5.3 Computing Bias

Computing bias happens when a computing innovation reflects unfair human bias, either through the algorithm's rules or the data the innovation uses. Bias can appear at any stage of software development, from planning and data collection to testing and deployment. For AP Computer Science Principles, explain where bias comes from, who is affected, and how programmers can reduce unfair outcomes.

Why This Matters for the AP Computer Science Principles Exam

Computing bias falls under the Impact of Computing big idea, which is one of the most heavily weighted parts of the multiple-choice section. On the exam, you may read a passage about a computing innovation and answer questions that ask you to evaluate it based on legal and ethical factors. Being able to explain where bias comes from and how to reduce it is exactly the kind of thinking those questions reward.

This topic also connects to your investigation of computing innovations during the course. When you study how an innovation uses data and affects people, recognizing possible bias helps you describe harmful effects clearly and accurately.

more resources to help you study

practice multiple choiceFRQ practice & scoringcheatsheetsscore calculator

Key Takeaways

• Bias in computing comes from two main sources: bias built into the algorithm itself and bias in the data the innovation uses.
• Bias can be embedded at every level of software development, from early planning to the work done after release.
• Computing innovations can reflect and even reinforce existing human biases, not just create new ones.
• Programmers are responsible for taking action to reduce bias as a way of combating existing human biases.
• Using diverse, representative data and reviewing algorithms for unfair outcomes are practical ways to reduce bias.

What Is Computing Bias?

Bias is a tendency or inclination, especially one that is unfair or prejudicial. Everyone has biases, but biases based on someone's identity can be harmful to society.

Computing innovations use data from the world around them, and people choose which data to feed into those innovations. Because that data and those design choices come from people, computing innovations can reflect existing human biases. Bias can enter in two main ways:

• Bias written into the algorithm: the rules or logic of the program treat groups unfairly.
• Bias in the data: the information used to build or train the innovation already carries unfair patterns.

A key point for the exam is that bias can be embedded at all levels of software development. It is not just a data problem or just a coding problem. It can appear when a team decides what problem to solve, what data to collect, how to build the algorithm, and how the product is used after release.

Examples of Bias

These are examples that show how bias can appear in real computing innovations. They are illustrations of the concept, not required AP content.

• Criminal risk assessment tools try to predict the chance that a defendant will re-offend, and that prediction can influence decisions in the justice system. These tools are trained on historical data, and historical data can carry bias against certain races and classes. As a result, the tool may flag people from certain groups as higher risk more often.
• Facial recognition systems are sometimes trained on data sets that include far fewer images of women and people of color than of white men. When the training data is not diverse enough, the system can perform unevenly across groups.
• Recruiting algorithms sort large numbers of job applicants. If past successful candidates were mostly men because mostly men applied, the algorithm can "learn" to prefer male applicants. You can read about a real case of this with Amazon's scrapped AI recruiting tool.

In each example, the innovation did not invent the bias on its own. It reflected bias that already existed in the data or in the choices people made while building it.

How to Reduce Bias in Computing

Programmers should take action to reduce bias in algorithms as a way of combating existing human biases. The first step is acknowledging that bias could exist at all. From there, these are practical actions:

• Use diverse and representative data: data that reflects the full population helps reduce bias in the models built from it.
• Review and test algorithms for bias: checking algorithms and testing them on diverse data can reveal unfair outcomes before release.
• Apply fairness checks: comparing outcomes across groups helps catch results that treat people unequally.
• Stay aware of human bias: since people design these systems, actively looking for human bias is part of the job.
• Bring in more perspectives: a team with varied backgrounds is more likely to spot bias that one narrow group might miss.

How to Use This on the AP Computer Science Principles Exam

MCQ

Expect to read about a computing innovation and then answer questions about its effects. When a question asks you to evaluate an innovation based on ethical factors, look for where bias could enter: the data used or the algorithm's logic. Be ready to identify that bias can appear at any stage of development.

Source Analysis

When a passage describes how an innovation was built and what data it uses, ask yourself who or what might be left out of that data. Connecting biased data to unfair outcomes is the core skill this topic tests.

Common Trap

Do not assume that "the computer is neutral" or that bias only comes from bad intentions. Bias often enters unintentionally through the data, and a fair-looking algorithm can still produce unfair results.

Common Misconceptions

• Computers are objective, so they cannot be biased. Computing innovations reflect the data and design choices made by people, so they can carry the same biases people have.
• Bias only comes from the data. Bias can also be written into the algorithm itself, and it can be embedded at every level of software development.
• Bias is always intentional. Much computing bias is unintentional, like training a system on a data set that does not represent everyone.
• There is nothing programmers can do about it. Programmers are expected to take action to reduce bias, such as using representative data and testing algorithms for unfair outcomes.
• Fixing bias means the algorithm is broken. Reducing bias is part of responsible development, not a sign that the program failed. It improves both the program and its impact on society.

Related AP Computer Science Principles Guides

• 5.6 Safe Computing
• 5.1 Beneficial and Harmful Effects
• 5.5 Legal and Ethical Concerns
• 5.2 Digital Divide
• 5.4 Crowdsourcing
• Big Idea 5 Overview: Impact of Computing