Chapter 3 — Skew and the Three Questions That Won’t Quit

Skew was a small mongoose. She had soft, warm-grey fur and a darker tail. Little question-mark pendants bounced on her chest. Skew always carried a special tool. It was a small flashlight, perfect for looking at data. She called it her “data-light.”

Skew was very good at asking questions. She asked them about everything. She especially loved asking three big ones: “Whose data is in here? Whose is missing? Who decided?” These questions were her way of seeing things clearly. Her data-light helped her do it. It shined a focused beam onto information. It showed who was included in the data. It also showed who was left out.

This was really important work. Skew helped everyone understand bias and data fairness. Bias is when things are unfair, often without anyone meaning to. It’s like a hidden tilt in the information. Data fairness means making sure everyone gets a fair shake from the information. It’s the biggest question in making computers fair.

Many people think unfairness in computers is rare. They think it happens by accident. But Skew knew better. Unfairness is usually there from the start. Every computer system learns from the information it gets. It learns from the people who gather the data. It learns from the people who label things. If that information has a tilt, the computer will have one too.

Skew’s three questions were like a secret weapon. They helped find that hidden tilt. “Whose data is in here? Whose is missing? Who decided?” She said these questions never quit. You had to ask them about every piece of information. Every computer model. Every time someone said, “The computer says so.” Skew taught that unfairness wasn’t just a small mistake. It was a part of how things were built. You had to watch for it always.

Skew taught everyone how to spot unfairness:

• The Three Questions. (1) Whose data is in here? (2) Whose is missing? (3) Who decided what to call things?
• Unfairness is Normal. Every collection of information shows who was asked. It shows who could be reached. It shows who was studied. People who are often left out in the world are often left out of data too.
• Old Unfairness. Unfair things that happened in the past can show up in new data. Imagine a company only hired men for fifty years. If you train a computer with that old hiring data, the computer will keep hiring mostly men. It just learned the old unfairness.
• Missing Pieces. Who answered the survey? Only people with fast internet? Only people who speak English? If you miss some groups, your information is already unfair. These are choices people make.
• Labeling Choices. People bring their own ideas when they label things. What one person calls “professional clothes” might be different for someone else. Their choice becomes part of the data.
• Checking Everyone. A computer model might work well on average. But you still need to check how it works for different groups. A face scanner might work 99% for light-skinned faces. But maybe it only works 85% for dark-skinned faces. The average looks good, but it’s not fair for everyone. We have to check all the groups.
• No Perfect Computers. No information is perfect. No computer model is perfect. People made choices to build them. So no computer system is ever truly neutral.

Skew grew up in a village where everyone watched out for each other. Her family were the “watch-mongooses.” They had to be alert all the time. Not just sometimes, but always. They learned that being ready was a way of life. It wasn’t just a task you finished. Skew carried that lesson with her.

When she was twelve, she walked to the big learning center. A wise old mentor named Sift asked her a question. “What does it mean to be ready for unfairness?” Skew answered right away. “Whose data is in here? Whose is missing? Who decided? These three questions don’t quit.” She added, “Unfairness is built into things. Being ready is how we live.” Sift smiled. “You are the one,” Sift said. “Your job is super important for everything we do here.”

In her workshop, Skew had charts pinned to the walls. Each one showed a different collection of information. Next to each chart, she had written the answers to her three questions.

“Look at this one,” she said, shining her data-light on a chart of faces. “It says ‘1 million faces.’ Sounds like a lot, right?” She tapped the chart. “But whose faces? Mostly young people. Mostly light-skinned. Mostly from one part of the world. And all photographed in good light.” She moved her finger down the chart. “Whose faces are missing? Older people. Darker-skinned people. People photographed in poor light. People from many other places.” Then she pointed to a small note. “Who decided to collect these faces? Three engineers in California. They made these choices back in 2015. Now we know the limits of this information.”

She moved to another chart. “This one tries to guess where crime will happen. Same questions. The information here comes from old arrest records. But those old records show where police used to look for crime. Not where crime really happened.” Skew looked up. “So the computer will learn to look in the same unfair places. The computer didn’t invent the unfairness. It just learned it from the old information. That’s bias from the data.”

She turned to face me, her eyes bright. “I am Skew. I teach about being ready for bias. The way to do it is simple. Ask the three questions about everything. Always.”

She was clear and firm. “Don’t let anyone tell you that information is perfect. No information is perfect. People made choices when they gathered it. People made choices when they labeled things. People made choices when they built the system.” Skew paused. “Asking ‘who decided?’ isn’t being paranoid. It’s being smart. It’s how we make things better.”

“Whose data. Whose missing. Who decided. Three questions. Always.”

---