Lesson Plan: How to detect deepfakes and avoid disinformation

Materials

• Lesson outline
• Student handout
• Projector, Computers, and Internet Connection
• MediaWise – ”Is This Keanu Reeves Account Really a Deepfake?” video
• Reuters – Deepfake Explainer
• Application Activity Resources (Test all links inside the school building)
  • TikTok – Unreal Keanu Reeves account
  • Business Insider – There’s a Terrifying Trend on the Internet that could be used to Ruin Your Reputation, and No one Knows How to Stop It
    • University of Washington – Lip-syncing Obama: New Tools Turn Audio Clips into Realistic Video
    • Mark Zuckerberg deepfake example
    • Kit Harrington as Jon Snow deepfake example
• PBS NewsHour – Why Deepfake Videos are Becoming More Difficult to Detect

Key Vocabulary

Deepfake: Manipulated and computer-generated images/videos used to resemble someone else’s likeness.

• Five techniques of Deepfakes:
  • Face swapping — Deep learning software learns the characteristics of the source and
  • target faces. Software then regenerates the target face using movements mapped
  • from the source face.
  • Face reanimation — Deep learning software learns the movement of the target face
  • and the movement of the source face then regenerates the source face using
  • movements from the target face.
  • Object removal — Deep learning software analyzes the characteristics of a
  • background scene. A target object is cut out of the footage and software generates a
  • new background.
  • Synthetic images — Deep learning software learns characteristics of human faces
  • through being fed thousands of images. A generator then creates synthetic faces
  • using these rules. Images are rejected by the network until they look authentic.
  • Synthetic audio — Artificial Intelligence learns characteristics of a person’s speech then generates new speech using different inputted words.

• Vetting: Critical and precise examination of the validity of something.
• Misinformation: Inaccurate info, often presented through deliberate means.
• Media literacy: The practice of proper contextualization of media content and its purpose.

Activator  (5 minutes)

List two elements about this fake Keanu Reeves profile that make it seem “Legit”Avatar looks like KeanuHas 8.4 million followers

List two elements about this fake Keanu Reeves profile seem “NOT Legit”“unreal_keanu”“Parody, Life & Eternal Youth”

Explain that deepfake videos, images, and audio can be convincing and when they purposefully spread, misinformation they can have serious consequences for national security, democracy, and invasion of an individual’s privacy or reputation. However, even when a deepfake is not trying to harm anyone intentionally a lot of people can be fooled.  

Acquisition (25 minutes)

Display this on the projector and walk through it together or ask the students to work independently or in small groups. Check to see the link and the examples in the resource works in the classroom.

Read through this Reuters lesson about deep fake technology and answer the following questions.

1. The word “deepfake” is a blend of what two words:
   1. Deep learning
   2. Fake
      
2. Define these five types of deepfake techniques in your own words:

Face Swapping	Deep learning software learns the characteristics of the source and target faces. Software then regenerates the target face using movements mapped from the source face.
Face reanimation	Deep learning software learns the movement of the target face and the movement of the source face then regenerates the source face using movements from the target face.
Object Removal	Deep learning software analyzes characteristics of background scene. A target object is cut out of the footage and software generates a new background.
Synthetic Images	Deep learning software learns characteristics of human faces through being fed thousands of images. A generator then creates synthetic faces using these rules. A discriminator network rejects them until they look authentic.
Synthetic Audio	Artificial Intelligence learns characteristics of a person’s speech then generates new speech using different inputted words.

3. List two potential positive uses of this technology:

Advertising and Filmmaking

Trying on clothes or visiting a location virtually

4. List two potential negative uses of this technology:

Sway voter intentions

Blackmail a person or damage their reputation

Watch the MediaWise Teen-Fact Checking Network (TFCN) video

Tell students you will play the video through once without pausing. Then ask them to play it again in small groups or independently to answer the questions and pause it as needed. Circulate around the classroom as students work.

Answer the questions while watching Sophia’s TFCN video. Watch it all the way through once first, then play it again, pausing it  if needed tthe o answer the questions.

