Yes We Can

Hi everyone, thank you so much for tuning in to this week's Yes We Can. I'm Michelle Zatlyn, I'm one of the founders of Cloudflare, also our president, and I'm just so honored to have Carolyne here today. Hi Carolyne, welcome. Hi there, thank you so much for having me. I can't wait. Today we're talking about artificial intelligence and what that means and how you can get it, if you want to learn more, what you can do, if you want to go back to school, what you can do, and all the different practical applications of it. So I'm really excited to have you here. You're based in Montreal right now, correct? I am, yeah. How's everything in Montreal? The snow is melting, it's getting nice out, so I'm looking forward to just getting outside after the winter we've had. Good, that's good. Well, it's a beautiful city, so if the list audience has ever been, it's definitely a good one to add to your list. All right, well let's dive in. So you're currently finishing up your master's program with AI, which is artificial intelligence, that's what AI stands for, for anyone tuning in and wasn't sure. And can you maybe just start by telling the audience what is artificial intelligence and your description of it? Sure, absolutely. First off, I find the term a little misleading because it's all about human-generated data, human -implemented models, human-implemented algorithms, and I guess it's artificial in the sense that when you put all of these components together, it's making automated decisions. But it was motivated by biological events, so the activation functions we use to propagate the signal in a model were inspired by neurons in the brain. And going back to your question about examples, think of speech, so your Google Home, your Alexa. In vision, we have self-driving cars. In text, we have Google Translate, and it's such a big industry. That's cool. Okay, so I like how you said how it's actually, there's a lot of human in artificial intelligence, which is a little misleading. I've never really heard anyone point it out, but as soon as you said it, I thought, huh, that's actually a good point. Yeah, thank you. It is. Okay, so one of the things I was so excited to have you on the topic, because I do think AI is something I hear so much about, but it does feel unapproachable, a little scary. It's like, oh, I'm not an AI expert. I don't even know where to start, and it seems so technical. And it's also such a growing field. And so do you agree with the sentiment that it's a really nascent field, growing, becoming more and more important? Has that been your experience as well, kind of from your expert perspective? Yeah, absolutely. I mean, I think it's just going to affect every industry, and if it hasn't already, I mean, it certainly will. And I really think about it as enhancing human capabilities, so automating human tasks, really helping the human doing their work in an augmented way. That's good. Well, if you have to have the humans to generate the data, then it helps make automated decisions. It's interesting about all the different applications. I always worry when people say it's going to impact everything. That sounds scary. Sounds like I'm going to lose my job. Why do you need people? How do you kind of reconcile those two things? You know, I think it's really going to shift industries, but I see it creating jobs as well. So as you mentioned about the data, there's a lot of work that needs to be done in data annotation, cleaning the data, many things that can be done without a background in computer science. We need to build multidisciplinary teams, so we need to have social sciences on our side. You know, it's really a collaborative field, and that's how it should be built. That's good. Well, sometimes when I hear what you're saying, I think about, okay, people are listening, and they don't really know a lot about it. And they're like, oh, this sounds interesting. It's a great topic to learn more, because it sounds like you agree it's going to become more important and become bigger parts of our lives over time, not less. So it's kind of say, if you don't already know a lot about it, no better time to learn more today. Absolutely. I feel the same with technology, where some people are shy away, think, oh, I hope the Internet goes away. I'm like, the Internet's not going away. You got to just embrace it and become a better digital citizen. And I think it sounds like it's the same with AI. It's not going away. It's only going to become more important to all of us. Definitely. Good. Okay, good. So you've been studying in your master's program, and one of your areas of focus is actually in gender bias in AI, which is so topical, not only because it's Women's Empowerment Month, and so gender is a big topic for this month, both at Klappler and around the world, but it's also something that comes up a lot around AI of, are the algorithms inherently have biases in it, and what are the implications of that? And so your area of focus is to detect it and to help mitigate these biases. So can you maybe start by telling us, why do gender biases in data sets even, why are there biases to begin with? For sure. Yeah. I like to start off by saying that, first of all, it's a really complicated problem, and it should be tackled not only by machine learning researchers, but we should be building these multidisciplinary teams of social scientists, in my case, people in gender studies. And again, because I study gender bias, it should really reflect an inclusive definition of what gender is. So for example, perhaps there's better ways of encoding gender than the binary version we see today. So maybe a continuous scale would be more expressive. So you have on one end male, one end female, and then you have everything in