AI & Digital Ethics Grade 9 Deepfakes and Misinformation at Scale

Study Guide: Deepfakes and Misinformation at Scale
Grade 9 | AI & Digital Ethics

1. The Driving Question

What happens when anyone can make a video of the president declaring war, a celebrity saying something they never said, or your best friend asking for money—and no one can tell it’s fake? How do we decide what to believe when technology can fabricate reality faster than we can verify it?

2. The Core Idea — Built, Not Listed

Imagine you’re scrolling through your phone and see a video of your favorite musician announcing they’re canceling their tour. The video looks real—same voice, same facial expressions, even the same lighting from their last concert. But when you check their official account, there’s no announcement. The video was a deepfake: an AI-generated clip trained on thousands of hours of the musician’s real interviews, then manipulated to say something new. Deepfakes don’t just trick individuals; they can flood social media in minutes, making it impossible to separate truth from fabrication. This isn’t just about one fake video—it’s about scale. AI tools now let anyone create convincing fakes, and social media algorithms spread them faster than fact-checkers can debunk them. The result? A world where seeing isn’t believing, and trust in information erodes.

Key Vocabulary:
- Deepfake
Definition: A hyper-realistic digital forgery of a person’s likeness (face, voice, or both) created using AI, often to spread false information.
Example: A deepfake of a CEO announcing a company’s bankruptcy, causing its stock price to crash before the hoax is exposed.
College Shift: In advanced media studies, deepfakes are analyzed as part of synthetic media—a broader category that includes AI-generated text, audio, and even entire fabricated events.

• Misinformation at Scale
  Definition: False or misleading information spread rapidly and widely, often amplified by algorithms or bots.
  Example: A fake tweet claiming a hurricane will hit Miami in 24 hours, shared by thousands before meteorologists can correct it.
  College Shift: In political science, this is studied as computational propaganda, where AI and automation are used to manipulate public opinion.

• Generative Adversarial Network (GAN)
  Definition: A type of AI where two neural networks compete—one creates fakes, the other tries to detect them—improving the quality of forgeries over time.
  Example: The AI tool DeepFaceLab uses GANs to swap faces in videos, making it possible to create deepfakes with just a few photos of a person.
  College Shift: In computer science, GANs are explored for their potential in art, medicine (e.g., generating synthetic medical images), and even cybersecurity threats.

• Liar’s Dividend
  Definition: The phenomenon where real evidence (like a genuine video) is dismissed as fake because deepfakes have made people skeptical of all media.
  Example: A politician caught on camera making a racist remark claims it’s a deepfake, and supporters believe them—even when forensic experts confirm it’s real.
  College Shift: In law and ethics, this concept is tied to epistemic crisis—the erosion of shared truth in public discourse.

3. Assessment Translation

Format: This topic appears on AI ethics assessments (e.g., ISTE standards, state digital literacy exams) and AP Computer Science Principles (as part of the "Global Impact" unit). Expect: - Short-answer questions analyzing a deepfake scenario (e.g., "How might this video be used to manipulate an election?").
- Evidence-based writing (e.g., "Argue whether social media platforms should be legally required to label deepfakes").
- Multiple-choice questions testing definitions and real-world applications (e.g., "Which of these is an example of the liar’s dividend?").

Distractor Patterns in Multiple Choice:
- Confusing deepfakes with cheapfakes (low-tech edits like speeding up a video).
- Assuming all AI-generated content is malicious (ignoring creative or educational uses).
- Overestimating the role of detection tools (many deepfakes are still undetectable by current tech).

