Pod Insights Practical AI The world needs an AI superhero About this episode From drug discovery at the Quebec AI Institute to improving capabilities with low-resourced languages at the Masakhane Research Foundation and Google AI, Bonaventure Dossou looks for opportunities to use his expertise in natural language processing to improve the world - and especially to help his homeland in the Benin Republic in Africa.
Sponsors:
Fastly https://fastly.com/ – Our bandwidth partner. Fastly powers fast, secure, and scalable digital experiences. Move beyond your content delivery network to their powerful edge cloud platform. Learn more at fastly.com https://www.fastly.com/?utm_source=changelog&utm_medium=podcast&utm_campaign=changelog-sponsorshipChangelog++ https://changelog.com/++ – You love our content and you want to take it to the next level by showing your support. We’ll take you closer to the metal with no ads, extended episodes, outtakes, bonus content, a deep discount in our merch store (soon), and more to come. Let’s do this https://changelog.com/++! Featuring:
Bonaventure Dossou – Website https://bonaventuredossou.github.io, GitHub https://github.com/bonaventuredossou, LinkedIn https://www.linkedin.com/in/bonaventuredossou, X https://x.com/bonadossouChris Benson – Website https://chrisbenson.com, GitHub https://github.com/chrisbenson, LinkedIn https://www.linkedin.com/in/chrisbenson, X https://x.com/chrisbensonNatalie Pistunovich – GitHub https://github.com/Pisush, X https://x.com/nataliepisShow Notes:
Bonaventure Dossou | Instagram https://www.instagram.com/bona.dossou2020 — ongoing: My Year of Fame and how I joined the world of Research https://medium.com/masakhane/2020-my-year-of-fame-909f6cc670edUpcoming Events:
Register for upcoming webinars here https://practicalai.fm/webinars!
Transcript Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.
0:00 Speaker 00 I feel like AI can help like not only economically improve or develop Africa, the continent, but also improve the access to education. Today you're talking about AI to people or to young African men or women, they'll feel like you are a god, like you can do anything. People usually come to me and be like, oh, can you create this model that does this and that? That powerful side of what AI can can do for Africa.
0:31 Speaker 01 Big thanks to our partners, Leno, Fastly, and Launch Darkly. We love Leno. They keep it fast and simple. Check them out at leno.com/changelog. Our bandwidth is provided by Fastly. Learn more at fastly.com and get your feature flags powered by Launch Darkly. Get a demo at launchdarkly.com.
0:48 Speaker 02 Hey, Jared here. One of the things we can count on in the software industry is change. The state of the art changes so fast, in fact, that keeping up can feel like a whole other job on top of your actual job. That's why we created Change Log Weekly. It's our totally free newsletter that we drop in your inbox each and every Sunday. We link to the latest news, the best articles, and the most interesting projects that you should be aware of. We also add a little commentary from us saying why something's important, pointing you to other instances of a trend, or just making a dorky joke to keep it lively. So if you haven't yet, I recommend subscribing to Change Log Weekly
1:27 Speaker 02 and help us help you keep up with the latest. Head to changelog.com/weekly and sign up today. Again, it's totally free and we never spam you. Yuck. One last time, that's changelog.com/weekly.
1:54 Speaker 02 Welcome to Practical AI, a weekly podcast that makes artificial intelligence practical, productive, and accessible to everyone. This is where conversations around AI, machine learning, and data science happen. Join the community and Slack with us around various topics of the show at changelog.com/community and follow us on Twitter. We are at Practical AI.
2:20 Chris Benson? Welcome back to another edition of Practical AI. We are the podcast that tries to make AI practical, productive, and accessible to everyone. I'm Chris Benson, one of your co-hosts. Today, Daniel Whitenack is not with us, but we have one of our former guests, Natalie Pustunovich. If I got the last name right, Natalie? You did. I did. Excellent. Zero learning shot, zero shot learning. Awesome. Natalie is here to co-host, and I would like to dive right into introducing our guest. His name is Bonaventure Dosu. Did I get that close to right there?