between, and maybe you're not on the scale. There's other ways to encode gender that we really need to think about. And to your question about, why is there bias in the data sets? I mean, so if you have a frequency imbalance between the association between male and female associated with the term nurse, for example, that's going to be reflected in our models. So part of the problems is that the statistics in the data set, they show themselves in the model. And lastly, just to wrap this up, if the third person plural pronoun, for example, isn't in your training set, then that won't ever show up in your model at inference time. So that's non-inclusive to people who identify as non-binary and use that pronoun. So it's a problem, but we're working on it. That's good. Well, it's so interesting because when we were prepping and I was learning more about this field, because I know a little bit, but you've really helped open my eyes. I thought the example, how you kind of broke it down, you're like, look, artificial intelligence needs models to train from. It's got to go get the data. The data is all human generated. And so, I mean, let's just use your nurse example. If you go scrape the web, the Internet for a bunch of content on nurses, there are more content. There's just inherently more content about women nurses than men's nurses. Is that a good example? Exactly. Yeah, that's correct. So then when you ask a model that's been trained on this data, can you predict the pronoun within the sentence, and there's the word nurse in it, it's going to predict a female pronoun almost every time. And the nurse in that, the issue was that, and again, I'd like to specify that this is mainly for English language. So it's not always the case for different languages, but because in English we have ambiguity with our pronouns sometimes, that's where the problem occurs. That's good. So I think this is a good example just to help contextualize. It's like, yeah, there's just more content about women nurses online than men. You scrape the data from that inherently, then it predicts when it says, oh, nurse, and then they put she or her in front of and that excludes men nurses. Or as you said, some people who might not even identify as a, they might identify in some other way, either he or she, maybe they may. Absolutely. Got it. So that seems like a problem. So how do you, I mean, not that we're going to solve it on this call, but maybe you can just share a little bit of like, should we be optimistic for the future? Or is this all doom? Is there a solution? Is there a path forward to help solve this? Because I don't want a world where all engineers are men or all nurses are women. I think that that doesn't actually reflect the world that I know. So how do you, are you optimistic for the future or what do we do from here? Yeah, absolutely. It's a big area in research right now. Industry, academia, everyone is looking at this. And again, it's really about building these multidisciplinary teams. Tech is an industry where it's kind of like move fast or die. And some of these models crept in without perhaps having the right people in the room when they were being built. It's understandable, I suppose, but these models have been retracted. And I think there's a hopeful path for the future to fix the biases that we've been seeing. That's good. Well, I feel some of that. You asked a lot about being a woman in technology and I kind of feel the same way. When I started 10 years ago, no one really talked about it, but now there's a lot more, there's a lot more willingness to talk about our awareness. And so it's, I think that is a step one of helping to fix it. So it sounds like similar with some of these biases that exist on the AI side. It's good. I'm glad you're there, Carolyn. That's important for all of us. Okay. One more talk about this and then we're going to switch gears to kind of how you fell in love with computer science. But before we do, as you've clearly gone, what I would describe down the rabbit hole of artificial intelligence and become so enthralled, and then again, studying the gender biases and seeing it and being like, oh my God, how are we going to fix it? Has anything surprised you as you've kind of fallen in love with this field? I mean, I think, like I just mentioned before, what's surprising a bit that some of these models have crept up in production. So these models are affecting people's lives. And I understand, I guess why it happened, but it's surprising that both in academia, in industry, third party researchers have published some of these models. And now we're in a situation where we kind of have to like, go back and fix it. But these models are out there. So it's the situation that we're in. Mm-hmm. Okay, good. That's good. Well, thanks for sharing that. Sometimes I feel like I talk to people and they think about everything, all the problems have been solved in the world. And it's just like, I think, no, there's a lot of problems that need to be solved. Exactly. We need great, talented people around the world who care about these different areas to go solve them. It's what makes the world go round. So I think that's a good, highlighting a good point of no, it's not, it's definitely not perfect. Okay. Switching gears. When we were getting to know each other, the exact words were like, I fell in love with computer science, which made me so happy. I smelled when I heard that. And then there was a second part of that sentence of I fell in love with computer science later in my life. Not when I was young, a little bit later when I was in my twenties. So maybe you can share with the audience, how did that happen? Yeah, I'd love to. Yeah. So my education started off actually in anatomy and cell biology. And after