Decoding Fake Pain

How easy is it to fake that you are in pain? This question has interested people for a long time, particularly those who have to make decisions about compensation claims related to chronic pain (where large amounts of money are often at stake) and anybody who feels a disconnect between other people’s pain expression and objective evidence for the presence of pain (or better, the lack thereof).

Marian Stewart Bartlett and colleagues tested a computer system devised to differentiate true from fake facial expressions of pain (published in Current Biology, open access). In two experiments, healthy participants were shown short video clips in which faces were shown of individuals who either experienced genuine pain while holding their arm in ice water, or pretended to be in pain. Participants were asked to indicate whether they thought the facial expression was true or fake. In a second experiment, participants were shown pairs of videos of the same person. In one instance the person was faking the expression of pain; in the other video the pain was real. This time, participants were told which of the two videos showed the fake expression of pain, to train them in differentiating between the genuine and the fake expression.

The same videos were also shown to a computer vision system called the Computer Expression Recognition Toolbox (CERT) that can analyse facial expressions from video in real time. How does it do that? In short, this system analyses facial muscle movements that are characteristic for the expression of a certain emotion (in this case, pain). Using a pattern recognition approach, the computer “learns” which muscular activity is characteristic for both conditions, and subsequent video sequences are compared against these “templates”.

The study produced two remarkable findings. First, humans are incredibly bad at telling true and fake expressions of pain apart. With an accuracy rate of 51% (that only increased to a meagre 54% with training), their judgements were not better than chance. Second, the computer system outperformed the human participants by far, and categorised 85% of the expressions correctly. Needless to say, this result is all grist to the mill of those who advocate the need for objective markers of pain in clinical settings and courtrooms. Instead of telling their story, patients would simply have to express their pain in front of CERT and we would know if they were faking the pain or not. Really? As much as I share the authors’ enthusiasm for CERT as a research tool and find it fascinating that there are subtle but detectable differences between real and faked pain expression, I fear for those who express their (real) pain in an unusual way.  As in most decoding approaches, idiosyncrasies are likely to be disregarded and individuals with an unusual presentation run the risk of being misclassified. However, such techniques are gaining ground and have recently stirred up fierce discussions in the context of “brain reading” (i.e. the application of pattern recognition approaches to brain imaging data), which might be a great topic for another blog post …

Let us know your thoughts about “automatic detection of deceit” in clinical and/or legal settings in the comment section below.  And just in case you were wondering which movement it is in particular that reveals an expression is faked: it was the all-too-common mouth opening – a finding which led the authors to put forward that “training physicians to specifically attend to mouth-opening dynamics to improve physicians’ ability to differentiate real pain from fake pain” might be the way forward. In times where a large number of patients complain that their doctors rarely look at them during consultations, focusing on the patients’ mouths seems an odd starting point to remedy this shortcoming.

About Katja Wiech

Katja WiechKatja Wiech is Associate Professor at the Pain & Mind, Nuffield Dept. of Clinical Neurosciences & FMRIB Centre, John Radcliffe Hospital, University of Oxford.  She is also Section Editor here at BiM!


Marian Stewart Bartlett, Gwen C. Littlewort, Mark G. Frank, and Kang Lee (2014) Automatic decoding of facial movements reveals deceptive pain expressionsCurr Biol.; 24(7): 738–743. Published online 2014 Mar 20. doi:  10.1016/j.cub.2014.02.009 Open Access!


  1. Seamus Barker says

    Great article. There’s a very interesting book called, “Pain: A Political History” by Keith Wailoo. One of the key points Wailoo makes is the way that pain has been centrally constructed in the U.S. in the 20th century around political battles over the status of pain and disability claims. In seeking to reduce spending, there have historically been attempts to find ‘objective’ signs of pain, and to invalidate the legitimacy of subjective symptoms. Tests such as that described above certainly play into the culture of seeking to distinguish ‘fake’ from ‘real’ pain. Previously, I’ve seen this very dubiously attempted using MRI scans of lower backs as a putative gold standard in Independent Medical Panel hearings. The potential for the abuse of the above technology is troubling, but, equally, perhaps brain-imaging of pain states could provide the validation so many sufferers need (not just emotionally but legally!).

