What AI needs to focus on right now

Note: Transcripts are automatically generated from episode audio, and are not fully corrected for spelling, grammar, and formatting.

RANA EL KALIOUBY: Hi there and welcome to Pioneers of AI! I’m Rana el Kaliouby – an AI scientist, entrepreneur, investor … and now podcast host. Think of Pioneers of AI as your guide to all things artificial intelligence. We’ll be here every Wednesday, bringing you conversations with some of the brightest makers and thinkers in the AI space.

[THEME MUSIC]

The episode you’re about to hear is a conversation I had with Reid Hoffman, host of the podcast Masters of Scale. He’s the co-founder of LinkedIn and Inflection AI … and also a dear friend and mentor of mine. You’ll hear about my back story and my take on why we need human-centric AI. Plus you’ll get a taste of what’s in store for Pioneers of AI. Okay, onto the show!

EL KALIOUBY: What does an AI first human machine interface look like, right? If you’re designing an interface from scratch, is it really going to be like your smartphone or what else could it be.

REID HOFFMAN: That’s Rana el Kaliouby, a leading innovator in the field of emotion AI — as in, how machines can learn to measure and simulate human emotions. She’s also my friend and colleague.

EL KALIOUBY: AI can unlock human potential, and it can be both massively economically successful, and right and good for people. So it can help us be healthier, more productive, more knowledgeable, good for the planet.

And then building empathy into these devices too. I think that’s so important.

HOFFMAN: We often talk about the importance of IQ when it comes to AI, but Rana wants us to also consider EQ. Emotional intelligence is key to making more human centric AI.

Rana has dedicated her career to this conviction. She co-founded Affectiva, a company building technology to measure complex human emotions. She’s also an investor and author of the best-selling memoir Girl Decoded.

And now she’s adding a new title to the list: podcast host. Her new podcast — Pioneers of AI — goes behind the scenes of the AI revolution and gives you the tools to understand what’s coming next.

Rana is a longtime friend of Masters of Scale. Today, we want to re-introduce her and welcome her new show Pioneers of AI to the podcast pantheon. And anytime I get to speak with Rana about where AI is headed, especially on the fronts of empathy and emotion, it’s a thrill. Let’s get to it.

Rana el Kaliouby to Masters of Scale. I’ve been looking forward to this for a while.

EL KALIOUBY: Me too. Hi, Reid.

Copy LinkFrom early curiosity to pioneering emotion AI

HOFFMAN: So, we’ve covered this a little bit in the earlier Masters of Scale. That was long enough ago. Let’s do a little bit of the origin story. Say a little bit about your path to becoming an AI academic, like where you grew up and how you chose your PhD, that kind of stuff.

EL KALIOUBY: Well, let’s see. So I was born in Egypt, as you know, grew up around the Middle East. Both my parents were technologists, so we were always exposed to a lot of technology early on. I wanted to be an academic, so I decided to study computer science as an undergraduate.

Got really fascinated — I still am — by this human machine touchpoint, human machine interfaces and how it not only affects our communication with technology, but our relationship with one another too. So that became my area of focus, and I left Egypt to go do my PhD at Cambridge University, in computer vision and machine learning over two years.

25 years ago. So people think AI is like this new thing. I’m like, some of us have been doing this for a while. And I pioneered an area or a category within AI called emotion recognition or artificial emotional intelligence. During my PhD, I met this MIT professor. She was visiting Cambridge University to give a talk and we clicked and she invited me to join her as a postdoc at MIT and that’s how I ended up in the U.S.

Copy LinkWhy emotional intelligence belongs at the center of AI

HOFFMAN: One of the things that I think is also important for people who want to go deeper into your story is the memoir Girl Decoded — obviously a great, clever title in lots of different ways. People way under-describe the EQ issues within AI. How much of that understanding, in addition to the path of your life, would people find from the memoir?

EL KALIOUBY: Yeah, the memoir is this juxtaposition of my personal journey with the journey of building the technology, because it’s so intertwined. So I talk about, okay, why do emotions matter? They influence every aspect of our lives. Everything from our little decisions every day to big decisions we make, how we learn, how we connect with one another.