this, I joined a startup where we were building a marketplace to trade intellectual property. So that was my first introduction of, of tech, the whole tech world. I was actually on the customer service team, product feedback team. But, you know, I was just so inspired by the fact that we could build something ourselves and other people were paying for that. And it really just set off this fire in me where I was like, wow, like, I really want to learn more about this. And I just, I need to know more. So when the startup was acquired back in 2015, rather than joining the new company, that's when I decided to go back to school in computer science. And yeah, and I was 27 at the time when I started learning how to code. When you go back to school, when you say go back to school, just to make it clear for the audience, you don't mean a bootcamp, like a six week bootcamp. You mean literally back to college or university? Yep. I was 27 back in undergrad. And, and so what was that like for you? What was it like going back to, I mean, that, that takes a lot of courage, first of all. And, and obviously you're also very lucky to be able to get to do that, but a lot of courage to go back to undergrad as a 27 year old changing, you know, you went studied science and now you're back for computer science. Like what was that like? Yeah. You know, I feel very fortunate that I was able to go back to school. And, and, you know, you're in a situation where you're 27, this is where your friends are like hitting their stride and you can't help but compare yourself. You're happy for them, but you're just like, all right. And, you know, it was, it was a humbling experience and I would have, I'd do it again. Like it really changed the direction of my life and it affected my personal relationships, my every aspect of my life, my personality, everything. So I would, I would do it again. And yeah, I mean, the first, when you're first learning how to code those first computer science courses, they're difficult and it's sort of this process where you have to rewire your brain. Like for example, the equal sign, it's no longer inequality. It's an assignment operator where you're assigning a value to variables. So there's this whole rewiring aspect and you really need to give yourself the time to, to learn. You need to have grit to kind of push through those, those tough assignments. But it's possible, you know, I hadn't, I hadn't touched those, those math classes in a while, you know, calculus and stats and everything. So I had to relearn everything kind of quickly. It's possible, but you have to spend the time. I was really, that's all I did for X amount of years when I was learning. Wow. That's good. I'm so glad you're willing to talk about this and share this because I, I think there's a lot of people who maybe would love to go learn about something else, but they don't even think it's, they kind of like, I can't do that. When, when you told your friends or even your family that you want to go back to school, were they supportive or did they try and convince you not to do that? You know, everyone was very supportive. I was, I, I had to move back home pretty much. Like, first of all, I was very lucky that I was able to move back into my dad's house. And, and, and my family was super supportive because they could see how passionate I was in it. And I was like, this is my path. My friends are like, all right, you know, you're back in undergrad, let's do it kind of thing. I'll still be your friend. Exactly. You know, and then I also made wonderful new friends in computer science who are some of my best friends today. So yeah, it was, it was an experience. For you. That's great. And, you know, again, I, one of the things that you were describing just like the, like, how did you know to do that? Like it's, I mean, obviously you've been working at the startup and you have loved this experience, but that's, again, that's a big decision and a multi-year decision. That's not a six month decision. It's something that lasts longer. Like how, what kind of gave you the conviction or was there kind of like an aha or a couple of has, or you're just like, I just have to go see where this goes. You know, like to be completely honest, like I, I needed a change in my, my personal life. And I saw how, how the tech world I started as a gateway to kind of like being able to do better for myself. And I just find that it's such a great industry for women, especially because the pay is awesome. There's amazing, like parental leave options. You can work remotely and it's so creative. It's so fun. You can build whatever you want. I just like, I saw all these things and also you don't need to spend a million years in med school or like take the bar exam or do all these things. You know, it's kind of like a very quick ROI. And you're willing to work the lifestyle afterwards is, is really rewarding. Cool. Good. Thanks for sharing that. That's good. I love that. Okay. So fast forwards today, you're so in addition to you went back to school, you become an AI expert. We're going to talk about your, the AI program masters program that you're part of, but you're also now the CTO of a high growth tech startup called a script scrap snow scripts swap. It's hard to say. I'm glad I'm not the only one because I, people say that about Cloudflare. They all say that's hard to say once people say cloud fair and I am empathetic to that. And so I've been practicing this, but obviously I need to script swap. So maybe tell us a little bit, what does squirt swap do? Absolutely. Yeah. So script swap simplifies the jargon in contracts. Our, our customers, our contract managers at big banks, insurance companies, et cetera, who want to increase the speed of negotiation and time to signature