MediaBerkman

Radio Berkman 148: Lies, Damned Lies, and Technology

April 8th, 2010

In an age when every conversation, email, and tweet could be digitally archived, how honest we are – or how deceptive – is open for scrutiny. But there is still a lot we don’t know about the nature of deception.

How can we tell if someone is telling the truth? Are there verbal cues, in addition to the sweaty palms and rapid heartbeat? Is there a difference between lies, or is every lie the same? And how does the medium of conversation – an email, a text message, a phone call – affect the type of lie we might tell?

This week on the podcast, Judith Donath interviews Jeff Hancock, of the Social Media Lab at Cornell University, on how we lie, and the role technology plays in the evolution of the lie.

Listen:
or download
…also in Ogg!

Reference Section:
More on Cornell’s Social Media Lab
Watch Jeff’s April 5, 2010 talk: Technology and Deception

CC Music this week:
Neurowaxx: Pop Circus
Robert Rich: Cowell Piano

Photo courtesy of Flickr user JaeYong

Subscribe to Radio Berkman

FULL TRANSCRIPT
(Some errors)
Lies, Damn Lies, and Technology

Truth seems like a straightforward enough concept. But try to teach a machine what the truth looks like.

It turns out that the difference between a lie and the truth is a lot more subtle than we might be conscious of. We think we can tell when someone is lying to us in person – there’s plenty of non-verbal cues there. Sweaty palms, a twitchy eyebrow, fidgeting –

But with computers – all we can really rely on are the words coming out of our mouths and off our fingers. And as far as a computer knows, every word it sees represents an immutable fact.

Turns out that’s not really true after all. Not only are there plenty of verbal cues that indicate a deception. The very medium we use to communicate and they type of message we are communicating can signal a lie.

You are more likely to lie about your height, weight, and age on a dating site – that might sound obvious. But you’re also more likely to fib about job qualifications on a paper resume than on a professional social network like LinkedIn. And that text message you send saying “you’re already on your way” – when you’re really just getting out bed – happens a lot less over email.

Today’s guest, Jeff Hancock studies deception and social interaction with the Social Media Lab at Cornell University. He spoke with Judith Donath about how people lie, and whether deception is evolving with technology.

===

JEFF HANCOCK: a lot of deception detection has focused on non-verbal cues. so the way in which people move their facial expressions the last forty years have really been focused on that And I think there’s also this assumption of what we say we control that we’re conscious of but the last say ten years of psycholinguistics have shown that actually most of it on automatic pilot we get to choose and are aware of very few of the specific words we end up using. And so when we started looking at the language of deceptions along with JD Pennebaker and Judy Bragoon(?) and there’s others in the deception space like Johnson and Ray they did this thing called Reality monitoring so whether you actually experience something versus whether you just imagine you did it changes the way you you talk about that experience. And sure enough we find these sort of manifestations of deception in the language we use. In the actual words that we choose and a lot of times the best action so to speak the biggest differences are in the little words. So these are called function words – it’s things like prepositions and articles and pronouns and these are particularly unconscious – we have no control over those in terms of hearing them or producing them so I have no idea how many times I’ve said I to you and you would have no idea, and although they’re small they make up a huge amount of language that we use about fifty five or so percent of all words are these function words, even though there’s only about five hundred of them total.

JUDITH DONATH: it seems in the context of lying though that prepositions may be one thing but that I is a very important word and I think you said that you are seeing fewer personal pronouns when people are lying. is there any understanding of why that is?

JEFF HANCOCK: Yeah so there’s this notion of psychological distancing. It’s a very old idea actually, from this guy Morabian that talked about how when we want to get away from something we’re saying or even somebody else we talk differently and so reducing self referencing is one indication of that as is passive voice and the first person singular – if you’re feeling uncomfortable Iie and you don’t feel good about that and you distance yourself from that and I think that this I or me is the biggest indication and you know it’s amazing. we see the first person singular thing across almost every context. so online dating you’re sitting there and actually have time to construct the sentence you get to think about it you’re not you know in real time conversation but you see it in real time conversation you see it in online chat. Pennebaker has seen it in perjury testimony. We’re now seeing it with politics so politicians at least the Bush administration for this particular false statement that our that Iraq had weapons of mass destruction that they said fewer I’s when making those claims. And like a lot about what Bush did is that they’re politicians so they’re already living in this world where everything they say and do is recorded so even though they know they’re being recorded all the time they weren’t either aware or able to sort of control the language that they’re using when making false claims.

