Name: CEO Initiative 2019: A Closer Look at Emotion-Recognizing Technology
Uploaded: 2019-06-11
Duration: 14 min 1 s
Description: Rana el Kaliouby, Co-founder and CEO of Affectiva, explains how ‘Emotion AI’ will change consumer engagement.

Emotion AI, tell me what that is and why the people in this room should care about it, what it means for business. Well, if you think about human intelligence, it's not just about your IQ, right? I mean, as leaders and in our professional and personal lives, your emotional intelligence matters perhaps more importantly than your IQ. And that's true for technology, as well. As AI becomes more mainstream and starts taking on roles that were traditionally done by humans, like acting as your personal assistant or hiring your next coworker, driving your car, you know assisting with your health care. It's really important that technology doesn't just have lots of IQ but also EQ and so we're on this mission to bring Artificial Emotional Intelligence to our devices. Well, so how you can a device to recognize emotions? How can you exhibit EQ when you're a machine? Yeah, so the way people communicate nonverbal signals, only 7 % actually of how we communicate is in the choice of the words we use. It's 7 %? Is the actual text, only 7 %. Okay. 93 % is nonverbal, and that's split between your facial expressions As well as gestures and vocal intonations. So a big part of what we do is focus on the face and using deep learning and machine learning to build algorithms that read your expressions. Can you show me how that works? Yes, let's try a live demo. All right. All right. I've been warming up my face all day for this, so- Yeah. [ LAUGH ] Here's hoping. [ COUGH ] Yes, let's give this a shot. All righty, so let me come in. All right, so start with your best poker face. My best poker face, okay. Okay, great, [ LAUGH ] so as you can see, the software has basically identified Matt's Face, it identified your gender as male based on how you look. So far so good. [ LAUGH ] And then it's locked onto the main feature points on your face. So your eyebrows, your eyes, your mouth, your lips. And then let's test the software, let's start with the simplest expression, which is a smile. Okay. Pretty universal, all right 100 %, that's a great smile. Yes. [ LAUGH ] And then let's cycle through those. Okay, look surprised. Okay, looking surprised. You can see that it's all mapping Matt's dominant expression to an emoji. Always one of my own, okay. How about a brow furrow? Brow furrow. Which is the corrugator muscle. It's an expression of yeah, anger, exactly, or confusion. Smirk if you're skeptical, and it's asymmetric little corner pull. Yeah, there you go. Disgust. Disgust? [ SOUND ] Okay, that's like a [ UNKNOWN ]. That was disgust and disgusting. Okay. And then, this is my favorite, this is the lip pucker, which is the Kardashian duck face. That's awesome, all right 100 % That's the one I practiced most Thank you Matt. So this is just kind of a toy app to demonstrate how the technology works. When we first started it could only identify three expressions, And using a ton of data and deep learning, it now can read 20 different facial expressions, seven emotions and we keep adding more and more. So you've gathered this data, this wonderful gallery of my own photos now, what would Affectiva, your company, do with it? How would you translate that into something that a business could use. Yeah, there are a lot of use cases. So 25 % of the Fortune 500 companies, so a number of you here, are already using this technology to test how people emotionally engage with content. For example, video advertising, and then the data drives decisions around media spend advertising spend targeting. So that's one of the initial use cases of the technology. So these are the kinds of decisions and working in media I know this that are often made in agonizing ways by dozens of people sitting in a room is this kind of a shortcut. Cut to the answers or decisions? It's a data- driven approach to making these decisions, right? Often, as you said. And often, actually, you're making the decisions and you're not the target audience, right? So if you're targetting millennials, you may not know what will resonate. But now, you have a way to test that And of course, you can test it through surveys, but that's tapping into your cortex. It's not really tapping into the visceral, emotional response people have. Sure, that's the [ UNKNOWN ], you're in the verbal realm, the 7 %- Exactly. Of what we feel as opposed to the 93 % that we- Absolutely. Express, that's interesting. So you've talked a little bit about how your own company's working a little bit more in the automotive sector. Tell me a bit more about. The applications of that there. Yeah, a number of years ago, we started getting approached by global automakers, especially in Asia, actually. And they said, this is very interesting what you're doing with