I tried to prove I'm not AI. My aunt wasn't convinced
Rapid advances in AI-generated audio and video are making it increasingly hard to know whether a call, message, or online clip really comes from a human, raising fears about scams, political manipulation, and a broader collapse of trust in digital communication. Commenters weigh potential defenses — from family “codewords” and shared memories to cryptographic signatures, verified camera hardware, and regulation mandating watermarks — but note that social engineering, device compromise, and user apathy can undermine most technical fixes. Many predict a shift back toward in‑person interactions and tightly bounded trust networks, with significant economic and social costs as remote hiring, online evidence, and everyday internet use become more suspect.
Shared secrets, shibboleths & social defenses
- Many argue families should share offline “codewords” or shibboleths to authenticate unusual calls (e.g., emergencies, ransom scams).
- Others prefer shared private memories instead of new passphrases, doubting people will reliably remember codes.
- Several note scammers create urgency and emotional pressure so victims ignore or can’t recall codewords.
- One concern: once a phrase is used over a network, it can be captured in breaches; the secret degrades over time.
- Some already use shibboleth/duress words with alarm companies or joke about spy‑style countersigns.
Cryptography, cameras & technical fixes
- Strong sentiment that society failed to adopt widespread cryptographic signatures early enough (email, VOIP).
- Proposals: signed emails, signed media, tamper‑proof or authenticated cameras, and provenance chains from capture device to viewer (possibly blockchain‑backed).
- Objections:
- Device or key compromise and social engineering remain weak points.
- People may let their own AI agents send signed messages.
- Editing photos/video breaks simple hash‑based verification and professional workflows rarely use straight‑from‑camera output.
- Centralized signing by major platforms could further concentrate power.
Trust, scams & economic / social impact
- Multiple anecdotes of hijacked email accounts and AI‑crafted, highly personalized scams.
- Call‑center and “grandparent” scams now potentially enhanced with AI voice and video.
- Some foresee shrinking trust spheres: only in‑person or local, verified relationships are considered reliable.
- Predicted impacts include more in‑person interviews and travel, difficulty trusting online information, and heavier economic and psychological costs.
- Others say misinformation, doctored media, and spam are old problems; AI mainly lowers cost and scales them.
Detection limits & human psychology
- Many feel “spotting AI” visually is already unreliable; context and provenance matter more than artifacts like extra fingers.
- Discussion of phenomena like analysis paralysis, flip‑flopping under doubt, and how manufactured urgency suppresses skepticism.
- Some suggest “reverse captchas” using taboo or disallowed topics to distinguish humans from safety‑constrained corporate models, though this fails for uncensored/local models.
Regulation, watermarking & policy
- Suggestions include legal requirements for AI watermarking and platform‑level labeling of synthetic media.
- Critics argue bad actors will ignore rules, watermarking is technically fragile, and laws may only burden compliant companies.
Cultural, political & philosophical angles
- Worries that deepfakes erode courtroom evidence and public accountability (e.g., plausible deniability for incriminating footage).
- Some see AI as accelerating a “post‑truth” or “spam‑saturated” world; others push for “touching grass” and prioritizing offline community.
- A side debate questions whether human–human interaction is intrinsically more valuable than interaction with convincing AI facsimiles.