without necessarily incurring in-house legal fees. So that's what script swap does. And these simplification transformations simplify the not only grammatically, but also formatting wise. Yeah. So that's a bit of a recap of what we do. And, you know, when I came on as CTO, the team had already built these wonderful rule-based heuristics that take a complicated paragraph as an input, and then it spits out the simplified version. And the beautiful part about this is that they got these transformation checked by plain language experts using our, our custom, we have, we built a custom data annotation tool. So, you know, when I came on with a background in NLP, I was kind of like salivating at this dataset, like it's large NLP dataset, it's clean. And so I saw it as a great opportunity to leverage the latest NLP models to learn this transformation. So now we don't use rule -based heuristics. We're using the big transformer models to learn the transformation. And we'll do, we'll be deploying our models through an API at the end of the summer. Well, that's great. So this is like, you fall in love with computer science, how creative what you can build, you became an AI expert, you fell in love with that. And now you're using it in a real life application to help make businesses and people's jobs more productive and help, help detect complicated jargon and contracts to surface it and say, Hey, let's look at these areas for legal teams. Absolutely. Yeah. It's been really fun. It's been a good journey. Right. Right. So if people want to learn more about ScriptSwap or you, what's the best way to, to maybe you can share the website or if you're on Twitter, help me in. Scriptswap.com. And then I'm also on LinkedIn. You can find me Carolyn Peltier. Yep. That'd be the easiest way. That's perfect. Great. Awesome. We have about 10 minutes left. And so we're going to go back to you. So again, you just, you've had such a journey and it's, it's inspiring to hear about how you just went for it and why, and that you're loving it. And there's a lot of people who maybe want a change, a new chapter in their life. Right. And, and so when you went back to school, how did you then, when you graduated your computer science degree, how'd you decide to go do a master's program in AI? And then you ended up at Mila, which is a program in Montreal, beautiful Montreal. And how did you pick Mila? So maybe start with that. Sure. Yeah. So when I finished my computer science undergrad I knew I wanted to get into AI, had taken a couple of courses in undergrad and I could sense like this was really going somewhere. And it's going to be important, can be more important 10 years from now than a place to bet your career on. Yeah, absolutely. And I looked around and I was so lucky to find it's called the AI for social good lab. It's a lab mainly to elevate women in the industry of AI. It's a program in the summer. And that's where I met my colleagues. We co-founded Biasly AI. So this is a group of researchers that are looking to tackle bias in machine learning. So not only gender bias, but racial bias. And so we co-founded that at this lab. We then go to a conference and see professor Yoshua Bengio presenting. After he's done, we kind of just approach him. We, we pitch him our, our Biasly AI project. And he, he accepted us as a research intern in this humanitarian AI group at Mila. So it was, I mean, it was such a great moment for us. You know, we were all still in school. None of us were actually, we had one, one of us was doing her master's, but in a different area. And to be able to come in as research interns, all of us together at Mila under professor Yoshua, you know, Bengio's supervision, it was, it was an incredible experience. And, you know, so that's, that's how I got into Mila. That's, well, I mean, I came in as a research intern, but then I decided to pursue a master's. So that's when I, you know, started my master's in machine learning. It's amazing. I love that. It's so entrepreneurial. I mean, you just pitch the, first of all, this professor you're talking about is like, I mean, the godfather of AI, perhaps. I don't know if I'm overstretching, but he's- Turning award winner, which is like kind of the prize, the Nobel prize of computing. Yeah. Big deal. He's, he's a big deal. I would, yeah. And so the fact that like at a conference, you go pitch this very prominent figure and I mean, to his credit, he listened and said, why don't you come work in my lab? That's pretty cool. Absolutely. I mean, he, he really gave us the time to explain our project. He took a meeting afterwards and he's a busy person. So we were so grateful and he's been one of our biggest supporters. It's been really great. That's great. Okay. So if others are thinking about, well, let's start with not even, let's say that we want to go, we'll get to the master's program, but let's say they just want to go learn more. They're like, oh my God, like obviously you're so passionate about this and you just want to go learn. Where can people go? Like, what should they read? Where can they go learn more? And, and, and just to kind of see, learn more about artificial intelligence, maybe understand some of the technical underpinnings, where would, where would, where would you direct? If they're specifically interested in Mila, you know, they can go to the Mila website, but Mila, so it's an AI Institute. It's based in Montreal. It's a non-for-profit. It's made up of the University of Montreal, McGill University, along with UMI Technique and ASHA SAFE, founded by Professor Yoshua Bengio. Like you mentioned, he's the Turing Award winner. And it's really cool. I mean, it's, it's where Ian Goodfellow came up with the concept of the GAN models. So the, this is all like the, those generative