JUDITH DONATH: its possible they’re not even aware that they [CROSSTALK] But that the use of I is so important because it lets you still feel you’re making a factual statement but you’re not taking responsibility for the statement.

JEFF HANCOCK: That’s right. Most people have no clue that it’s indicative of a deception and some researchers that do deception detection like for the government and that are trying to catch truly bad guys, this is a big dilemma. I know some that I respect a lot that are deciding do I publish this, where I will you know inhibit the ability to use this cue to detect deception or do I you know not and keep it secret which is problematic as an academic. but you know then the idea is are there countermeasures. I’m of two minds on one hand I just don’t think even a good actor I’m guessing, I don’t think you can control these function words to the degree that you can say pass something like this. And then there’s another issue like once you start doing this over control it’s not going to look like normal discourse either, so I think as soon as you start controlling you’re discourse because you’re lying or because you’re trying to hid lying knowing what the cues are I think it will look different. so it’s a very interesting space I mean now that we have computer scientists that can use all these tools – natural language processing, I mean they can go through texts produced in in almost any domain and have a ninety percent accuracy of determining this is an I, versus a first person singular pronoun versus this is an emotion word, and so you have these really amazing ability to tag and categorize language and they’ve also been doing all this work on statistical classifications where you have neural networks and support vector machines and decision trees that are really good at looking at hundreds and hundreds of cues simultaneously and trying to decide something does this message look more like a deceptive type message this category, or is it more like a truthful one. So I like other people point out that deception detection is probably impossible like perfect deception detection is probably impossible but at the same time I really I’m intrigued by this notion that everything that we’re going to be saying will be recorded and that we’ll have these very powerful algorithms and tools to analyze the language. so is it possible? no I still don’t think it’s possible one hundred percent deception detection I think that the ways that we can lie and the reasons that we lie who we lie and what we lie about are just so numerous that there’s no single way to always be right. So first person singular is a big one. negative emotion terms – we don’t always find this. So negative emotion terms should be related to deception if the deception is big and bad so if the person’s motivated to do it if they caught they’ll be punished if they don’t get caught they’ll get a great reward and when those kind of lies are taking place the negative emotion words tend to leak out, you see more negative emotion words. The exclusive words, that category is surprisingly consistent it’s like the first person singular and this gets to the fact that lies tend to be more simple in their structure whereas truth tends to be more complex and exclusive words like except, but, however, those are related to complexity. Other words that have been popping up are less well understood so motion words, things like going, doing, and relative space words like where something is they are almost always different across deception and truth but they swing like sometimes they’ll be highly, like really different but more when they’re lying and other times they’ll be really really different huge effects but less.

JUDITH DONATH: I mean is this in looking at that the types of ambiguous things, where people are ambiguous about location or are you looking at big lies?

JEFF HANCOCK: The big lies, like the Bush ones, the motion words were much higher and we’re still trying to out whether that’s because of this idea of going to war we need to go there, which have changed would have changed it relative to say a lie about whether I had sex with that woman.

JUDITH DONATH: It’s also so complicated for instance with the Bush one because even if you can say where I know this person said something false and if you can even say I know that they knew it was false, it’s hard to classify it in that domain of did they think they’re doing a good thing. So because if you lied to your spouse you’re pretty sure that if you get caught you’re in a lot of trouble and you did something you really shouldn’t do, it’s a tremendously guilty guilt inducing situation. Whereas the going to war issues are very complicated because you don’t know if the person saying them was in all honesty thinking they were doing the right thing that this was really important to do it was real important that people be motivated to do it and was the only way to get people to do the right thing it is very different than if they’re sitting there thinking it’s really horrible, its a bad thing for the country but I get really rich if we go to war which might have a very different underlying emotional state so-

JEFF HANCOCK: and we see this I mean when we look at the emotional work, we’ve been breaking the cues down into a cognitive related ones and emotion related ones. And so for instance online dating we actually saw the negative emotion words flipped from what we expect. We usually expect more negative emotion words when people are lying. Online dating when they were lying about what they physically looked like how old they were those sort of things they actually used fewer negative emotion words and going to your point I think in that case I think it’s much less about emotion and actually feeling whereas it’s more probably about that presentation they were trying to give, like I’m a positive person even though that they were being deceptive about well you know how tall they were how much they weigh. so I think your point about emotions more generally is true and this gets back to just the weird types of deceptions and there are so many rabbit holes we can go down because this is so much things to lie about and how they would make you feel.