2:58 Speaker 00 Yes.
2:59 Chris Benson? How would you pronounce it, just so our listeners can hear you do it better?
3:01 Speaker 00 Bonaventure Dosu. Actually, it's not even a French, it's not even a French or English name, so the pronunciation is okay.
3:08 Chris Benson? Okay, so before we started the show, you mentioned something about your name and what it means, and I would just like to throw that out because I thought it was pretty cool. You want to, you want to tell the, tell the audience?
3:19 Speaker 00 So my first name, Bonaventure, is actually a French name that is right at like as Bon Aventure. And if we split into two like Bon Aventure, it means actually good adventures.
3:33 Chris Benson? Okay, so we're gonna have a good adventure on today's podcast.
3:35 Speaker 00 I guess, yes.
3:38 Chris Benson? So now that I've totally put you on the spot in the beginning, we like to have fun. You want to tell us a little bit about, you know, your background, kind of how you got to this point where you're at right now, and then we'd like to dive into some of the stuff that you've been doing after that.
3:51 Speaker 00 Okay. How I got here, that's a very long story.
3:54 Chris Benson? You don't have to go all the way back to birth. It's okay. If it's interesting, we'd like to hear it.
3:59 Speaker 00 Yeah, so I can say that I did not at all plan, predict, like where I am right now. I was meant to be a geneticologist, someone who helps pregnant women to give birth. I was trained in high school to become in biology and to become, yeah, a doctor. I sucked at math, and of course, I love computer science because I had my first, first computer when I was six. I envisioned myself doing it, but not that much, you know, medicine was my prior step. So a lot of things happened, family and stuff, personal issues, and finally,
4:37 Speaker 00 Jumped right into the field of mathematics. So I then got a scholarship to study my bachelor, study for my bachelor in mathematics in Russia, where I spent five years.
4:47 Chris Benson? That is cool.
4:48 Speaker 00 Yeah, and the education was full in Russian. So there was no English or such things.
4:54 Speaker 04 How long did it take you to get up to speed in Russian?
4:57 Speaker 00 Well, I can say seven or eight months, at least to know the basics of the language and everything. And but then, throughout the years at the at the university, like you don't you don't have any other choice. We did not have any other choice. I don't like to speak. So I could say we're also frankly, we're also like lucky enough because studying math is a kind of studying on universal language, right? So whether you speak good or not the language, I mean, you can still go to the board and write and like you can understand each other. But there are some classes for us, like logic or things like that where you need actually to understand the language and understand the subject of the problem.
5:39 Speaker 00 So it was not so easy, but it was also interesting to try something new out.
5:46 Chris Benson? So before you go on, I'm just curious, you made a jump in there from medicine to math.
5:54 Speaker 00 Yeah.
5:54 Chris Benson? And somewhere you had to make a serious decision on what that was. What? How did you make that leap? What what happened there?
6:01 Speaker 00 So, like I say, I studied math.
6:04 Chris Benson? And then he's majoring in it. I'm struggling to follow. This is a great story, though.
6:09 Speaker 00 My dad, I mean, he's actually retired, but he's an architect. So I was living around mathematics. My sister works in finance, she's an accountant. So math is also a thing, but I was the only one who was backing behind. So what happened is I luckily like met a professor when I was back in high school, who actually showed me that people don't come into this world. They they they don't have the gift of being super duper good at math like at birth, right? It's something actually that you get trained on. It's something, it's a science, so you learn it, right? It's not like someone comes with it. So he transmits, transmits me that passion.
6:53 Speaker 00 And that will actually to keep learning, learning, learning. And you know, the most interesting fact about him was I could completely fail a, let me say, an exam. But he would focus on where I succeeded and give it as example. I mean, it was not only me, but it was many others. So actually, he was not pointing out the fact that the overall like grade or score was bad, but he was pointing out a strength. And I knew then that I was sucking at geometry, but then good in arithmetic. And that's where the whole story started because I love to do that computation, calculation, things like that. And then I started like a lot of training. I could do a bunch of exercises and everything.
