency Means Learning to Question the Output

You do not have to become a software engineer to begin. You do need to move beyond accepting an answer because it sounds confident.

Here is a practical place to start:

• Learn a tool through work you understand. Try it on a lesson, a business task, or a creative problem where you can judge the quality of the answer. Notice both its usefulness and its mistakes.
• Look for the assumptions. Ask what the answer assumes about language, ability, family, neighborhood, or access to resources. Look at who is represented and who is missing.
• Compare responses carefully. Try equivalent examples while changing one relevant detail, such as a name or dialect. Keep the underlying meaning consistent and compare patterns across several attempts.
• Document what you find. Save the prompt, response, tool, date, and why the result concerns you. One troubling answer can raise a question; a careful set of examples makes the problem easier to investigate.
• Seek opportunities to shape the work. Look for evaluation roles, training-data work, community research, advisory groups, or further study. Ask whether your input can change a decision and whether the role is paid.

You do not have to find your way into this work alone. There are organizations already building it:

• Black in AI works to broaden Black participation and influence in AI development, deployment, and policy. Explore its community and programs if you want to connect with people working in the field.
• The Algorithmic Justice League combines research, art, education, and advocacy to expose AI harms. Its resources and harm-reporting pathways offer a place to learn and contribute to accountability.
• Data for Black Lives connects data work with organizing and change in Black communities. Explore its research, events, and movement work if you want to connect technical learning with community needs.

These are different routes into an existing movement. You can begin by reading their work, attending an event, and identifying where your skills or learning goals fit.

These are skills we can practice and teach. They help us explain a concern clearly enough that it can be tested, challenged, and addressed.

They also help us recognize useful technology. Learning AI gives us more choices about where to use it, where to question it, and where to refuse it.

Participation Needs Real Power

People of color are not one community with one experience. Black, Latino, Indigenous, Asian, and other communities bring different histories, languages, and needs. Representation has to make room for that range.

And the responsibility for fixing bias cannot be placed on the people most likely to experience it.

The companies building these systems and the institutions deploying them are responsible for their decisions. They should provide accessible training, paid opportunities, independent testing, and a clear response when harm is found.

Inviting us to review an answer means little if we cannot influence what happens next. We need a say in what gets measured, which failures are unacceptable, and whether a system should be used at all.

We Should Help Shape What Comes Next

Whether we are working with today’s tools or preparing for AGI, the idea of AI that could handle a broad range of intellectual tasks, our task is the same: build the knowledge and influence to question what a system learns before we give it more authority over people.

I believe AI can expand access, creativity, and opportunity. I also believe its benefits depend on the choices people make around it.

That is why I want more people of color learning these tools, becoming proficient, helping train them, and testing them with the care our communities deserve.

Our knowledge of how bias feels is a starting point. Our ability to identify it, demonstrate it, and help change a system is power we should keep building.

We deserve more than a place in the data. We deserve a hand in shaping what the system learns and what it is allowed to decide.

Ty Boyland is a Creative Industries AI consultant, youth engagement specialist, and music industry leader based in Memphis. He helps organizations move from AI curiosity to practical, responsible use through consulting, training, and facilitation.

Read more and work with Ty at Ty Boyland Consulting.

Read and join the conversation on Medium.