JUDITH DONATH: The questions about personal appearance are important, I think a lot of times when people think about lying and on line dating I think it’s about people who say I’m not married and they are. Or they have a whole family. Those are some of the big really big ones that are out there in the you’re going to be in trouble, there’s something very guilt inducing here and some of these others especially the weight ones it’s hard to tell how much of it is self deception and the sense that well I might really weigh two hundred pounds now but I know that in a week it’ll only be a hundred and seventy because I’m going to be really good.

JEFF HANCOCK: we were really worried about that but I feel pretty good that people know pretty well where they’re at with weight and you know there’s a couple reasons. I mean they tended to err on the side that is most advantageous for them. So it didn’t seem random. And also when we ask people you know how deceptive were you it matched, like they were aware I mean one of our biggest concerns always is we ask people were you lying or not, you know in the lab or are all these places, who knows whether they actually are telling the truth. in this case we documented here’s what the discrepancy was and when we asked them separately you know how accurate is this it matched up pretty nicely so I think actually we know pretty well and even the theory behind self deception what it takes for an organism to be self deceptive is astonishing. And so I think as we dig into that we’ll start getting more insight into the ways language can change depending on if it’s self deception versus whether it’s this intentional deception that we can talk about. we’ve been talking about lies and differences between them as if when you lie everything you say is a lie. In fact when you look at a lie most of what the person says is quite truthful and I think we’ve started to tie this to your idea of signaling can be related to deception. As you indicate with a number of different signals and perhaps ones that are more reliable that makes the statement appear true but then there’s the statement within it that is deceptive and in fact if you start thinking about it it’s actually hard to tell a lie without saying anything truthful. I mean having to make up everything, so usually it’s like this happened and this happened and then the lie then this happens and so we’ve started to do is think about our work is actually naturally confounded. We’ve been looking at say the Bush statements that were false versus the Bush statements that were not false or we’ve been looking at online dating profiles that were truthful versus weren’t, or that lie or tell the truth. in fact when there’s a lie there’s two types of language there. There’s the statements that are deceptive and there’s the truth around them that are serving the deception. So if I were to ask you to lie to somebody about the most important person in their life there would be two types of words you would use ones that were that actual lie and the others that were sort of supporting it surrounding it. And so we’ve started to look at whether those are different and they really are. The I thing for example that is all carried out in the deceptive statements, and the words around them tend to have a lot of I’s. So they actually make up for almost around. So we’re seeing some really cool differences in within a lie statement there’s these two types languages, and I think all of our work before has been sort of like naturally confounded across those two. And then the other thing that we’ve been working on is that lies don’t take place all at once, they tend to be told through time and at least that’s how conversation works so when we’re working with instant messaging it was really clear that you know there’s this sequence of statements, messages basically and our question was do lies, do they tend to clump together like in a burst? Or are they leaked out like here’s a deceptive message and then there’s a bunch of honest ones then there’s another deceptive message. And if you were thinking sort of cynically that probably the best way to lie is to spread them out and hide them in the truth, but cognitively that’s probably difficult to do. So anyway we had these two ideas, these two hypotheses – Bursty where they would all occur in clumps, or spread out and we just looked at people instant messaging in regular life and then record when they lie and we looked at it as a sequence and sure enough it’s very bursty. So you tell a lie and then you have to tell several statements around that and then it’s you know truthful for a long time and then boom another bunch of other lies.

JUDITH DONATH: Well thank you very much, this was fascinating.

===

Jeff Hancock is an associate professor at the Department of Communication at Cornell University, where he helps run the Social Media Lab. You can find out more about him, and hear a talk he gave earlier this week where he goes into a lot more detail on this very topic – it’s well worth a watch – at our site blogs.law.harvard.edu/mediaberkman.

This week’s episode of Radio Berkman was produced by me Daniel Dennis Jones, with Judith Donath and the Law Lab at the Berkman Center for Internet and Society at Harvard University in Cambridge, Massachusetts.

Be Sociable, Share!

• 
• 

Entry Filed under: audio,Berkman Center,radioberkman