models that we're seeing. It's also where the early concepts of attention came, were introduced by Dimitri Bandano. And, you know, Mila has great relationships with industry. So we have Facebook upstairs. We have Microsoft across the street, Google's not far, but also like smaller SMEs and other not-for -profits. So there's this big collaborative environment, you know, and lastly, there's the AI for Humanity group, which is the group that we came in as research assistants. And so those projects and climate change, COVID-19 and, and the Biasly AI project is actually continuing on. It's been funded and there's, there's other interns there now. So it lives on. You know, early on, you were talking about like earlier on in our conversation, how important it is to have diversity, like looking at AI and multidisciplinary. And, you know, as you just speak about Mila, obviously you've had a very positive experience and it's amazing how multidisciplinary it is between the research side and then also the industry side and the practical application it's, and all the different use cases, whether it's from legal contracts to climate change, just like, whoa, it's, I think you're doing a good job of saying, wow, this is, seems like this is huge. And, and, and it really is going to transform industries and the way we approach problems. Absolutely. Yeah. The other thing that's super interesting is, and I did not appreciate this until I got to know you better is just how these clusters where, where I think Montreal is a city where there's a cluster where they really, where there's this cluster of AI experts and to learn. And, you know, there are other clusters, but I think if somebody really wants to go learn a lot and really get into this field, trying to get embedded into one of these clusters is potentially a good route. Yeah. And there's other clusters too, right? Montreal is not just the only, there's, there's Toronto, there's the Vector Institute. And then, I believe it's Amy, I'm not sure off the top of my head, but out West, there's also another Institute. So, you know, it's all funded by CIFAR and it's, it's, it's growing and like where Canada is, is the home of top AI talent and I'm not sure everyone knows that, so we should be proud. Because they tuned into Yes, We Can and heard it from you. It's good. We have about three minutes left. And so, you know, what I, I, I started Yes, We Can during this pandemic, and it's been a huge highlight, just meeting amazing women around the world, working on areas that, again, your passion for what you're doing really shines through. And so it's always a highlight of my day. And one of the, every conversation has been different. But there's one question I ask everyone. And so I would love to hear, you know, as a woman in technology, where has industry lived up to your expectations? And where has it fallen short? Yeah. So I had a vision of what tech would be like when I before I went back to school. And like, let me tell you, it has surpassed it. And it's such a creative field. I met so many smart people, compassionate, interesting people. And it's way more collaborative than I've ever imagined. And I guess like, where it's fallen short is, you know, I'm sure we've, we've all heard the statistics, LMNCI in 2019 estimated that only 12% of machine learning researchers were, were women. But you know, I'm really hopeful for the future. We have programs like the AI for social good lab that are working to promote women in AI. And we have leaders in the field, like Professor Yasha Bengio, who truly care about these issues and are doing everything, you know, are doing what they can to, to close the gap. Increase it for 12%. It needs to be higher than 12%. Otherwise, yeah, 20 years now, we're going to be in the same spot. And they're just gonna like, Oh, my God, why didn't we do more? That's good. That's good. You know, do you ever think back and think why you didn't fall in love with computer science earlier, like in high school? Or do you ever think back to why it took? Why a little way a bit lucky to find it, you know, you kind of, you went to a startup, and you're like, Oh, my goodness, but why you weren't able to find it earlier? Yeah, you know, I think about this. And to be honest, it was not on my radar in high school. And I went to high school where there was the international baccalaureate program. And it was sort of a situation. So this was back in 2002 or 2003. And it, I didn't know it existed. It was sort of like this thing where like, my peers, so all of us women went into health sciences, we all wanted to go to med school. And then all the guys went into STEM, and like, you know, went into engineering, and it just, it was not an option. I didn't even think of it. And, you know, one of my goals is to really bring awareness to to computer science to everybody earlier on in life, because the sooner you start, the better it is, you won't be 27 in your dad's basement. Although it sounds like things worked out and that you're very happy, but I agree. You know, people often say, if you can see it, you can believe it. And, and I, I, a lot of what you said resonates where, you know, in high school, I didn't even know that tech was a job option, like, or and what all the different jobs within tech. And so we've just scratched the surface today. So thanks so much for sharing that. All right. Well, you are inspiring, Carolyn. Thank you so much. I can't wait to go learn more about AI. Thanks to everyone for tuning in. And so you can come back every week to Cloudflare.tv for Yes, We Can. And if you have ideas for guests you'd like to see, email me at yeswecan at Cloudflare.tv. And thanks so much, Carolyn, for being here. This was fabulous. Thank you so much, Michelle. Thanks, everyone.