Everyday Guidebooks
Cover of AI for Beginners

Business, AI & Entrepreneurship

AI for Beginners: A Course in Bias and Fairness

Understand How Data, Labels, and Measures Shape Automated Decisions

by Todd Michael

Available now

When an AI system makes a recommendation or helps decide who gets an opportunity, how can we tell whether it is fair?

This ai for beginners course offers a clear, practical introduction to the choices behind automated decisions. You will explore how samples leave people out, how labels reflect human judgments, and how a measure of success can hide unequal outcomes. Accessible case studies bring these ideas to life in settings such as hiring, lending, healthcare, and education.

Each chapter includes questions for reflection or group discussion, helping you identify who benefits, who may be harmed, and whose voice is missing. Along the way, you will encounter competing definitions of fairness and learn why better data or a technical adjustment cannot always resolve a problem on its own.

No coding or advanced mathematics required—just curiosity and a willingness to ask better questions.

Contents
  1. 1. AI Decisions in Everyday Life
  2. 2. From a Human Problem to a Model
  3. 3. Who Gets Counted? Sampling and Representation
  4. 4. What the Records Leave Out
  5. 5. Labels Are Human Decisions
  6. 6. Proxies and Hidden Assumptions
  7. 7. What Does Success Measure?
  8. 8. When Fairness Measures Disagree
  9. 9. Communities, Context, and Unequal Harm
  10. 10. A Closer Look at Hiring
  11. 11. A Closer Look at Healthcare
  12. 12. What Technical Fixes Can and Cannot Do
  13. 13. Accountability Beyond the Model
  14. 14. Put the Course to Work
  15. Other Books by Todd Michael

Read a sample: 1. AI Decisions in Everyday Life

Imagine opening a music app and finding a playlist made for you. Some songs fit your taste; others leave you wondering what the app was thinking. You can skip a track, save a favorite, or ignore the list altogether. The recommendation influences what you encounter, but it does not require you to listen.

Now imagine applying for a job and learning that software will help sort the applications. You may not know what information the software uses, how its suggestions are reviewed, or whether a person would notice an error. The system is still producing an output, but the stakes and your ability to respond are very different.

Both examples invite the same starting question: What role is the technology playing in a human decision? A recommendation is not the same thing as an order. A prediction is not the same thing as an observation of the future. And a score does not, by itself, explain what should happen next.

This course is about learning to ask those questions with care. You do not need to program a model or memorize equations. You do need curiosity about how a system is built, what it is asked to do, and how its use affects people whose circumstances may differ from your own.

The term artificial intelligence, or AI, covers a broad range of technologies. In everyday discussion, it often refers to systems that perform tasks associated with human judgment or perception, such as recognizing patterns in images, generating text, or suggesting what a person might want to see.

Many systems discussed in this book use machine learning: an approach in which a computer model learns patterns from examples or other data rather than relying only on rules written out one by one. A model might learn associations between information in past applications and outcomes recorded by an employer, or between a viewer’s activity and the films they choose to watch.

The model’s output depends on the task it was built for. It might recommend a film, assign a category to a document, produce a score, or estimate the likelihood of an event. Some AI systems generate new material, such as a draft email or an image, in response to a request.

Not every automated decision uses machine learning. A fixed rule that moves every application received after a deadline into a separate folder is automation, but it need not involve a model that learned from data. The distinction matters when investigating how a particular tool works. For the person affected, however, a fixed rule can still have consequences worth questioning.

Consider a few familiar settings:

- Recommendations: A service ranks songs, products, videos, or news stories so that some appear before others. - Screening: A tool helps sort applications, requests, or cases for further attention. - Prediction: A model estimates an outcome, such as whether an appointment is likely to be missed. - Classification: A system assigns an item to a category, such as sorting a message into a spam folder. - Generation: A tool produces text, images, or other material that a person may use in a later decision.

These descriptions tell us what a tool appears to do, not whether it does it well or fairly. Two systems can perform the same kind of task while using different information, serving different goals, and exposing people to different risks.

Suppose a clinic uses a model to identify patients who might benefit from a reminder before an appointment. The model produces a list. A staff member then calls people on that list, perhaps prioritizing those with appointments the next morning.

The model’s output is one part of the process. People also decided which records to collect, what outcome to predict, how many patients to place on the list, and what staff should do with it. The clinic’s staffing and phone hours affect what happens next. So does whether patients can answer calls during the day.

A useful way to describe the situation is to distinguish the model from the system around it. The model generates an output. The broader system includes the people, procedures, interfaces, resources, and policies that shape how that output is used.

The distinction can prevent two opposite mistakes. The first is blaming a model for every problem in a process without examining the decisions around it. If staff do not have time to reach patients, a better prediction alone may not solve the problem. The second is treating an output as harmless because a person technically makes the final choice. If a staff member has only a few seconds per case and routinely follows the score, the tool may carry considerable influence even though a human clicks the final button.

Human involvement can be valuable when a reviewer has relevant information, enough time, and the authority to disagree. Merely placing a person at the end of a process does not tell us whether those conditions exist.

A prediction is an estimate about an outcome, not a statement of certainty about a person. A model might estimate that an appointment is more likely to be missed under certain circumstances. It cannot know in advance that a particular patient will miss it. A bus may arrive late, a work shift may change, or the patient may find a way to attend despite obstacles.

Predictions can be useful when they help people prepare or offer support. They can also become…

More Business, AI & Entrepreneurship

Cover of AIAIAvailable nowCover of The New Gold RushThe New Gold RushAvailable nowCover of Leadership Intelligence for Succession PlanningLeadership Intelligence for Succession PlanningAvailable nowCover of Remote Jobs and the Scam TestRemote Jobs and the Scam TestAvailable nowCover of The Entrepreneur Mindset for Patient DecisionsThe Entrepreneur Mindset for Patient DecisionsAvailable nowCover of Price Without ApologyPrice Without ApologyAvailable now

About Todd Michael

Todd Michael is an American author and the founder of ToddMichael.tv. He writes practical, plain-English books on money, health, preparedness, technology, games, humor and more.

About the author Todd Michael · Todd Michael’s author site

Part of the ToddMichael.tv network.