Proficient vs. Developing Responses:
| Proficient | Developing | |----------------|----------------| | Explains how GANs work in simple terms (e.g., "two AIs competing to make fakes better"). | Describes deepfakes as "videos that look real but aren’t" without explaining how. | | Identifies both the risks (e.g., election interference) and potential benefits (e.g., AI-generated voices for people with disabilities). | Focuses only on risks or dismisses deepfakes as "just a prank." | | Proposes a solution (e.g., "watermarking AI content") and explains its limitations. | Suggests "just don’t believe anything online" without a practical strategy. |

Model Proficient Response (Short Answer):
Prompt: "A deepfake video of a mayor accepting a bribe goes viral. How might this affect a local election, and what could the mayor’s team do to respond?" Response: The deepfake could sway voters by making the mayor seem corrupt, even if the video is fake. Since deepfakes spread fast, the mayor’s team should: 1. Release the original footage (if available) to prove the video was altered.
2. Work with fact-checkers like Snopes to debunk the fake.
3. Use the same platforms (e.g., Twitter, TikTok) to share the truth, since algorithms favor viral content.
However, even with these steps, some voters might still believe the deepfake because of the liar’s dividend—people may assume all videos are fake now.

4. Mistake Taxonomy

Mistake 1: Overestimating Detection Tools
Prompt: "How can social media platforms stop deepfakes from spreading?" Common Wrong Response: "They can use AI to detect deepfakes automatically." Why It Loses Credit: This ignores that detection tools are often reactive—they can’t catch new deepfakes until they’re trained on them. It also doesn’t address how platforms might slow the spread (e.g., labeling unverified content).
Correct Approach: Platforms can: - Label AI-generated content (e.g., TikTok’s "AI-generated" tags).
- Delay viral spread by limiting shares until fact-checkers review content.
- Educate users on how to spot fakes (e.g., checking for unnatural blinking or lighting).

Mistake 2: Assuming All Deepfakes Are Malicious
Prompt: "What are the ethical concerns of deepfake technology?" Common Wrong Response: "Deepfakes are always bad because they spread lies." Why It Loses Credit: This ignores positive uses (e.g., AI-generated voices for ALS patients, or restoring old films). A strong response acknowledges both risks and benefits.
Correct Approach: Deepfakes raise ethical concerns like: - Misinformation (e.g., fake news swaying elections).
- Consent (e.g., using someone’s likeness without permission).
But they also have benefits, such as: - Accessibility (e.g., AI voices for people who’ve lost theirs).
- Education (e.g., historical figures "speaking" in documentaries).

Mistake 3: Confusing Deepfakes with Cheapfakes
Prompt: "Which of these is an example of a deepfake? A) A video slowed down to make someone look drunk. B) A video of a celebrity’s face swapped onto another person’s body." Common Wrong Response: "A, because it’s a fake video." Why It Loses Credit: This confuses deepfakes (AI-generated) with cheapfakes (low-tech edits). Deepfakes require AI, while cheapfakes use simple tools like speed adjustments.
Correct Approach: The answer is B. Deepfakes use AI (like GANs) to create hyper-realistic forgeries, while cheapfakes rely on basic editing tricks.

5. Connection Layer

1. Within AI & Digital Ethics → Algorithmic Bias
   Why it matters: Deepfakes are often trained on biased datasets (e.g., more images of white faces), making them less accurate for people of color. Understanding this reveals how all AI systems can reinforce inequalities.

2. Across Subjects → History (Propaganda in WWII)
   Why it matters: Deepfakes are the digital version of WWII propaganda posters—both use emotional manipulation to spread misinformation. Studying past propaganda helps predict how deepfakes might be used in modern conflicts.

3. Outside School → TikTok’s "AI-Generated" Labels
   Why it matters: TikTok now tags AI-generated content, forcing users to think critically about what they see. This is a real-world example of how platforms are adapting to deepfakes—something you’ll encounter every time you scroll.

6. The Stretch Question

If a deepfake video of a world leader declaring war goes viral, and no one can prove it’s fake within 24 hours, should social media platforms be legally required to remove it—even if it might be real?

Pointer Toward the Answer:
This isn’t just about technology—it’s about power. If platforms remove content too quickly, they might censor real emergencies (e.g., a leader actually declaring war). But if they wait, the damage could be irreversible. Some argue for a "trust but verify" system: platforms could label the video as unverified and slow its spread, while governments and fact-checkers investigate. Others say this is too slow—by the time the truth comes out, the war could have started. The real question is: Who gets to decide what’s true in a crisis?