Human Emotion Recognition With Technology
Clearly anything that can read emotions could be applied for both human good, but also for nefarious purposes. Our work is to maximise the good, and help protect against the bad.
Graeme CoxCEO, Co-Founder emteq labs
My Thoughts on the Conversation
Our emotions are constantly on display, whether we realise it or not. As humans, we interpret each other’s emotional states with every interpersonal interaction - consciously or subconsciously. Giving computers access to this level of personal and intimate data has been a prickly conversation in the past few years, especially with the rise of surveillance and facial recognition technologies.
However, there is a positive side to technology that could enable automated human emotion recognition.
In this talk with Graeme we uncover how the combination of sophisticated sensor technology and AI can be used within immersive environments, such as virtual reality, to create and deliver personalised therapies, helping people overcome anxieties, phobias or to deal with difficulty, dangerous scenarios, safely and more effectively.
Related article: Find out more about emteq's emotion recognition technology in this previous article.
Graeme mentioned that globally, 7 out of 10 hospitalisations are a result of preventable habits or behaviours, and being able to reduce this by just a small amount could have very significant positive societal, health and economic impacts.
As such, a big part of Graeme’s work at emteq is working with research and health institutions to create more effective treatments for people suffering from behavioural disorders, mental health conditions, fears, and anxieties - something that has sky-rocketed in the last couple of years.
As I see it, the combination of emteq’s emotion recognition technology with virtual reality therapies could be used to provide clinically controlled therapies at home, without the complications of asking anxious, confused patients to attend regular clinics in strange or unfamiliar settings. As Graeme explains in the video, this could not only reduce the overheads of already thinly-stretched health practitioners, but could also increase the completion rates (and therefore the success rates) of therapeutical programmes and reduce costs.
Beyond the therapeutical applications of emteq’s technology, there are huge potentials for education and training, too. Understanding an individual’s emotional response to a given scenario holds the promise of creating immersive tools that adapt to how the participant is reacting, making the experience more challenging, more stimulating, or providing additional support.
There are many studies that look to the power that immersive training and therapies can deliver (some are referenced in emteq’s Emotional and Psychological Well-Being white paper), but today’s technology lacks the emotional feedback loop. Implementing emotional feedback in virtual environments could create experiences that are uniquely tailored to the participant’s needs, thereby creating even more rewarding and effective programmes, that are truly unique for every user's specific need.
But with these intimately personal insights, there are also risks.
Graeme mentioned a Stanford VR study that was able to successfully identify a participants sexual orientation just by monitoring eye movements within a VR experience.
The level of detail emteq are able to extract is significantly deeper than just eye-tracking, so the potential for wrong-doing is much greater. Whilst we did touch on this and emteq's work with XRSI.org a little during the video, I spoke with Graeme afterwards and he pointed me towards a white paper that accompanied a roundtable they took part in on this topic. You can find the joint paper here.
About the transcript:
This video transcript is automatically generated by YouTube, so grammatical and typographic errors are often likely. If in doubt, please consult the original video above.
Interviewing Graeme Cox from emteq labs about Technology to understand human emotions and emotion recognition.
