I’m auditing a University of Columbia course on data science at the moment. This lecture is on detecting truthfulness in human speakers. The researchers created a corpus of speech, some of which was truthful and some of which was not, and analysed various features to see how easy it would be to distinguish.\u00a0<\/p>\n\n\n\n
<\/p>\n\n\n\n
They also had human judges score the speech for truthfulness, and found:<\/p>\n\n\n\n
<\/p>\n\n\n\n
– people performed, on average, worse than chance<\/p>\n\n\n\n
– some people, however, performed much better than others<\/p>\n\n\n\n
– training did not improve performance<\/p>\n\n\n\n
– neuroticism in the scorer reduced performance<\/p>\n\n\n\n
– agreeableness and openness in the scorer increased performance\u00a0<\/p>\n\n\n\n
– post-test confidence did not correlate with ability.<\/p>\n\n\n\n
<\/p>\n\n\n\n
In a meta-study, criminals were found to have the best ability to distinguish between truth and lies, and parole officers the worst (less than 50%, possibly because they expect people to lie more than they actually do).\u00a0<\/p>\n\n\n