New game
Download
SCORM download
To be used in an LMS (Moodle, Blackboard...)
Regular download
To be published on a website or blog.
Downloading games is an exclusive feature for users with an Academic Plan or a Commercial Plan.
Get your Academic Plan or your Commercial Plan now and start integrating your games into your LMS, website or blog.
Get Academic Plan
If you wish, you can download a demo game here and test its integration:
Share game
Quiz
Quiz

Bias Detective: What Went Wrong?

Integrate it into your platform
G. Classroom M. Teams
You can integrate the game into an LMS compatible with LTI 1.1 or LTI 1.3 such as Canvas, Moodle, or Blackboard. This way, the scores will be automatically saved into the platform’s gradebook.
Download
You have exceeded the maximum number of games you can integrate into Google Classroom with your current Plan.

To integrate as many games as you want in Google Classroom, you need an Academic Plan or a Commercial Plan.

Get Academic Plan
You have exceeded the maximum number of games you can integrate into Microsoft Teams with your current Plan.

To integrate as many games as you want in Microsoft Teams, you need an Academic Plan or a Commercial Plan.

Get Academic Plan
Downloading games is an exclusive feature for users with an Academic Plan or a Commercial Plan.

Get your Academic Plan or your Commercial Plan now and start integrating your games into your LMS, website or blog.

If you wish, you can download a demo game here and test its integration:

Get Academic Plan

Bias Detective: What Went Wrong?

Quiz

Played 36 %Accuracy 75 Average time 01:10
Create copy

About this activity

Machine learning systems do not intentionally discriminate, but they can reflect patterns and problems present in the data they learn from. In this activity, you will analyze real-world scenarios and identify what may have gone wrong. Think carefully about whether the issue is related to biased data, poor data preparation, flawed design, or something else entirely.

Created by

International American Institute
International American Institute
Ecuador

Download the paper version to play

Make your own free game from our game creator
Compete against your friends to see who gets the best score in this game
Make challenge

Top Games

 %
Anonymous
Anonymous
 %
%
%
You have exceeded the maximum number of games you can print with your current Plan.

To print as many games as you want, you need an Academic Plan or a Commercial Plan.

Get Academic Plan
Print your game
Bias Detective: What Went Wrong?
 

Bias Detective: What Went Wrong?Online version

Machine learning systems do not intentionally discriminate, but they can reflect patterns and problems present in the data they learn from. In this activity, you will analyze real-world scenarios and identify what may have gone wrong. Think carefully about whether the issue is related to biased data, poor data preparation, flawed design, or something else entirely.

by International American Institute
1

A facial recognition system performs well on lighter-skinned faces but frequently misidentifies darker-skinned individuals.

2

A student performance prediction model uses grade records that contain many missing values and duplicated entries.

3

A loan approval system denies more applications from a specific neighborhood because historical data shows higher default rates there.

4

A recommendation system repeatedly suggests similar content, making it harder for users to discover new topics.

5

A health prediction model was trained mostly on data from adults but is used to predict conditions in teenagers.

Feedback

This problem likely occurred because the training dataset did not include enough diverse examples. When certain groups are underrepresented, the model performs poorly on those groups. This is a classic example of sampling bias.

Missing and duplicated data can distort patterns the model learns. Without proper cleaning and preparation, the model may draw inaccurate conclusions from incomplete or messy data.

The model is learning patterns from past data. If historical decisions were influenced by inequality, the model may reproduce those patterns. This is an example of bias embedded in historical data.

When users interact mostly with similar content, the system continues reinforcing those patterns. This creates a feedback loop that limits diversity in recommendations.

If the training data does not represent the population where the model is applied, performance may be unreliable. The model lacks relevant examples for teenagers.

 
Are you sure you want to leave the page?

If you leave the page, you will lose your game progress.