uh everything from skin flush to facial expression to the posture we read these subconsciously and understand the uh the emotional response of the of the person we're interacting with uh we're looking at ways that we can objectively measure these and turn these back into medical grade biomarkers of literally how we feel i am very excited today to be joined by graeme cox from emteq labs graeme it's very nice to have you on today uh please do tell me a bit more about yourself introduce yourself and what you do at m tech thanks scott it's great to be here um so uh as you say i'm the ceo and co-founder of the company emteq labs uh personally i'm uh um my background is in artificial intelligence my degrees in ai from back in the 1990s and i've been working with big data machine learning and deep tech for my entire career both in large consultancies and um in several of my own startups along the way um like the first startup i was involved with was actually uh whilst i was still in uh in the university and that's been a kind of a feature of my my life ever since um i built a cyber security startup um around the turn of the century and ran that through until i sold it to uh to dell which now forms a core part of dell secure their site the global cyber security division um provides some of the core monitoring expertise and algorithms into their solutions and um yes the whole area of uh of of of human performance and behavior and uh effective computing has been a has been a kind of major thread in my life uh from from the the start of my career through to now um i uh either working as a consultant in understanding how to improve processes uh people and technology in in situ or in uh trying to build and operate teams of people in the companies that i've created myself and indeed in my uh in my sport i've been personally doing triathlon for about 20 uh 20 years now and i've been a triathlon coach for uh nine years and uh in the whole process of improving performance changing behaviors for the positive et cetera has been a it's kind of like that central thread of my interest in uh in both work and personal investigation and so m-tec comes around really as i think the culmination of all of that um a fortuitous meeting with a a brilliant chap called charles veduca who is a facial surgeon who i met in the school playground as you do about six seven years ago down in brighton where i now live led me to realize that we were talking about new ways of reading expressivity on the human face and how important that is to the fundamental understanding of human behavior and therefore how you can modify it um what we like and what we dislike fundamentally drives our behavior so that gut reaction we have to whatever it is the things that we do the the the food that we eat the things that we experience this drives how we're going to operate in the future and that's fundamentally important to understanding the human condition and how to um how to both improve people's lives and welfare but how to actually also get the best out of them as well on a purely health basis 70 7 out of 10 of the leading causes of avoidable death are driven fundamentally by our behaviors um and though by those behaviors i mean things like what we eat and the drugs that we take into our bodies whether we smoke or not whether we do exercise or not the choices that we make the little ones day-to-day that lead up to though to your to your long-term health and determine whether you are likely to die of stroke or heart attack or emphysema or drug abuse or you know whatever it is it's the vast overwhelming uh load on the nhs and the burden on human health is fundamentally driven by the choices that we make as individuals and by understanding better how to nudge behaviors in the right direction to help people help themselves to help them be the best version of themselves is the basic mission that we set out with emteq to achieve wow okay that's fantastic um you touched on health there and the choices we make and actually there's been a couple of interviews i've done recently about metabolic health and how we can be better at making decisions about how we manage that and the impact of that on our long-term health as well um so maybe we'll come on to that in a minute because i'd love to get your kind of professional and your triathlete opinion on that um one of them a company called levels is very much focused in in your kind of persona the high performing athlete so we'll talk about that um but tell me more about what is um what is the technology that you do at m tech and how does that how would people see that yes absolutely so uh and obviously i've given you a very high level view and it's like well what is it what is it we're actually doing so fundamentally the core technology that we have been uh developing which is which has a lot of novelty around it and has led us to creating a lot of new intellectual property along the way is in uh is in improved ways of reading the physical signals of human emotional response we like to do that with sensors primarily rather than with cameras uh and we are reading things like the electrical muscle activity inside the face to understand the tiny um reactions that we have the the micro expression reactions that we have that indicate our actual gut response to whatever the stimulus is we're being faced with um features of uh changes in our heart rate and the variability of that heart rate and the heart rate itself and the subtle features within it that indicate um the level of excitement or arousal that we have when we're faced with whatever stimulus is uh and uh and and movement the body movement of our particularly of our upper body and head that indicates things like whether we're attracted into something or whether we're repelled from it and understanding those different features the the kind of things that a somebody with a high emotional iq subconsciously reads and interprets extremely well and at the other end of the scale somebody with a degree of autism reads exceptionally badly this is these cues have been developed so a