And I’m sure our listeners will recognize that it’s not just your IQ that matters — your cognitive intelligence — but your emotional intelligence, your ability to tap into other people’s emotional experiences. And that’s how we motivate behavior change, influence people, persuade people to do things. People who have higher EQs tend to be better leaders and better partners. And I do believe that’s true for technology as well.

So what’s your view on that?

HOFFMAN: Well, one of the things that I think is funny is the people who say emotions don’t affect how I make decisions — it actually affects them a lot and they just don’t have visibility. The question around what words you choose or how different words affect you in different ways, even if they have largely similar semantic meaning. For example, very classically, I’d say, “I think I might have this point of view” versus “I have this point of view.” Very different emotional impacts.

EL KALIOUBY: But that’s not what people typically focus on.

HOFFMAN: Rana believes that AI models need to be designed to pick up on such nuances — models that can understand the difference between a user saying “I think” versus “I know.” She sees this as fundamental to AI’s development.

And while others have ignored emotion in AI, she’s dived in headfirst. The company she co-founded, Affectiva, revolutionized emotion AI.

Rana has always been a trailblazer. From academia to entrepreneurship, she’s plowed her way in a field where she was often the only woman and the only Muslim in the room.

EL KALIOUBY: I used to wear the hijab. I wore it for 12 years. I put it on willingly and I took it off when it just didn’t feel like me anymore. I also moved to the U.S. as a single mom with two young kids — they were 10 and four at the time, and now they’re 21 and 15.

So I’d like to believe that I’ve had this entrepreneurial spirit in me and when I have conviction, I just go for it. I think one of my core values is having courage to chart a path that maybe hasn’t been charted before. So when it came to Affectiva — here I am at MIT, funded by the National Science Foundation and Rosalind Picard’s lab — we started to get a lot of commercial interest in our technology, which is this facial expression recognition technology.

All these Fortune 500 companies wanted to use it for a variety of applications. I remember we walked into the Media Lab Director’s office at the time, it was Frank Moss, and we said, Frank, we need more dollars to hire more PhD students. And he thought about it for a second and he said, this is not research anymore.

There’s a commercialization opportunity here. And my initial reaction was like, whoa, wait, I’m about to apply for faculty. Don’t mess with my academic plans. But then I realized — and that was the tipping point for me. And I think that’s my message to academics who are listening to this: entrepreneurship is an amazing path to actually bring your technology to scale.

Like, you take it from this academic setting where you’re publishing papers and maybe 10 people are using it to scaling it worldwide. And that to me is what’s so magical about starting a business.

Copy LinkTurning academic research into a scalable company

HOFFMAN: That sense of taking insight and thought research that you get from the academic side and then applying it to how do you build scale companies, scale technologies, other kinds of things.

EL KALIOUBY: So, to say it wasn’t easy — by the way, I want to share a story, more of a confession really. I’m a first time founder when I started Affectiva, it was my first time doing this. And there was a lot of stuff I didn’t know. For example, we had one of the investors who were escorting us at the time send us an email and he said one line, send me your BS.

And I was like, I have no idea what this guy’s talking about. Turns out he was asking for our balance sheet. Although for very early startups, BS also means the other thing, right? I was like, the only BS I know of I can’t really attach in an email. But then our very first product roadmap — we were like, Q1, we’re going to penetrate the advertising research market.

Q2, we’re going to do education. Q3 will be health. It was so naive and we had so underestimated how hard it would be, but I actually think that’s okay. Because that gives you that conviction and that belief that you can go for it. That’s what keeps you going, I think.

HOFFMAN: Yeah. I think a lot of the entrepreneurship stuff you cannot fully prepare for. There’s no, I prepare everything. You could prepare some, but you just have to have that chutzpah and grit and very broad learning capabilities to do that kind of thing.

EL KALIOUBY: Yeah, that’s right.

Copy LinkHow Affectiva brought emotion recognition into the real world

HOFFMAN: So go into a little bit more depth about both your research and Affectiva and what the shape of that has been.

