Figure 1. illustrates the results of an LSTM that's sexually attracted to tall muscular guys, accusing everyone else of being deceptive. > Deceivers put their

Figure 1. illustrates the results of an LSTM that's sexually attracted to tall muscular guys, accusing everyone else of being deceptive. > Deceivers put their hands in their pockets and look around more than the participants in the natural condition. Doesn't it depend on _context_? If I recall correctly, one of the main giveaways that lead to the identification of the Boston bombers was precisely their composure during the whole situation, which investigators deemed "off" in that context and made them persons of interest. > Each participant was randomly assigned to walk either naturally or in a deceptive manner, as described below. So it's like training a "terrorist detection" system by asking your buddies to act "terroristy"? The experiments trains on people displaying behaviors they think convey deception, while deception really tries to convey deceitful behavior. A better experiment would have been not telling the subjects about the nature of the experiment. Then create a situation where there's a real incentive for the subject to be deceitful. An example would be an experimenter explaining there's a payout for the experiment. The payout must be either monetary or something related. Like saying the study will be at some exotic location, all accomodations taken care of, for a month, and you can bring your significant other. Then stopping on one answer in particular that disqualifies the subject from the pool of people eligible to that payout. And then explaining that the experimenter will let this slide but their colleague will probably not; they then instruct the subject and coach them on what they need to avoid saying on several points and before they're done explaining, the "colleague" enters the room and starts going through the checklist. Now you have incentive to be deceitful and a slightly less crappy experiment than asking people to be deceitful.