digital biomarker that indicates emotional response that can be used at a clinical level to to to help drive therapies and behavior change programs that sounds very important very important especially in today's in today's world where i think we're all suffering from conditions and anxieties and issues that we probably didn't even know we we could be succumbed to a year ago but in terms of just trying to frame it there's some there's a little chart that i use um all the time it's the united nations sdgs and here it is here's one i printed out earlier on um i can imagine you know good health is a is an obvious um it's an obvious bucket here that you would fit into are there any others in here that you think you know emteq could fit into maybe quality education or reduced inequalities how how do you feel about that um yes or well certainly um the ability to take that next level of human communication and quantify it so to go beyond um speech analysis which is currently state of the art with alexa siri etc and move that through into an emotional understanding of the of that next layer of communication the body language and um subtle vocal and physical cues that we give out that indicate how we really feel about things that does that that allows us to better influence behavior at all levels and i see i see the uh that there's a as a spectrum where at one end at the one end of the spectrum you are dealing with clinical issues so the vast quantity of uh anxiety issues that exist in the world today um in 2018 it was recognized that over 300 million people worldwide had a clinical anxiety problem aside from just the general day-to-day anxieties and issues we deal with that are at a non-medical level that's before coronavirus hit us you know the level of anxiety uh and tension in the population today as we speak you know we can all see it it's uh it's risen massively over the last year and that is a real uh problem of how we deal with that but but we're on a spectrum and as you say in those stg's the spectrum of behavior change goes from dealing with actual clinical issues at one end to improving performance of otherwise normal people at the other end so and those performance improvements include improved educational outcomes by tailoring educational content to understand your audience to actually play to your audience the ability of content delivered to adapt to whether the specific audience you're playing it to is actually enjoying and engaged with your content or not provides a potential future generation of educational delivery in in immersive technologies virtual reality augmented reality etc or 2d content that allows an adaptation and uh personalization of delivery that that we that we don't we're not able to see today so is that where we would find your technology in the immersive world integrated into virtual and augmented reality systems so so our first uh products to market are in as exactly as you say scotland are integrated into virtual reality and that's for a number of reasons so the first is that um a lot of the commercial use of our technology today is in is in research whether that is academic research um clinical research in uh drug and therapy trials or or indeed in commercial research in you know market research around content testing etc and immersive environments like virtual reality provide the ultimate experimental environment so for our clinical work where we're dealing with things like post-traumatic stress disorder it's extraordinarily difficult and expensive to put somebody back into a situation that gave them a trauma in the first place you know you can't you know you wouldn't want to put somebody in a war zone in order to help treat their uh their their war zone related trauma you wouldn't want to put them back in an accident scenario in order to do that immersive technologies allow you to provide access to those traumatic experiences in a safe and controlled manner similarly when you're dealing with um anxieties or phobias so social anxiety for example uh mixing mixing with crowds social interactions etc you can do that in a safe controlled and cheap and effective manner in virtual reality that um that physically doing that with therapists in in live scenarios is is just logistically very hard and ultimately we see ways for people to start doing self-treatment for anxieties phobias uh and traumatic um episodes using home-based systems that allow them to to run through protocols at their own pace in their own time and to uh to gain the benefits without constant oversight from uh from a therapist that is our end goal and similarly again on that spectrum at the other end of that self-paced educational performance improvement uh tools like training people for better public speaking for example training um frontline personnel to be able to deal uh to be more resilient in dealing with emotionally difficult uh environments so how do firefighters cope with a uh with a with an accident scene the moment they arrive being able to train uh in simulations gives huge benefits to both the organizations that are paying for that training and indeed to the individual themselves in the the number of times they can work through that the flexibility they have to access that content uh and uh you know there's there's a huge amount of evidence to show that simulations do evoke those same gut emotional responses as the as the real thing so we're a fan yeah and i've seen some pretty astonishing figures about the efficacy of virtual reality training compared to even face-to-face classroom training um and the numbers are quite you know quite amazing i think it's um 50 reduction in costs four times better learning outcome faster learning the ability to translate all of these um actually to carry over from the virtual world into the real world these training is quite astonishing with your technology does that make it even more effective because you can tune it to exactly my response which may not be the same as yours or the person next door well primarily in the work we're doing today we're we're adding we're adding a