EL KALIOUBY: So the core technology we developed at Affectiva was essentially an emotion recognition system that tracked and identified your various facial expressions and then mapped those to any number of emotional states. And when we first started, it could only recognize expressions — smile, an eyebrow raise, a brow furrow.

But over time, it now has a repertoire of over 40 different emotional states. And the range of applications is numerous. We service about a third of the Fortune 500 companies globally to understand the emotional connection their consumers have with their products, videos and content. And then a number of years ago, we pivoted to the automotive market where we now build driver monitoring systems and cabin monitoring systems to really understand what’s happening in the car and with the drivers — is the driver attentive, are they falling asleep?

Think of it as being a co-pilot to both driver and also passengers in the vehicle.

HOFFMAN: In 2021, Affectiva was acquired by a company called Smart Eye. But the closing of one door meant the opening of the next.

Rana has since launched her career as an investor, focusing on companies revolutionizing the field of AI.

EL KALIOUBY: So I’ve been investing for the last three years. I started a fund with a friend of mine, Rob May. We’ve made 40 investments. And then I’m now partnering with Gabby Seiderveld, who was part of my team at Affectiva, and we are launching Blue Tulip Ventures to invest in human centric AI — so entrepreneurship again, but in a different form.

I just love being in innovation spaces.

And I also love supporting founders on their journeys. I found from my experience that some of my investors were so instrumental to our journey at Affectiva — being thought partners, being strategic advisors, being connectors and door openers. And I want to do the same for founders of early stage companies.

I also think it’s an amazing time in AI, and actually you’re one of my inspirations and mentors when it comes to that. It’s an amazing time to be investing in AI. It’s still early days. I think the next generation AI companies and transformative companies of our generation are being born now.

And how cool would it be to play a part in these companies’ journeys.

Copy LinkWhy now is the moment for human centric AI investing and storytelling

HOFFMAN: Now, part of what we’re doing in this podcast is we’re going to be introducing another podcast because in addition to your hats as academic, inventor, entrepreneur, CEO, investor, you’re also becoming a podcaster and part of what you’re going to be doing is Pioneers of AI.

So say a little bit about what the podcast is. Why now? What path, what journey will the podcast be heading out on?

EL KALIOUBY: Yeah, it’s another kind of entrepreneurial journey, I would say. So Pioneers of AI is a guide for all things AI for people who are both experts in the AI space who kind of just struggle with keeping up with all the innovations and what’s happening in AI.

But ideally, one of our goals is to also expand the audience for AI and make AI more inclusive, so that you can come to the podcast to understand AI, embrace AI, anticipate what’s coming next with AI.

We definitely want to take a balanced view on all the potential good that can come out of AI, but we also want to be underscoring and highlighting where things can go wrong so that hopefully as a society we can guard against that. We will be featuring AI builders, of course.

So those in the kitchen, building the next generation of AI technologies and use cases, but also AI thinkers who are particularly thinking about the implications of AI on culture, on society, on our arts, on science, on education, on parenting. So we want to be tackling some of these everyday questions as it relates to, okay, how is AI going to impact my life on a personal and professional level?

I also am very passionate about platforming voices that we don’t typically hear from. So that’s going to be one of our goals as well.

Copy LinkWhat trustworthy AI agents will need to truly know you

HOFFMAN: Pioneers of AI is a podcast that focuses on all of the big questions around AI and the latest trends. Many technologists will say that the biggest trend in the forecast is Agentic AI — that is AI that can perform tasks for us without any assistance.

It’s an important topic and one that Rana is addressing head-on in her podcast.

EL KALIOUBY: The best AI agent, in my opinion, would be one that knows you really well. And one way to know you really well is, of course, to get access to all of your information and maybe whatever health device you’re wearing and your bank account so that it can act on your behalf and whatnot.

But another way it gets to really know you is through your emotional experiences — what’s your emotional state? Are you stressed? Are you happy? Are you anxious? So that it can really customize and personalize its interactions, just the way an awesome friend would do. And I think that’s still missing from where AI is.

And memory plays a big part of that. How do you build actual models of memory and getting to learn a person?