    Gerry Daly Reply:

    Could it be, could it possibly be, that when operators hit a blank wall when trying to identify a physiological source for any pain event described in a patient’s pain narrative, that the option for assuming a ‘fake’ narrative offers up an easier route to ‘closing the file’ and relinquishing any lingering sense of failure to treat as presented ? There are many medical conditions, particularly neurological conditions, which behave in ways where it is difficult to define a definite source. That can encourage any operator to seek an ‘outclause’ , rather than get bogged down in a puzzle of possible sources, where mistakes are easily made. By transferring the ‘onus of integrity’ back onto the patient, in such encounters, by means of questioning the veracity of the presented narrative (even, if not spoken), the operator may well free themselves to move on to other clients…..but they may also have dented the confidence and self-trust of the patient in the process.
    ‘Bedside manner’ has always been seen as an extremely useful tool in the operators tool-box….and ‘bedside manner’ usually works best with, at least, a pretence of acceptance of presented narrative. That most encounters are in the consulting room, rather than at the bedside, doesn’t negate the advantages of credibility…both ways !
    Seems to me that a lot of time and resources could be wasted trying to identify a minority demographic of ‘fake’ presentations, whilst, at the same time, arousing a mistrust on both sides. Dissonance Plus beckons !

  2. The question here should really be, not whether an expression of pain is fake or not, but whether there is any substance behind a ‘fake’ expression of pain. A patient might well attempt to imitate expressively the pain they were experiencing previously, and expect to again, in order to ensure that relevant treatments are explored. That can seem like an exaggeration at a time when the patient is not actually in pain, but, at the same time, might be the only clue to an underlying ‘hidden’ condition. Would it be right for a patient to say…’I’m not in pain at this moment, so I can’t describe what I’ve experienced over the last few weeks’. Of course not. And ‘imitated’ or ‘fake’ facial expressions are a useful means of communicating previous pain experiences. Probably best to first assume veracity, before considering any manipulation of trust…..isn’t that what the Hypocratic Oath suggests ?

  3. Lesley Singer says

    I don’t see the use for this type of technology for telling us if people are in pain. Imagine you are going to tell a person truly experiencing pain that they are not really because some machine told them. What is the real utility of something like this when it cannot be totally accurate. People in pain already are up against people who don’t believe them and it creates more psychosocial barriers . This would add to them.

  4. Thank you Katja for a very interesting piece. I find the whole idea of using pattern recognition approaches to identify pain and pain expression quite intriguing but at the same time I am left with a lot of doubts and questions.

    You already mentioned the difficulty of averaging across individual differences in pain expression and the problems it causes for individuals who have “uncommon” ways to express their pain. Similarly, I feel that the distinction between “real” and “faked” pain expression is grossly oversimplified. Sure, in a lab setting one can easily create sets of true and false pain expressions but I wonder how these actually model reality. We know already that pain expression, especially in the face, bears no 1:1 relationship with actual pain experience (if there is such a thing at all). And even though this is not stated explicitly, the practical implications imply that this tool might be used to infer actual pain experience, which seems a questionable inference to me. Most of all, pain expression is (also) a tool for communication and thereby influenced by numerous social and psychological factors. Faking might be one example but exaggeration (conscious or otherwise) or inhibition of painful facial expression is just as, if not even more, common and should by now means be confused with faking. A simple example: If I were pleading my case in front of a court and were being tested for authenticity, I would very likely be tempted to exaggerate my pain expression even though I might feel actual pain. I wonder in which category I would fall based on pattern recognition.

    I like the parallel you mention with regard to “brain reading” of pain itself, which is subject to the same criticism. While I can see the gain of this methodology in some contexts, overall I feel it could easily be abused and misinterpreted by reducing something inherently complex such as pain experience and expression to a black vs. white decision process. I feel there is a big risk of ignoring tons and tons of research by reducing pain expression to a dichotomous decision and especially, by making the inference to “real” pain. I guess the literature on lie detection paradigms might offer a good cautionary tale in this context.

    So all in all, while I can see the fascination and utility of these developments, I think caution is justified in limiting them to the scientific domain for the time being.

  5. Ali Khatibi says

    I believe learning to fake pain is something we do throughout our daily life. This will keep us away from conditions which can be potentially uncomfortable for us. So, I would see the first conclusion in another way: we are poor in detection of fake pain because we are good in faking it and we learned how to do that to not to get caught! (it can be tested by having people confronted with computerized detection system and train them to mislead that system and then asking them to show pain and test their expression by machine!).
    A second point came to my mind after reading the article can also be linked to the first point: most of studies on computerized emotion classification use the same or a similar expression database to test the classifier after training! It is necessary to test them in new setting and with different stimuli!