Questions for ‘Artificial intelligence is making it hard to tell truth from fiction’

Taylor Swift is looking at the viewer.Her hair is up and she is wearing huge diamond and purple gem earrings. with a navy blue outfit is covered in glittering beads.

Earlier this year, mega-celebrity Taylor Swift (shown here) became the target of a deepfake disinformation campaign. People circulated compromising and realistic-looking AI-generated fake photos of her on social media sites. Increasingly, such AI-generated fake images, text and video can seem more convincing than the real thing.

Matt Winkelmeyer/Staff/Getty Images Entertainment

To accompany Artificial intelligence is making it hard to tell truth from fiction’  


Before Reading:

  1. Disinformation refers to lies; misinformation refers to inaccurate information. Give one example of disinformation that you’ve encountered, followed by one example of misinformation.
  2. Artificial intelligence has advanced rapidly over the last few years. To what extent do you think this technology contributes to increased false information, such as in news and other media? Explain your answer.

During Reading:

  1. Give two examples of how deepfakes have been used to spread false information.
  2. What does it mean to say modern AI technology is “generative”?
  3. How do generative AI models produce text?
  4. Why is a diffusion model described as a “forward and a backward” process?
  5. What were the results of Federico Germani’s study, where people tried to judge whether tweets (now X posts) came from an AI model or real humans?
  6. What did Michoel Moshel discover about people’s ability to discern real human faces from AI-generated faces?
  7. What does the phrase “liar’s dividend” mean? Name two potential consequences of the liar’s dividend.
  8. What did Zain Sarwar’s team learn when it tested several AI-detection tools? Where did these tools perform well? Where did they fall short? Which tool performed the best and why?
  9. What are digital watermarks? How could they be used to help people spot AI-made content? What is a limitation of using watermarks for this?
  10. Give one example of how existing laws have been used to deal with AI-related problems.

After Reading:

  1. A generative adversarial network — or GAN — relies upon two AI models rather than just one to generate new content. Based on how these models work together, why is this type of network described as “adversarial”? Explain the value of having two adversarial models working to create new content rather than just one model. Would you expect the quality of content produced by a two-AI model such as GAN to increase, decrease or remain the same over time? Explain your answer.
  2. How might the rise of AI technology contribute to a liar’s dividend-type situation? How do you think this might affect you? Imagine a world where such disinformation is far more prevalent. How might interacting with others in such a world differ from how we interact today?