HOFFMAN: Yeah, because part of how trust is established is that you believe that the person knows you, cares about you, understands what’s kind of good for you and your interests and so forth, and will act in those ways. And memory, of course, is essential to all of that, right?

EL KALIOUBY: And if it’s not there, then the interaction suffers.

HOFFMAN: You know this as deeply as anyone in the world, but obviously one of the things about this agentic future is going to be establishing trust. So what do you think is going to be important? Not just for gaining trust, but actually being a faithful holder of trust.

EL KALIOUBY: I would say really knowing what’s happening to your data, first of all. Understanding who’s behind these — the company that’s building these technologies — I think that needs to be trustworthy. Knowing where and how your data is being used.

That’s so important. So I think a certain level of transparency around that is important and also control over that. The relationship between a person and their AI agents is so important. Like, how much autonomy do you want to give your AI agents? And does that build over time as you trust them more and more?

So there’s a lot of questions that we ought to be asking as design decisions when we’re building these AI agents.

I am so looking forward to Rana’s podcast conversations on Pioneers of AI. She’s just getting started.

[AD BREAK]

And when we come back from a break, Rana takes the mic and asks me a few of her questions about AI.

Stay with us.

[AD BREAK]

HOFFMAN: We’re back with a special Masters of Scale introduction for Pioneers of AI, with AI scientist, entrepreneur, and investor Rana el Kaliouby. You can find this interview on the Masters of Scale YouTube channel and more about the new show at pioneersof.ai. We’re going to switch hosts and you’re now the host for the second half of this podcast.

So I am handing the mic over to you.

Copy LinkChoosing a future shaped by possibility instead of fear

EL KALIOUBY: So exciting. And I will start by saying, you also have a new podcast joining our network. Share a bit about Possible. What’s it about?

HOFFMAN: So one of the things that I think is a great disservice to humanity — that’s happening across media, across government discussions and so forth — is a fear and dystopia approach to technology first, versus a possibility and creation approach. And it doesn’t mean that it needs to be blind optimism or blind utopia, because navigating how you get the good is really important. Like, for example, when you’re driving somewhere, you don’t get somewhere by going, I’m first going to figure out all the things that could go wrong before I get in the car, right?

Like, I’m going to think, well, it could spin out of control, I could hit a tree. You’ve got to start driving towards the location and you have to have some place you’re going. You can’t just go, well, if I eliminate all the bad, then the good will happen. That’s not actually how a journey and a creation and invention happens — it’s a going towards something.

So the real question is, what should you shape towards? And part of the reason I decided to add Possible into the pantheon. So we have Masters of Scale for scale companies and entrepreneurs and journey and impact.

And then Possible is, okay, what could possibly go right? Of course, but what should we be building towards? And it’s definitely positive — but it’s positive in a, how do you chart the journey, how do you navigate kind of way. And that’s why I added it into the pantheon.

EL KALIOUBY: I love that because sometimes you need to create the possibility, like literally for people, and then you plant that idea or you create that image or picture of the future and then people can work towards it. But right now there is so much fear. So you talk a lot about AI unlocking human potential.

What do you mean by that?

HOFFMAN: So last year I published the first book on AI co-written with AI. My co-writer was GPT-4. And I understand why John McCarthy and others called it artificial intelligence, especially back then.

But it actually now has a little bit of a misnomer. So I called it amplification intelligence, because it’s an amplifier of human capabilities. And that doesn’t mean that just like other technologies, sometimes parts of it so outstrip human capability that the technology just does it itself.

I mean, for example, we don’t have humans plowing farm fields anymore. So it’s like any bit of technology kind of moves this and it changes that landscape.

But in the change of the landscape, part of the discourse has been almost like the movie Idiocracy. Like, is AI going to be doing everything? And are we going to be sitting on the couch, drinking Gatorade and watching numbing Aldous Huxley, Brave New World, becoming just sensate bundles of emphatic lymphatic response.