layer to that uh training or therapy so rather than providing uh behavioral training for purely um physical process oriented tasks we're able to provide behavioral training to improve uh improve emotional response to reduce your stress to reduce your anxiety to improve your ability to react accurately and appropriately in the moment bringing in that extra dimension but you're right as well that in the future the ability to understand the level of engagement that an individual has with the subject allows the opportunity for the content to automatically adapt to that person so for example if you want to put somebody through a series of more complex and demanding tasks being able to estimate the cognitive load of that individual as they go through these tasks allows the adaptive content to deliver the right level of complexity the right level of stress the right number of distractors to the individual to suit a personalized model for them and that's absolutely a goal that we have with us with our technology so on the therapy side um i'm a mild sufferer of um fear of heights and i think if you get to ask my family they may say it's slightly more mild slightly more than mild um but yeah i know people who are absolutely terrified of going up a ladder i can do a ladder as long as it's not too high but when you get me climbing up castle ruins and things like that then i get a bit edgy now if i was to um want to overcome those fears i understand that that kind of exposure you have to just keep nudging yourself a little bit further my nudge would be completely different to somebody else's nudge who may be better or worse of that so is that the kind of benefits that you you're seeing that ability to personalize that treatment for an individual so they can be more effective at overcoming a fear or a phobia yes absolutely so if you think about it so so we'd call this virtual reality exposure therapy so the principle there is in the same way as
some people with milder allergies might train themselves to deal with a with a mild peanut allergy for example that that slow and gradual exposure to small quantities of the stressor allows your system to build up resistance to that to that stress or over time it's exactly the same it's a desensitization process whether it's fear of heights for flying fear of crowds whatever it might be that by gradually introducing yourself to the thing that causes the anxiety you can learn to overcome and manage that anxiety particularly if you're given some mental tools um to to to to keep yourself calm and to handle it when you find yourself in that scenario the trick however for success is to be given the right level of stimulus so if you over stimulate somebody you know so i suddenly put you standing one foot on the top of the grand canyon from from scratch you freak out your your fear of heights is much worse than it's ever been before there's no way you're ever coming back to that therapy again on the other end of the spectrum if i you know i put you on top of a you know a wooden wooden crate in the simulation you're a half a foot off the ground you're standing there thinking well and what's the point of this so under stimulation and over stimulation both lead to people not completing the protocols not getting the benefits that today um skilled human interaction is required in order to gauge that level of uh exposure so a behavioral therapist cognitive behavioral therapist a psychotherapist etc would lead you through a clinical process in helping you treat that phobia and they would guide saying right i can see that you need a bit more stimulation a bit less etc and today and in the near future our technologies will support and augment that therapist by providing the therapist with more objective information our potential end goal here is that we can take that therapist intelligence into the ai of our system itself and the system can automatically read you and understand the level of stimulus that's appropriate to you so that you can go through self-guided therapy in a safe but effective manner ah i'm glad you came back to ai that's one of my questions all of this sounds like sensory fusion and augmentation again something i studied at university around the same time as you um i was wondering how the ai piece comes into this and what benefit that adds to it absolutely so what we're doing here is developing a a personalized model of uh of of of emotional response which to the best of my knowledge and the in the research that we look at around the world has not been effectively delivered before so um in the field of uh emotion ai as it might get called generally um the world is is is mostly focused on a single approach a unified approach to classifying everybody's emotional response regardless of your age gender nationality racial background etc and whilst things like um the major expressions like a smile for example they are universal features a smile is a smile uh in its basic form wherever you go in the world the way that we use our expressivity the the quantity we express and the uh and the the way that that is subtly delivered it's it's not only different in japan as it is from america but it is also different in men and women it's different in it's different every level it's different between you and me scott you know so you might smile three times more in a day than i do but that doesn't necessarily mean that you're a happier person than me that just might be that your base level of smileyness is set different to mine and and that and that that personalized approach to understanding um individual expressivity and emotional state i believe is absolutely key to success so we are building uh deep learning models that not only provide us with classification across demographics and allow us to understand broad uh sweeps of of of response to a given stimulus but actually also a calibrated learning model for an individual that that improves its understanding of you the more that you have exposure to it and that's a that's it's it's a it's a machine learning approach uh where we've moved we're