What is Diffusion? AI Image Generators in Action


This activity demonstrates image diffusion with a simulation of  iterative learning, a key concept in both human and artificial intelligence (AI). By simulating the AI process of refining randomness into structured output, it provides a hands-on understanding of complex AI principles. The task encourages critical thinking, problem-solving, and collaboration, mirroring real-world scenarios and teamwork. Importantly, it fosters a growth mindset, teaching students the value of persistence and adaptability in learning.

Materials Needed Time needed 
  • Computer with projector to demonstrate AI image generation process
Approximately 30 minutes 
  • Students will be able to explain the process of diffusion in both natural systems and AI image generation.
  • Students will be able to demonstrate an understanding of iterative learning by participating in an activity that mimics the AI diffusion process.
  • Students will be able to employ trial-and-error strategies to approach and solve a problem with limited initial information.
Key Concepts & Vocabulary 

Diffusion (in AI Context): Gradual transformation from randomness to structured output in image generation.

Noise to Image: The process of generating random noise (a completely disordered set of pixels), and then over a series of steps, gradually refining the noise into a coherent image.

Iterative Learning: Repeatedly adjusting actions based on feedback to achieve a specific goal.

Feedback: Responses or signals used to guide adjustments in a learning process.


Part 1: Introduction to Diffusion

Define Diffusion: Briefly explain natural diffusion.Say: When you put a drop of ink in water, it spreads out gradually. This is natural diffusion.Transition to AI diffusion Say: AI diffusion in images is the opposite process – Gradually transforming chaos (random noise) to order (structured image).

AI’s Learning Process: Describe how AI learns from data to transform noise into images, guided by user prompts.Humans provide feedback on computer-generated images, tagging portions of them as similar to text prompts.This is similar to people tagging a friend in an image on social media. If computers are allowed to “learn” from these tags, they will associate certain pixel patterns with specific text.

Part 2: Setting up the Activity

Activity Overview: Inform students they will engage in an activity to simulate AI diffusion.State that you have a goal in mind for them, and they have to figure out your thoughts through experimentation.

Expectation Setting: Clarify that you’ll provide thumbs-up for actions getting closer to your undisclosed goal, and thumbs-down otherwise. This is similar to a hotter-colder process of finding an unnamed object.Emphasize trial and error, mirroring AI’s learning process.

Part 3: The Inductive Process Activity

Begin Without Guidance: Start the activity without any instructions or prompts.Choose a goal action for the students to complete, such as “drawing a dog on the board,” but do not tell the students the goal. The instructions will use that action as a target activity. Other possibilities could include reading from a specific book, making  a tower out of objects in the classroom, doing a specific dance move, etc.

Feedback Based on Actions: Sit back and observe the students. Give a thumbs-up as students move toward the whiteboard or show curiosity about it.As students interact with the environment (approaching the whiteboard, picking up a marker, etc.), give thumbs-up or thumbs-down.Your feedback is based on actions leading toward the goal (drawing a dog).

Guiding with Feedback: Continue to give non-verbal feedback, guiding students towards drawing on the whiteboard.As they start drawing, your feedback helps them adjust their actions towards drawing a dog.

Concluding the Activity: Once a student successfully begins to draw a dog, conclude the activity.This activity could be completed again with other goal actions.Acknowledge the collaborative effort and learning process.

Part 4: Debrief and Discussion

Link to AI Diffusion: Discuss how the activity mirrors the AI diffusion process of transforming random actions into a structured goal.Illustrate the process by using an AI image generator (see links at the end of the lesson) to create an image based on student prompt ideas.Remind students that this process is the result of many repeated inputs by human users providing feedback on computer image generations.

Discussion Questions
How did you feel when you first started the activity without any clear instructions?What strategies did you use to figure out what the teacher was thinking?How did the teacher’s feedback (thumbs-up/thumbs-down) influence your decisions during the activity?What role did communication play in your group’s ability to reach the goal?Would this be a good way to learn things in school? In what ways is human learning different from how AI learns?Were there moments of frustration, and how did you deal with those feelings? How is this different from how AI “learns”?Can you think of real-world examples where a similar approach (trial and error, learning from feedback) is used?
Supplemental Activity Ideas 

Advanced Art Project with Iterative Feedback: Students work together to create a complex art piece (like a collage or a sculpture) in stages. Designate one or two students as the goal setters, and then have other students make attempts toward their goals, refining as they go.

Interactive Story Writing: Students write a story in a choose-your-own- adventure style, where the plot changes based on reader decisions. They can refine the story branches based on feedback from initial readers.

Debate on AI Ethics: Organize a debate or discussion on the ethical implications of AI, especially regarding learning from data and making decisions. Encourage research and preparation, followed by iterative refinement of arguments based on peer feedback.

Sources to Learn More
Ideogram – AI image generator that illustrates the diffusion process well by gradually improving the image.
Perchance – Simple AI image generator that does not require a loginInformation on what AI diffusion models are doing – https://learnopencv.com/image-generation-using-diffusion-models/