1.  (1:10) What is a “deepfake” and how is it created?

Synthetic media that’s generated using artificial intelligence. In deepfake videos, the person’s face can be swapped with another  to make it look like they said or did something they didn’t. Deepfakes are created using deep learning algorithms, which teach themselves from large sets of data, including images and video, so they can create new, authentic-looking media.

2.  (1:37-2:24) List four other examples of how deep fakes have been used recently.

Advertisers to sell products

Ukrainian President Zelensky fake video telling Ukrainian soldiers to surrender

Movies to save money on production

Virtually try on clothes or eyeglasses

3.  (2:40-3:45) – List four red flags used to identify Keanu’s account as a deepfake.

User name “unreal_keanu”;

The word “parody” in the bio

Blotchy skin

Unnatural hair or light on skin

4. List four questions to ask that can tip you off that a video or image is a deepfake:

Do the eyes blink or are they static?

Is the skin tone blotchy and is the lighting off on the face?

Do  fine details may off or undefined such as:Hair not move or seem too perfect to be real?Teeth look weird or undefined?Are there imperfections in earrings or glasses?

4-  Does the video show the person acting “out of character”?

Application (10 minutes)

Display this on the projector and walk through each example or ask the students to work independently or in small groups. Check to see the links and the examples in the resource works in the classroom. Circulate the room as students work.

Application: For each of the deepfake examples, list clues you see that help identify the video as a deepfake. Then explain how a deepfake of this nature could be used in a positive or negative way. The examples are from this article of Business Insider.

Deepfake Example	Clues the video is a deepfake	How could it be used in a negative or positive way?
Former President Obama deepfakes created by the University of Washington		Spread misinformation about public policy or relations with foreign nations. Threaten democracy and national security.
Mark Zuckerberg’s speech about the power of Facebook		Misinformation about products or services could mislead the public and possibly create dangerous situations.
Kit Harrington as Jon Snow apologizing for how “The Game of Thrones” ended.		Individuals could appear to say things they did not or be in embarrassing or illegal situations that they were not.

Assessment (5 – 10 minutes)

This can take the form of a class discussion or a ticket out the door.

Assessment: “What’s the Big Idea?” Discussion Questions

1. How can deepfakes that are created just for fun or entertainment have negative consequences?
   People may not realize that they are actually fake and spread misinformation unintentionally online or by word of mouth.

2. List at least three solutions that could help protect people from being duped by deepfakes. Research this online with keyword searches to gather ideas and list the best solutions you find.

Congress can pass legislation

Social media platforms can develop algorithms or use deep learning software to detect deepfakes and remove them

Public awareness campaigns can be run to educate the public and media literacy taught in schools to educate youth about deepfakes (like this lesson!)

3. Describe why each of your proposed solutions listed above might be ineffective.

Congress is divided and how would the laws be enforced or detected. Even if punished the damage is done already.

We rely on business owners who have an interest in earning money to do the right thing. Also, the technology evolves so it must constantly be updated.

Cost of campaigns and educational resources is expensive. Schools are slow to adapt and change curriculum.

Extension and refining

Watch this PBS NewsHour Story about why and note three new things you learned about deepfakes.

Additional resources

• MediaWise “Is This Legit?” series
• How to combat political misinformation from PBS NewsHour Student Reporting Labs
• Fact-checking lesson for Student Journalists
• Misinformation Overload: Interviews from PBS NewsHour Student Reporting Labs
• StoryMaker Media Literacy Learning
• 10 tips to spot media misinformation
• Reuters Media Literacy Contest
• Journalism in Action, a history of journalism in the U.S., including political satire, from PBS NewsHour Classroom

---

MediaWise is a digital media literacy initiative of the Poynter Institute, a nonprofit, nonpartisan organization. Now in over 170 middle and high schools, PBS NewsHour Student Reporting Labs (SRL) is a national youth journalism program that trains teenagers across the country to produce stories that highlight the achievements and challenges today’s youth face.