
TED Tech Democracy in the Age of AI (2/5): When AI can fake reality, who can you trust? | Sam Gregory
15 snips
Oct 26, 2026 Sam Gregory, a technologist and human rights advocate, leads Witness, helping people use video to defend human rights. He explores how hyper-realistic deepfakes could fracture our shared reality, distort evidence and elections, and fuel denial. The conversation examines detection tools, media provenance, and accountability across the AI pipeline, asking how we can protect trust in what we see.
AI Snips
Chapters
Transcript
Episode notes
AI Can Fake Reality And Dismiss Truth
- AI makes reality easier to fabricate while making genuine evidence easier to dismiss.
- Deepfakes now affect elections, war-zone documentation, sexual abuse, and news broadcasts—not just isolated internet hoaxes.
Experts Could Not Reliably Separate Truth From Fakes
- Witness experts investigated three disputed audio clips from Sudan, West Africa, and India with uneven results.
- They authenticated Sudanese audio, found the West African clip inconclusive, and judged an Indian politician's recording partly real.
Deepfake Clues Become Obsolete Quickly
- AI deepfakes intensify existing partisanship and confirmation bias by making desired claims believable and inconvenient facts deniable.
- Visual clues such as blinking and unnatural hands quickly become obsolete.