moving into the deep learning space now which is very exciting as uh for me as an um with my ai background it's wherever we've been looking to get to for the last few years and it's starting to show real uh benefits in uh in delivery in practical terms i haven't really described you know physically what we're dealing with here so so you know to to put him in in the viewers mind the technology we're dealing with here we have vr headsets that have a set of sensors integrated into the facial interface the bit that actually touches the skin around the face and that allows us to read the electrical muscle activity of your face reads heart rate information reads your body movement skin flush etc and uses a combination of local processing inside the device itself and also cloud deep learning processing to interpret sensor data from the human body through into physiological data and from that physiological data into a personalized model of emotional response when we add into that the data stream that comes out of the virtual reality environment which provides us with context we're able to really understand your response to different scenarios so you know an emotional response is only valid if you understand what it is you're responding to you know so so if i just look at you and i don't see what it is that you're smiling at frowning at um shying away from whatever it might be i can't really understand you and your your your your range of emotional responses and the wonder of virtual reality is that you have detailed access to all of the data on what the stimulus was as well as all about biometric data and what the response is that together gives us our machine learning models that drive the company at its heart wow um now that raises a lot of questions in my mind one about are the developers ready to um to take this kind of information and build these personalized environments adaptive environments and then very much linked to that and everything you just said just then there's a lot of very personal and intimate data that's coming out of this right um and we don't live in a world where personal and intimate data is um very well managed so what are your thoughts on on kind of the responsibilities around around managing that data that's a that's a fantastic question scott and something that's that's very close to my heart um with my combination of my my my focus and my passion in in helping people through helping them understand their own behaviors with my background in cyber security as well uh the whole issue of data privacy and indeed data ownership is is very close to my heart so the approach that we uh that we take with our data is that all the data remains owned by either the individual who is self-managing using our our clinical end system or belongs to the researcher if it's a research system who and they are collecting data from a number of participants and although our data flows through our cloud infrastructure we do not retain any of the um user recognizable data features that uh that so that none of the raw data from uh collection ever hits m tech um accessible property so it goes through our uh through our deep learning engine and provides insights in terms of um affect emotional response back to the to the to the recipient but uh we only retain the higher level insights that are generated from that so the uh the ability to improve our personalized models is of course fundamental to the success of the company but the data itself remains owned by the people who collect it and that's a that's a that's a key um fundamental requirement for us we're actually members of the um uh the i'm gonna get this wrong the x x rsi which is a um a group fundamentally built for maintaining and developing the security and privacy of data collected using immersive environments because as you say these technologies provide a huge amount of personal information and even even in all an ordinary vr headset if you take out the biometrics so it's just a standard vr headset the quantity of personal information that could be captured by a developer is huge to the extent of things like um you know understanding gender and age so demographic information gleaned from how people consume vr content and sexual preferences educational background these things can be inferred from vr scenarios quite effectively and there's been some interesting research done on that so um who gets to own your data is fundamentally important i completely agree that's that's particularly eye-opening i think some of those things um inferring sexual orientation from what you look at i guess within a virtual environment absolutely well i mean this is a very crude example but you go into vr i present you with a woman in a bikini and the the way the the way your eyes linger and for how long gives a pretty strong indication of your um of your sexual preferences and uh and some information about your background instantly and there's been some solid research uh on exactly that interesting experimental setup but there you go
you know and i guess as virtual reality in these immersive technologies becomes more and more available and more and more pervasive in our lives this is a question that's not going to get any smaller is it it's it's going to grow totally i mean i i it does amaze me actually how accepting we are of our social media and the quantity of data that is being collected there i mean my understanding of the um of the algorithms that go behind tick tock for example to determine uh what the video feed is that you're given uh there's some there's some startlingly um insightful algorithms going into every movement that you make on screen how long you watch every every video for where your fingers hits on the screen to scroll um every part of it is analyzed in order to tailor that stream to you and and it's incredibly effective but you know it's if you uh uh if you know any tick tock users they get hooked pretty quickly yeah i have heard my oldest also keeps telling me exactly yes that is the demographic yeah so um i think lastly to touch on some of the challenges