emotions. Can you expand that to other states of mind, like distraction or drowsiness? And so now we're very focused on the automotive industry, which is undergoing a ton of disruption, as you know. One use case is around safety monitoring. So, driver monitoring for distraction and drowsiness, sometimes intoxication even. But what's really exciting is kind of the continuum of use cases through semi- autonomous vehicles, and eventually fully autonomous vehicles, where we're understanding how many occupants are in the vehicle, what's their demographic, how are they interacting with one another. And then the vehicle can personalize the experience and potentially monetize the riding experience. So this is a way to sort of change the atmosphere in the car, give, for example, Spotify a way to know which playlist to play or something like that. Exactly, changing the lighting, changing the type of riding experience. Maybe you want it to be quiet and zen cuz you're gonna get work done. I want it to be fun. Somebody else wants to sleep and the vehicle can adapt accordingly. And can you get that? From what you currently know about facial expressions, can you get that much data, that much information about a person from the images you can access today? Yeah, so the face has about 45 facial muscles that drive kind of thousands of ways to express emotions and cognitive. States and we're building our way up through these different kinda repertoire states. So right now we've collected over 4 billion facial frames from 87 countries around the world and that data's being used to train and kinda validate these algorithms. That 87 countries point that you just made, that's important from an ethical perspective and avoiding bias, right? Because you want a broad pool of images So that you don't accidentally build software that only recognizes white guys who make Kardashian faces- [ LAUGH ] Or something like that. Absolutely, I think this is one of the biggest concerns I have around AI in general at the moment. Which is accidentally perpetuating bias that exists in society in these algorithms because we're not paying enough attention to the divesity of the data. And so, we are very thoughtful about ensuring that the data's gender diverse, ethnically diverse, age diverse. And then, once we have the data, it's all about how to sample these deep learning algorithms in a way that You know equally represents different minority groups. I mean it was really telling actually yesterday that both in the opening sessions with Jenny and Brad, there was a lot of emphasis on facial recognition technology discriminating against you know women of color and and other minority groups, and it's not done on purpose, but we have to be very mindful about how we design these things. Sort of the unconscious bias. Exactly. I have another ethical guess concerned about empathy and emotion and AI, which is this. Not long ago, I was taught that empathy would be essentially the quality that would protect my job from being taken by a robot. My ability to read your cues and to read the cues of someone I was speaking to or talking to Was something that separated human intelligence from digital intelligence. And the thought was that, that was a realm that we kind of the empathy centric jobs essentially would be where humans would be concentrating in this eventual imagined Division of labour Yeah So with technology like this what is to stop an empathetic robot from taking my job or any [ UNKNOWN ] job I don't think of it as a competition between humans and robots or humans and machines. I really think of it with a lense of partnership, so. Because a lot of our communication is digital. How we communicate, again, personally and professionally is often done through digital communication. I think it's imperative that we have emotion AI technology to augment our EQ. What does EQ look like in a very digitally driven world? Right now when you're communicating online, you're EQ's probably like 20 points lower, right. Yes. Cuz you're missing all of these nuanced non verbal communication. So my hope is that our emotion AI technology augments people's ability to communicate with empathy and compassion online. I mean I actually, I'm a big believer that a lot of the bullying that we see And kind of the rhetoric that we see in online communications on Twitter and what not . Its because you don't see the effect you have when you send a person a really mean message. You don't know how you don't see that effect it has on them so. So with our technology could make up for some emotional deficits that are misuse of technology is. One of my favorite favorite examples around that. So imagine you have a nurse, and the nurse now has a number of home robots. Social robots that take care of patients at home. And have the ability to clue in that nurse if the if the patient requires actual human intervention. So it's a way to augment the nurse you know the