I just think that’s not very likely. You have all these discussions about like, you could just fire my marketing department and use GPT-4. Well, no, you can’t. Right. And by the way, marketing is a competition between multiple organizations. And of course, if you’re not using GPT-4 and other AI tools in your marketing today, you’re falling behind. You’re not using key tool sets, but it’s human beings doing it and being amplified in various ways. And you should be looking at what are the ways in this human competition — just like, for example, I use really good running shoes or I use flippers, or I use scuba tanks — to deploy this technology to be furthering my human objectives, my humanist objectives, my business objectives, etc. And so that’s what the unlocking human potential means. And by the way, AI and the applications it brings will have some of those challenges for people, full stop, and we need to help and be more sympathetic.

But ultimately, better for your children, better for society, better for industry, and if you keep a learning and evolving mindset, it can be better for you too.

Copy LinkWhere AI can expand access to health and education

EL KALIOUBY: So for some of our listeners who are still kind of wary of AI, what are some of the applications of AI that you’re most excited about?

HOFFMAN: Well, part of when I’m talking to government folks who are like, press is telling me that my primary job is to limit big tech — and part of the problem, of course, is a lot of the driver of AI, of this new cognitive industrial revolution we’re getting, is emerging from big tech — and I say, actually, while this is what everyone’s singing the song they should be doing, I have an alternative vision, which is: we have a line of sight, we could construct in a small number of months, a medical assistant that is as capable as an average GP or better, that could run for a small number of dollars per hour. How do you get that to be made and available to everybody who has access to a smartphone? Then you have a medical assistant for the billions of people in the world who do not have access to a doctor.

And by the way, even here in the U.S., there are millions of people who do not really have access to a doctor, other than an ER room, and they may not even have access to an ER room. So that’s one, and that will require some modifications of liability laws, incentives and other things. The second one, which I also think is super important, is a tutor.

Every skill, every capability, every subject for every age, infinitely patient, again, runnable on a smartphone. Wealthy people have always been able to afford tutors for themselves and their kids. Now it can be everybody.

EL KALIOUBY: Yeah. I love that you’ve highlighted those two applications in particular. I’m passionate about both of them. It’s this idea of democratizing access to quality health and quality education around the world. It’s very possible — right back to your new podcast, it is so possible.

Like the data’s there, you can create these co-pilots for health and co-pilot career coaches that can be learning companions and career companions. It’s so exciting. What are some of the things that concern you when it comes to AI?

HOFFMAN: Look, it is definitely possible that bad things can happen. Now, one of the reasons why I advocate very strongly against the existential risk, the robots are coming, Terminator movie etc., is because most of the real bad things to focus on are AI in the hands of bad human beings, whether it’s criminals, cybercrime, terrorists, a variety of things, rogue states, interference with elections — all of this kind of stuff in the hands of human beings with bad intent. Because those are the top concerns that I have. Now, a little bit of the tech lash is people tend to say, well, but what about things like hallucination or perpetuating a racial stereotype, and shouldn’t we not launch until there was zero of that.

And I tend to have the same view that I have on most things, which is: you can never get all these things in these complicated systems to zero. So the better thing is, make sure that you’ve got baseline good, and then be improving constantly on those. So you launch anyway. One of the things that I’ve helped set up with various 501c3s is hosted conversations across all of the major tech developers to talk about what kind of things could go wrong and what are safety cases and what should be on the table.

So people who obviously don’t have visibility naturally get some —

EL KALIOUBY: Sensationalized, right?

HOFFMAN: And look, having low visibility and a fear response is not irrational. Like, what are you guys doing? Are you just trying to build an AI that’s trying to manipulate me to buy things I don’t want, or get me really angry so I spend a bunch of time on social media? Right. Anyway, so those are all definitely in the concerns — but there’s the concerns that you go, let’s steer right now, and then there’s the ones that are dynamic and iterating. And the list of dynamic and iterating is very long. And the list of prevent before it could ever launch is compact and fixed.

Copy LinkHow to balance AI risk with continued innovation

EL KALIOUBY: I’m hearing you also say that there just needs to be a lot of intentionality about how we’re building these technologies and thinking ahead of time about what are the unintended consequences, what could potentially go wrong — but not letting that stop our innovation. So let’s switch to investor mode. The next trillion dollar company is going to be AI first, and it’s being born now. You’ve been investing in AI for many, many years. In fact, I believe you were one of the early investors in OpenAI. What do you look for when you’re vetting these AI startups? Because every startup today is an AI startup, right? So how do you truly vet what’s going to be transformative?

