Key claims
- Deepfake capabilities: AI can create convincing synthetic video, audio, and images of both real and entirely fictional people. The technology has advanced dramatically in just five years, enabling even untrained users to create professional-quality deepfakes via smartphone apps.
- Scale of visual media: Humanity consumed 1.4 trillion photos in 2020 alone. By 2030, video will account for 80% of internet traffic, and 2/3 of the global population will be creating and sharing video content.
- Projected growth: Experts estimate up to 90% of video content will be synthetically generated by 2030. Individual creators will soon produce content quality rivaling Hollywood studios.
- The liar's dividend: As deepfake technology becomes ubiquitous, the mere existence of convincing fakes enables bad actors to dismiss all authentic media — including evidence of real crimes — as synthetic hoaxes. Example: George Floyd's death video was claimed to be a deepfake by political candidates.
- Processing fluency bias: Humans are cognitively wired to believe media that looks or sounds authentic, regardless of its actual origin — making synthetic media particularly persuasive.
- Historical precedent: Media manipulation as a tool of power is not new (Stalin's photo-editing of political enemies), but modern technology makes it exponentially cheaper, faster, and more accessible.
Notes
Schick frames deepfakes not merely as a technical novelty but as a civilizational threshold: a moment when visual media — humanity's primary communication channel — becomes systematically unreliable. The talk is especially prescient on the liar's dividend: the asymmetry where authentic evidence can be preemptively delegitimized by invoking the possibility of deepfakes. This is a governance and epistemology problem, not just a detection problem. The George Floyd example is instructive — a real, documented crime, dismissed via deepfake accusation — illustrating the societal stakes beyond entertainment or fraud.
Schick's background as a political analyst (rather than technologist) gives the talk a policy-aware framing often absent from technical deepfake discussions. Worth comparing with Suwajanakorn's TED talk (detection-focused) and Graham's TED talk (artistic potential).