The Arizona STEM Acceleration Project

AI and Ethics: Day 4

Image Recognition, Text Classification, and Bias

A 7th & 8th grade STEM lesson

Janae Thomas

June 2023

Notes for teachers

• Context: This lesson is the fourth day to a multi-day lesson.
• Each “activity” can be completed within a class period, depending on time and choice OR over the course of a couple days.
• Incorporates (social studies) and reading
• This multi-activity lesson can be used as a stand-alone lesson to help students visualize and analyze the impact of societal bias on AI and what that can mean for our society.
• Lesson 1
• Lesson 2
• Lesson 3
• Lesson 4

List of Materials

• Slides
• Possible printed/guided notes
• Image Recognition: Devices to access Teachable Machine
• Text Classification: MIT Word Analogies
• (Remediation): Digitally saved pictures of dogs and cats — I recommend making a folder
• (Extension/Enrichment): Copies of the article “Even Kids Can Understand That Algorithms Can Be Biased”

AZ STEM Standards

Computer Science Standards

• 7.CS.D.1: Identify some advantages, disadvantages, and consequences with the design of computer devices based on an analysis of how users interact with devices.

​

​

Science and Engineering Practices

• ask questions and define problems
• engage in argument from evidence

​

• ​
• ​

ELA Standards

• 7.W.4: Produce clear and coherent writing in which the development, organization, and style are appropriate to task, purpose, and audience.
• 7.SL.1: Engage effectively in a range of collaborative discussions (one‐on‐one, in groups, and teacher‐led) with diverse partners on grade 7 topics, texts, and issues, building on others’ ideas and expressing their own clearly.

Objective(s):

• SWBAT develop algorithms in order to train an image recognition program.
• SWBAT share personal experiences and opinions about AI and technology utilizing evidence from their own experiences and texts.

Agenda (lesson time)

Day 4 Agenda

• Image Recognition: 20-30 Minutes
• AI and Bias: 5-10 Minutes
• Text Classification: 15-20 Minutes
• Journal Activity: 5-10 Minutes

• Day 1: Introduction to AI
• Day 2: What can be classified as AI
• Day 3: Ethical Dilemmas and AI
• Day 4: Image Recognition and Text Classification

Important Vocabulary Review

• Robots - machines that sense their environment, do calculations of some kind, and then perform an action
• Technology - skills and tools that people use to achieve goals, often to make life easier
• Artificial Intelligence (AI) - a program made by people that makes computers do things that seem intelligent, or smart, in the same way that humans are intelligent
• Ethics - moral guidelines for how people in society should behave if they want to be fair
• Algorithm - A set of steps or rules to follow in order to solve a problem or accomplish a specific goal.

Image Recognition �(20-30 Minutes)

This activity involves using a FREE website program called Teachable Machine to teach students about how AI uses image recognition.

Have students complete this activity in groups.

• Teachable Machine is A FREE WEBSITE. If it mentions anything about paying for anything, please don’t.

Image Recognition and AI

Just like a typical algorithm, you have:

INPUT

OUTPUT

CAT

Steps to Change Input

Learn features from training data

An image

A label for the image

Computers and Understanding Images

For a computer to understand an image:

1. It takes the image and turns it into pixels.
2. It uses the neural network to combine pixels into features.
1. All the “edges” are combined together in Layer 1 and 2 to recognize this feature.
3. With training, it uses features to determine the images’ class.
• After enough training (more images), it will begin to recognize certain features of “a woman”, for example.

Let’s Try it Together!

We’re going to use a program called Teachable Machine and train it to recognize and classify pictures using our webcams. We’re going to train it how to recognize:

​

• Rock
• Paper
• Scissors

Your Turn!

Start a new project on Teachable Machine and train it to distinguish between cats and dogs.

• Upload a variety of images.
• Think about the specific features that make the two species different.

After training the program, test it!

Image Recognition - Reflection Questions

• Who was able to program their AI to recognize a cat and a dog majority of the time?
• What was easy? What was hard?
• What were some mistakes you all made in the beginning? Were you able to fix the mistakes or adjust accordingly?

​

Write a 2-4 sentence reflection on the following question:

How can AI image recognition possibly cause ethical dilemmas in the future?

AI and Bias (5-10 Minutes)

Possible Discussion Questions

​

• What is algorithmic bias? What causes algorithmic bias?
• What are some examples of algorithmic bias identified in the video? Who did they impact?
• If algorithmic bias exists in technology, how might that impact society?
• Based off what you know about training algorithms, how could this problem be fixed?

​

Text Classification �(15-20 Minutes)

This activity involves using a free program from MIT.

Have students complete this activity in groups.

• This website is A FREE WEBSITE. If it mentions anything about paying for anything, please don’t.

Text Classification and AI

Same thing, it’s a simple algorithm:

INPUT

OUTPUT

Steps to Change Input

“Please come here”

“Forward”

Compare input to training data

User speech or text

Corresponding label

Computers and Understanding Words

For a computer to understand words:

1. It takes the input word(s) and change them into word vectors.
1. cat - “Living being”, “feline”, “gender”
2. It uses the input word(s) neighboring words to compare the word(s) to the training data.
• “cat” is very close to “kitten”

Word Analogies

Sometimes, problems can arise in text classification.

​

To visualize this, let’s use this website Exploring Word Analogies.

​

Take some time to look at the other classification, specifically the Jobs graph.

Word Analogies Discussion

• What did you discover when you looked at Jobs and gender in word analogies?

​

• What products do you know of that may use word analogies/embeddings?

​

• Sometimes we can’t fix training data. How else might we go about handling bias in AI algorithms?

​

Let’s think…what could be the problem?

• Google “Lawyers” vs. “Court Assistant”
• What do you see?
• Google “Doctors” vs. “Nurses”
• What do you see?

Journal Activity (5-10 Minutes)

• In 4-6 sentences, respond to the following prompt:
• More and more, we are starting to see more versions of AI being developed and, eventually, utilized by our society at large. Do you feel this a step in the right direction? Why? What are some possible benefits and consequences that can arise from using this technology more often in our society?

Differentiation

• Any of the slides contain notes can be printed to provide to students to take notes or follow along.
• Different grouping strategies can be used for all these activities.
• For the Image Recognition activity, pictures of cats and dogs can be provided as guidance.

Remediation

Extension/Enrichment

• For the Image Recognition activity, students can be encouraged to train the system to recognize other common pets, like fish, rodents, reptiles, etc.
• It is possible to upload the Teachable Machine models to block coding programs like Scratch and have the bots recognize an image and act accordingly. This takes a bit of “playing” with the program, but you would export the model and add the Teachable Machine extension to Scratch.
• The website Most Likely Machine goes into further details about algorithmic bias and gives excellent examples of how it affects people. It also has a cool little game to really show how algorithmic bias comes about!
• There’s this really good article from Scientific American about algorithms and bias and it can be provided to students using a reading or Kagan strategy.