MIT study: AI-edited photos and videos plant over twice as many false memories
Justgototheeffinmoon · reddit · 2026-11-18
A preregistered MIT Media Lab study (presented at CHI 2025, covered by the Washington Post) with 200 participants found that AI videos generated from edited photographs produced just over twice as many false memories as unaltered images.
- The experiment started from 24 ordinary photos (a wedding, a street, a race); groups saw originals, AI-altered versions (changed expressions, inserted military personnel, added climate-change signs), or short AI videos generated from the edited stills.
- The video group produced the most false recollections and was also the most confident in its wrong answers.
- Memory researcher Elizabeth Loftus notes memory isn't a recording device; Pattie Maes calls the implications "very dangerous."
More from Research
- In LLM quant research, eval design beats prompt engineering — bgoncalves · 2026-11-18
- Blink CPU speed test: model2vec 3.8x faster solo, Sentence Transformers 2.2x faster at batch 1024 — tomaarsen · 2026-11-18
- blink-8M tops seven static embedding models with 55.35 MTEB and 46.70 BEIR nDCG@10 — tomaarsen · 2026-11-18
- blink-multilingual-64M demo: English query retrieves Dutch and Spanish docs correctly — tomaarsen · 2026-11-18
- MinishLab ships Blink static embedding models: 8M and 64M params, fast CPU retrieval — tomaarsen · 2026-11-18
- Massive proof-dependency graph for Theorem 3.1.1 wows math Twitter as formalization scales up — littmath · 2026-11-18