nurses now these mini bots, right? But they really work for the nurse. Exactly. I have one more question about ethics and then I'd like to throw things open to the floor. Our COI knowledge partner is jam packed and they recently did a jam packed recently did a survey of businesses to ask them about their AI Strategies and one thing we found, is that only about a third of these companies have a framework in place to avoid the claims of bias problems ethical conflict problems that we've been talking about. If a CEO came to you, someone in this room, and asked you how to kind of come up with a framework like that, what would be your quick, 30- second piece of advice? I like to think of it, about the ethical development and the ethical deployment of AI. So with the development, it's things like data consent and data privacy, and mitigating bias. On the deployment, it's Just really thinking carefully about where do we deploy these AI algorithms, and where do we steer away from them? So, we as a company do not work in the security and surveillance, and lie detection space. Sure, sure. We feel that that breaks the trust We have with our users and consumers, so that's one. Development and deployment of AI. Is opting in an important standpoint from the person, you know. Absolutely! So everything we do is opt- in, and often not just like verbose language that you don't really understand, but actually really simplifying it. So you know the camera is gonna turn on, who's gonna get access to the data, Yeah, that's really important. That's great, thanks. I'd like to throw things open for questions, if anyone has any. And, as always, if you've got one, please stand up and introduce yourself. Anyone out there? I'm looking for the yellow paddle. If not, I have some others. [ LAUGH ] [ LAUGH ] Up here at that table, thank you Hi. I'm Paula Goldman. I am the chief ethical and humane use officer at Salesforce, so I think a lot about. Great. The questions you were just asking. And just noting that just in the last few months, facial recognition technology has become such a hot button topic. You have large portions of the transgender community saying kind of ban the technology altogether. You saw the city of San Francisco say. You know we're not gonna use it in law enforcement. So there seems to be a very polarized discussion, rather than a nuanced discussion about the pros and cons of the technology, and I wonder how you think about how we, Balance those things, so we get the good. Right. But also avoid some pretty extreme downsides that people are talking about. Yeah, absolutely.. I mean I do wanna underscore that there is so much potential good out of this technology. One of the areas I'm most passionate about, Is applying this to mental health, so the very first [ UNKNOWN ] case we explored was autism. How kids with autism understand non- verbal communication. But also things like facial and vocal biomarkers of depression and Parkinson's and flagging suicidal behavior. They're so much good and potential Transformative applications of this technology. So I think we have to as a community of AI thought leaders and innovators to really kind of be able to articulate what are the potential for good and what's the potential abuse. And be at the forefront of that, I mean I think legislation and regulation is really key, and it's coming. But I don't think we, as a community, have to wait for that. I think we have to actually kind of self- regulate and put some frameworks and guidelines in place. Have you, or I should say, has your company ever been offered an opportunity that you decided to turn down for these kinds of reasons? Yeah, we're routinely approached by kind of surveillance, and security, and law enforcement organizations. But the one example that sticks out, in 2011, we were raising venture money. We were about two months away from running out of funding, and we got approached by a free lettered Venture arm of a security agency and they wanted to give us 40 million dollars at the time which was a lot of money for us as a company. And we had to think really hard about whether, because the alternative was maybe we run out of money all together and cease to exist and I just couldn't. I kind of went back home and imagined what life would look like if we did that pivot. And I just felt that that was not why we started the company. It was not in line with our core values and so we turned it away and thankfully were able to raise money from like minded strategic inventor investors. I'm glad that turned out to be the case. Rana el Kaliouby is the CEO and co founder of Affectiva and were so greatful you were here, thank you so much. Thank you. [ APPLAUSE ] [ BLANK_AUDIO ]

CEO Initiative 2019: A Closer Look at Emotion-Recognizing Technology

A Closer Look at Emotion-Recognizing Technology

Tech

Rana el Kaliouby, Co-founder and CEO of Affectiva, explains how ‘Emotion AI’ will change consumer engagement. 