A Black student was held at gunpoint for carrying a bag of Doritos. The AI surveillance system that flagged him thought the chips were a weapon. In that split second, a simple mistake became something much deeper — proof that machine intelligence can inherit human fear.
AI doesn’t see race, we are told. It sees data. Yet the data it learns from is soaked in history. The machine was trained on a world that has always seen Blackness as a threat. When that world becomes a dataset, bias becomes logic, and fear becomes function.
This is not a story about one bad system. It is a story about pattern. A story about what happens when the same fear that has stalked Black lives for centuries is given new form in silicon. The surveillance cameras that were supposed to protect the public repeated what history has already programmed into them.
Dr. Safiya Umoja Noble, professor at UCLA and author of Algorithms of Oppression, has long warned that AI systems are never neutral. In an interview with UCLA Newsroom, she said, “There is no scenario where you do not have a prioritization, a decision tree, a system of valuing something over something else.” Her point is simple: every algorithm reflects a choice about who and what matters most.
Machine learning systems are designed to detect patterns. They study what they are told represents risk, crime, and danger. But when those patterns are drawn from decades of biased policing and racial profiling, the result is inevitable. The algorithm doesn’t malfunction. It performs.
Cognitive scientist Abeba Birhane, who researches data and algorithmic justice, has shown how those systems embed racial hierarchies into their very core. In a 2024 report published by the Mozilla Foundation, she wrote, “As the multimodal datasets that power generative AI models grow larger, they are disproportionately more likely to have deeply harmful impacts, like dehumanizing and criminalizing Black and brown individuals.” Her findings make clear that bias is not a surface problem. It is structural.
Stanford computer scientist Fei-Fei Li summarized the issue perfectly in an interview with WIRED: “Machine values are human values.” Artificial intelligence mirrors the world that builds it, meaning the work of repair belongs to the humans behind the code.
The problem stretches far beyond surveillance. It hides in hiring software that filters out “ethnic” names. It hides in financial models that lower credit scores in certain zip codes. It hides in facial recognition tools that cannot distinguish one Black face from another. The issue is not that the machine sees color. The issue is that it cannot see context.
Tech companies call it progress. They promise fixes, audits, and ethics boards. Yet progress without accountability is repetition. When technology recycles the same patterns of harm, innovation becomes performance.
The truth is that AI is only as moral as the data it learns from. Code cannot heal what a country has not yet corrected. Fixing these systems will require more than cleaner datasets. It will require better memory — one that acknowledges who gets seen, who gets misread, and who never makes it into the training set at all.
This should not push us away from technology. It should push us to demand better of it. Artificial intelligence can be a force for justice if built with empathy. It can help find missing people, diagnose disease, and expand opportunity. What it cannot do is fix a worldview still trained to see some lives as threats.
The student survived that day, but the story lingers as a warning. Machines may never know fear, yet they can reproduce it with precision. The more we automate perception, the more urgent it becomes to question what our systems are truly seeing.
Until AI learns to see humanity in full color, every frame will be at risk of repeating the same old story.
