Skip to content

1. Introduction to Fairness in Machine Learning

Playback Speed:
Transcript

Machine learning is changing our world in many ways.

You encounter the results of machine learning when you: Use a social media app that can recognize people’s faces ...

Have your signature verified when you endorse a check at your bank ...

Ask a question to a customer service chat bot on a website ...

Or see a recommended product based on your previous shopping habits.

Even Google Search uses machine learning models to give you results that best match your unique search history.

Many machine learning innovations are helpful, fun, and can protect personal information.

However, the outcomes produced by machines can sometimes be unfair.

An unfair outcome is one that shows favoritism, self-interest, deception, or bias.

Bias means preferring one thing and, therefore, not giving an equal chance to others.

For instance, many hiring managers use computer programs to preview resumes and identify the most qualified candidates for a job.

This can save people a lot of time, but the data used to train the machine is crucial.

If the machine is trained on data that suggests that a certain type of person is more successful at the job, the program will learn that and be more likely to exclude people who are different.

This example might sound surprising because you probably expect computers to be logical and objective.

But the machine is just learning what humans have taught it.

That’s why fair training data is so important when teaching a machine about any topic.

Even the most experienced programmers constantly work to achieve fairness in their computer programs.

The goal of fairness in machine learning is to train systems that make fair predictions across all groups.

In this lesson, you will reflect on fairness in machine learning and consider the benefits and drawbacks.

You will choose a type of item from a list ...

Collect images that represent that item ... And reflect on the images you chose, deciding whether anything was excluded.

For example, would the computer still recognize an item if it were a different color?

A different shape?

Or made of a different material?

Then, you will apply your discoveries to a real-world machine learning scenario.

Now, move on to the next video to choose your item.

Next