Real-World Practice:
Judging AI Credibility from Classrooms to Workplaces and Scams
AI literacy ultimately has to return to real situations. Students deal with assignments and research, professionals with summaries and meeting notes, and social users with scam ads and fake media. Each scenario has different judgment criteria. This chapter uses cases to practice what can be used, what must be verified, and what should never be accepted at face value.
- For each case, read the "Background" first and try to judge: Real / Fake / Mixed?
- Then read "Verification Process" and "Final Determination," compare your judgment with actual results
- Pay special attention to "Tools Used" and "Detection Key" in each case — these are directly applicable skills
Political Manipulation Cases
Background: Around Taiwan's 2024 presidential election, multiple AI-manipulated political videos and audio clips circulated. Publicly documented cases include altered campaign videos and a post-election audio clip claiming President Lai Ching-te privately complained about former president Tsai Ing-wen.
The verification point is not only whether the voice sounds similar. Check together: ① whether a credible original speaking occasion exists; ② whether official channels or mainstream media reported it; ③ whether detector scores are only supporting signals; and ④ whether the content fits the public record and known context.
RFA / Asia Fact Check Lab found no evidence supporting the 2024 "Lai complaining about Tsai" audio and assessed it as likely AI-generated. Taiwan FactCheck Center also documented AI-manipulated election videos, warning that political media cannot be judged only by whether it looks or sounds plausible.
- Source verification: Any "candidate statement" without corresponding news coverage, journalists present at livestream, or official release is highly likely to be fabricated
- Consistency check: Is this statement consistent with the candidate's known positions? Drastic position reversals without mainstream media coverage should be highly suspected
Disaster & Emergency Cases
Background: During the mass COVID-19 vaccination period, multiple videos circulated globally on social media claiming to show "immediate death after vaccination" or "severe adverse reactions," generating vaccine hesitancy.
AFP Fact Check examined a video claiming a girl collapsed after receiving a COVID-19 vaccine; reverse search showed the same footage had appeared before the pandemic, so it could not be evidence of a COVID-19 vaccine reaction. CDC guidance also notes that syncope can occur after vaccination and must be interpreted medically, not inferred from a single viral clip.
① Reverse image search: extract key frames and look for earlier appearances; ② Timeline check: confirm whether the footage predates the claimed event; ③ Medical context: separate "collapse," "fainting," and "death" from vaccine causation instead of inferring a medical conclusion from one clip.
This type of case demonstrates a classic false-context technique: real footage can create a completely different meaning when paired with a wrong time, wrong cause, or unsupported medical conclusion. Reverse search and medical sources must be used together.
Corporate Fraud Cases
Background: Between 2023-2024, scammers increasingly used AI-generated content to package identities, including fake investment advisors, fake company websites, forged credentials, altered images, cloned voices, and deepfake videos. These assets make fraudulent accounts look more like real professionals or legitimate firms.
- Real professionals have official records: Real lawyers are registered with bar associations; real doctors have license numbers that can be checked; real companies have registration numbers. Verify credentials first before deciding to engage.
- Images are not identity proof: Reverse-search suspicious "professional" photos, but no historical record is only a risk signal, not proof by itself.
- Video calls are not final proof: Live video can raise the bar, but it does not replace license lookup, company registration checks, physical contact information, and second-channel confirmation.
"Real But Falsely Accused of Being Fake" Cases
Background: After Russia's 2022 invasion of Ukraine, large volumes of photos and videos circulated on social platforms. Bellingcat and the Centre for Information Resilience (CIR) maintained the Russia-Ukraine Monitor Map to document, geolocate, and verify visual material through open-source methods.
Bellingcat describes this verification work as a combination of geolocation, chronolocation, satellite imagery, and cross-checking against other open-source evidence. The point is not whether a single frame "looks real," but whether the image can be placed into a verifiable location, time, and event context.
This case demonstrates why "trust your eyes" is not enough. The correct response is not to believe everything or doubt everything, but to run systematic verification. Geolocation, timeline reconstruction, and multi-source checks can expose fabricated content while also protecting authentic visual evidence.
Self-Test: Three Practice Questions
You receive a video in a LINE group showing a famous politician saying something "shocking." The video was just posted and there's no news coverage yet. What should be your first step?
Reference answer: ① Confirm source ② Search video key frame (screenshot then reverse image search) ③ Check official channels ④ Wait for at least one credible media report before forwarding
You see a photo claiming to show "a protest that happened yesterday," but the clothing styles and slogan language look a bit strange to you. What's your verification process?
Reference answer: ① Google/Yandex reverse image search ② TinEye "Sort by Oldest" to find earliest appearance ③ Observe era markers like clothing, signs, car models in the photo ④ Use FotoForensics for EXIF and ELA analysis
You receive a phone call where the speaker's voice sounds very much like your boss, urgently requesting you transfer funds to a "client account" and saying "don't tell anyone else." What's your response?
Reference answer: No matter how similar the voice, hang up and directly dial your boss's known personal mobile to confirm. "Urgent" + "Keep it secret" + "Transfer money" are three core fraud trigger words — when all three appear together, your alert level should be at maximum.
Congratulations on completing the seven-chapter media forensics course. Remember: media literacy is a skill requiring continuous practice. Here are recommended next steps:
- MIT Detect Fakes — Interactive deepfake identification training
- Which Face Is Real — GAN synthetic face identification practice
- Go to this platform and upload real images/videos, read AI forensic reports, understand each detector's output
- Read the Detector Details page to understand each AI model's academic basis and applicable scenarios