that many of us have faced in the last 12 months um and a broader conversation really um as we find ourselves many of us find ourselves working from home a lot more than we ever thought we would if we ever thought we did um there's a lot of challenges that we have to deal with there um isolation anxiety um separation anxiety from your team all of these big things that are that are kind of building up in in us um and then there's the role of the employer so i in some of these conversations i've seen we've had a lot of conversation about how far does that role of responsibility that moral and even legal responsibility of employer stretch into a working from home environment um so be good to get your idea of what how that landscape lies and then the role of the kind of technologies we're talking about um today does that fit in in that picture at all okay yeah really interesting stuff so my first thing my my personal philosophy on building and managing teams is that um empowerment and self-determination is the way forwards now i say that um in the full knowledge that the teams that i work with are educated to the highest possible level they're all career individuals they're all highly paid and therefore you expect them all to be highly motivated and individual uh very individualistic in their approach anyway but nevertheless for within that context i'm a very strong believer in personal determination about flexibility in working hours approach etc and as such i'm massively against employee monitoring so that to me that is uh that is anathema as a business owner and uh and a manager myself in terms of our technology and how that how it could be applied because clearly anything that can read emotions could be applied to both uh for both human good and also for nefarious purposes as well um i focus specifically on sensors rather than cameras because they the depth of information that we can find is so much greater which allows us to build our personalized models at a level that we just couldn't do i believe using features from um from from video footage however there's a second kind of um network effect benefit of of using sensors which is that there is a very strong sense of um personal acceptance in using our technology in the first place the fact that you physically have to put something on your face whether that is our virtual reality headset or the or the glasses that we have in prototype for real world estimation of emotional response means that you are physically accepting a level of computer monitoring when you put them on and hopefully doing it because you are you accept the benefit that comes with doing that you that you you put your donning your personal assistant who is working purely for your benefit to understand your responses and to give you cues that will help improve you it's very different from an employer deploying software to your work computer with the camera here and using that camera data to understand whether you're paying attention to your screen whether you're actually working whether you're looking grumpy or so cameras are intrusive fundamentally that's part of the reason why google glass failed uh when when it came around it wasn't that you know the glasshole approach it wasn't that the glasses looked weird i believe you know in silicon valley that's not such a big deal at all it's the fact that there was a camera pointing at the person you were talking to and that's the level of social intrusion that even in in the 21st century i don't think we're able to uh to accept um so um sensor-based technology focused on the wearer is actually much less intrusive because there involves a there's a there's a level of personal responsibility and acceptance in donning it and using it in the first place and i believe that's why we're actually really well suited to succeed in this field where a lot of camera technologies may fail because cameras are do not come with personal acceptance anyone can point one at you anytime and and they don't have that that ability to deliver personalized information it's much more generic um i've got a list of other things we could talk about but in the interest of time um let's kind of wrap it up and uh if any of the uh viewers are interested in finding out more about what you're doing at m tech or even xrsi the work that you're contributing to there where should they go what they what can they find um well please come and find us on our website first of all emteqlabs.com um anybody who is interested in contacting me please do email me graham emteqlabs.com i'd love to start a conversation with anybody's interested in finding out more about our technology or we can um we can help in any way um the xrsi is indeed xrsi.org and my co-founder charles sits as sits as the medical representative on that on that committee as i contribute to from my cyber security background um and um yes you know please start the dialogue it'd be lovely to talk brilliant well you can give me some links and i'll put them down here like magic videos they should be down here for you now um and um yeah so it just it's just um thank you so much for coming on and uh sharing your thoughts and ideas um i i found it fascinating and like i said i hope we can do another one later on to talk about the rest of these the rest of these things i'd love to scott brilliant brilliant um for the viewers please do reach out to graham using the links below um do like and subscribe to the youtube channel and there's a lot of other very interesting interviews coming online soon so once again thank you so much for your time today it's been one thank you very much
Keep up with the interviewsTake part in an interview, or get interviews and updates by email
People are hungry for inspirational good news stories.
I am on a mission to tell the story of the leaders, companies and technologies that aspire to change the world for the good. Working with me will give you and your brand additional visibility and kudos. Plus, I’ll share your message far and wide.
Find out more how I could help you and your vision get noticed, here.