Bias in AI
AI learns from human data -- and humans have biases. Learn how bias gets into AI systems, see real examples, and understand the impact on communities like ours.
Here's something that might surprise you: AI can be racist. It can be sexist. It can discriminate against people with disabilities, people from certain zip codes, and people who speak certain languages.
Not because someone programmed it to be that way. But because AI learns from us -- and we're not always fair.
How Bias Gets Into AI
AI systems learn from massive datasets -- billions of web pages, books, images, and documents created by humans. If those datasets contain biases (and they do), the AI learns those biases too.
Think of it like this: if you only read news from one political perspective your entire life, your worldview would be shaped by that perspective. AI works the same way, except it's reading everything -- including the worst parts of the internet.
There are three main ways bias enters AI:
Training data bias. If the data used to train AI over-represents certain groups, the AI works better for those groups. Most AI training data is in English, from Western countries, and created by people with internet access. That leaves out a lot of the world.
Historical bias. AI trained on historical hiring data might learn that "good candidates" look like the people who were historically hired -- which often means white men, because discrimination shaped those historical patterns.
Measurement bias. Sometimes the way we measure things is already biased. If "success" in a job is measured by metrics that favor certain work styles, AI will optimize for those styles and penalize others.
Real Examples That Happened
These aren't hypothetical. These are real cases:
Amazon's hiring AI was trained on 10 years of resumes. Because the tech industry has historically been male-dominated, the AI learned to penalize resumes that included the word "women's" (like "women's chess club captain"). Amazon scrapped the tool.
Healthcare AI used by hospitals across the U.S. was found to systematically give Black patients lower risk scores than white patients with the same conditions. The AI used healthcare spending as a proxy for health needs -- but because Black patients historically had less access to care, they spent less, so the AI assumed they were healthier.
Facial recognition systems have been shown to misidentify Black and Latino faces at rates 10 to 100 times higher than white faces. In Detroit, a Black man was wrongfully arrested based on a faulty facial recognition match.
AI-generated images of "professional" people tend to show white faces. Ask an AI to generate an image of a "CEO" and you'll likely get a white man in a suit. Ask for a "nurse" and you'll likely get a white woman.
Impact on Our Communities
In San Antonio, these biases show up in ways that affect real people:
- Job screening AI might rank resumes from certain zip codes (like the West Side or South Side) lower, even if the candidates are equally qualified.
- Loan approval AI might use patterns in data that correlate with race or ethnicity, even when those factors aren't directly included.
- AI-powered translation often struggles with Spanglish, code-switching, or regional dialects, making tools less useful for bilingual community members.
- Predictive policing AI has been shown to over-police communities of color, creating a feedback loop of more surveillance and more arrests.
What You Can Do About It
You might think, "I'm just one person using AI tools. What can I do about bias?" Actually, a lot:
- Be aware. Now that you know bias exists in AI, you can watch for it. That's already a huge step.
- Question outputs. If AI gives you results that seem to favor or exclude certain groups, push back. Ask the AI to consider different perspectives.
- Diversify your sources. Don't rely on a single AI tool. Different tools have different biases.
- Advocate. When organizations in your community adopt AI tools, ask them: "Has this been tested for bias? Does it work fairly for everyone?"
- Report problems. Most AI tools have feedback mechanisms. If you see biased outputs, report them.
The goal isn't to avoid AI. It's to use it with your eyes open.
Try This
Open an AI image generator (like Bing Image Creator -- it's free). Ask it to create an image of "a successful business owner in San Antonio." What does the person look like? Now try "a teacher," "a doctor," "a construction worker." Do you notice any patterns in age, gender, or race?
Talk About It
Have you ever felt like a technology didn't work as well for you as it does for others? Maybe voice assistants not understanding your accent, or autocorrect changing words in Spanish. Share your experience. How does it connect to what we learned about AI bias?