HOFFMAN: Yeah. And I did actually lead the first commercial round into OpenAI. So you look for a set of characteristics, which are sufficient knowledge of the technology, an understanding at least of what problems they’re going to need to be solving.

They may not have the solutions yet, but they have a good map. So part of the investor dialogue with them is, how do you see your game? How do you see what the challenge of this is? And another thing that is a classic entrepreneurial failure is, I’ll just build a great product and it all works out. You have to have a go to market strategy and distribution strategy, which is as important as your product in most cases, sometimes even more. One of the definitions I read about entrepreneurs, gosh, 20 years ago, is: an entrepreneur makes plans that outstrip their current resources. That is a truth. It isn’t the only thing about being an entrepreneur, but it’s like, okay, I’m going to have to be acquiring capital and talent and customers and go to market. And so you have to do all of those kinds of things. So all of that characteristic within the founding team is kind of a general across a lot of these things, although some of them tie into the specifics of what the game looks like now.

One of the things I love about entrepreneurship and technology is the game changes on a very fast time period, so an answer now evolves and changes. One of the things about entrepreneurship that’s underdescribed is that what you’re really seeking is to be competitively strong, which means competing against either nonexistent competition or weak competition. And if you’re jumping into a field where there are multiple really strong startup competitors, it’s much harder, right?

Like, for example, I’m a little bit more cautious about people coming in with a new LLM. There are a lot of those.

EL KALIOUBY: Right. Right. It’s out there.

HOFFMAN: For a few years now. So not never, but you really have to have something that’s an interesting play.

But to look at what are other areas that could be really interesting, and also when you’re charting that path ahead, you’re skating to where the puck is moving towards and where you’re building it towards, not to where the puck is today.

Do you have a good vision of that? So I tend to look for things that are not as populated. And it’s one of the reasons why I led the Series A in Airbnb, and I was the first money in, along with Peter Thiel, into Facebook, and part of the reason why I did the angel investment in PayPal and joined as an executive. In a venture thing, what you’re ideally looking for is that on day one of the investment, everyone thinks you’re a little crazy, right?

And then at like year two, it’s like, oh yeah, that’s credible.

And at year three to five, it’s obvious.

Copy LinkWhat investors look for in the next breakout AI company

EL KALIOUBY: Amazing. All right, we could go on forever, but I am going to wrap up our interview together. So if you could have AI do anything for you, what would that be?

HOFFMAN: I love Ethan Mollick’s line — today’s AI is the worst AI you’re ever going to use. And so this answer should change on a regular basis.

EL KALIOUBY: Oh, I love that.

HOFFMAN: Right. Today, what I would love AI to do — and I don’t think it’s quite there yet —

I would love it to be a front end to my complete communications inbox, everything from email to Signal and WhatsApp, Intelligently triage that, so that, for example, I wake up in the morning and I don’t have to check my email first thing. I just talk to the assistant and the assistant goes, here are three things that you really probably need to know right now, first thing in the morning before you go get your coffee.

EL KALIOUBY: Love that.

HOFFMAN: And by the way, here’s 10 things I might be able to craft good responses to, here are some drafts. What do you think? That kind of thing to facilitate communication, because again, that’s amplification of human potential.

And the idea is that you still keep me very centrally in the loop, but to give me superpowers.

EL KALIOUBY: Love that. Well, thank you so much for joining us.

HOFFMAN: Well, it’s a pleasure. I’m looking forward to becoming a deeply intrigued listener and occasional participant. So this is awesome. While we still have you, search for Pioneers of AI wherever you’re listening right now and subscribe. You can hear the first episode right now! It’s right there in the feed.

It features Dr. Joy Buolamwini — who is one of the world’s leading experts in algorithmic bias. This is a must-listen for anyone concerned about how to make safe AI that benefits all of us.