7:35 Speaker 00 So high school is divided into two main components or maybe parts. The first part, you first do a four year, that's more general. And after you have, you pass a national exam. And then the second part is where now you go into a more specialized field. Like if you want to go to do mathematics, advanced math, then there's a fit for it. Want to do biology, it's a fit. So I performed well at that national exam in biology, but not in math. So I was now with a written biology. But in SS one, I met him. So the transition from the last SS, SG, SG four to SS one was huge.
8:17 Speaker 00 Because people who were like, like top math student and everything, they saw that transformation. They were like, okay, so what happened? And that I was, I was, that's from where like the whole passion about math and like doing something that involves math like started. And then after I still wanted to do to become a doctor because that was, that's ever been my childhood dream. But then the came a conflict because I discovered computer science. I mean, I used to, but then I discovered the real computer science, like where you know the all the math and algebra behind. It gave me more, more passion. Eh on the other side, I also have few friends who are studying medicine back in Benin.
9:02 Speaker 00 And they used to cram books, like and I told me books and things like that. And that's not my thing. I'm not super duper craver, like I prefer when I had the freedom to design my own thinking way to go to the the target. Not that there's already pre-existing things you have to learn. I'm also someone who learns from mistakes a lot. So I was like, okay, if you become a doctor, you the mistakes here won't be called into the human's life. So, yeah, that was mainly when when like the the thing came. But the more the the big change in fact was when I wanted to come here in Canada to study medicine. And they told us that they do not give that privilege to international students.
9:44 Speaker 00 That I have to be a doctor to have at least a good background before coming over there. And then I'm like, okay, my dream was dead. So then I switched to mathematics, which actually found out turns out to be a good choice.
9:58 Chris Benson? Yeah, you know, I just looking through some of the the stuff that you that you have on your resume. It sounds like you've made the best of a bad situation, I'd say.
10:07 Speaker 00 Yes. Even when I was in Russia, I wanted to do mathematics and computer science. Because I've never started in the mindset of going or becoming a math teacher, right? So I wanted to do something applied. But then they put me, my government somehow put me in maths and mechanics. So I was not doing any more applied maths, but I was doing pure theoretical math. Well, I had a choice like to change and everything, but I still said, I was like, okay, let me take it as a challenge. And, yeah, it ended up like good.
10:42 Chris Benson? So you're in math, but I know that right now you're in Quebec.
10:46 Speaker 00 Yes.
10:46 Chris Benson? Working in one of the coolest places on the planet Earth, doing deep learning. There's still a little bit of a leap to get from math to what you're doing. And I know I haven't revealed like who's advising you and things like that. I'll let you do that. But, you know, Quebec, hint, hint, folks. Anyway, go ahead and tell us a little bit. How how did you get into that?
11:05 Speaker 00 So the core of ML of AI is actually maths, like differential equation, statistics, uh probabilities. And at first, me and my friend Chris, who also knows uh Mr. Daniel, we were like, okay.
11:24 Speaker 00 We were one year from our graduation. And we knew that we wanted to do something that actually was on trend. Like something a lot of people had attention to and things like that. So we did not want to do like the classical software engineering or things like that. We wanted to be more free. And we were think people like who like giving crazy ideas. So like at nine, at three A M, won't be sleeping, we just jump up and then like. Like bring up a crazy idea or things like that. So we found that there doing research will be a better choice, right? Because that's when you can try out a lot of ideas.
12:08 Speaker 00 I mean, we did not want to do research in maths. And then which field of computer is actu- was actually like, let me call it a hotspot. It was A I. Then we looked at A I and we're like, okay, how can we connect A I with the math? That we are currently doing. That's where everything started. We now started reading books, watching tutorials, watching classes. And then we now connected the dots between poor theoretical math and application of deep learning. For instance, when you want to use same driving car, you know, like sometimes you have to actually be able to. And measure the instance like shift or movement of the car or maybe of the camera, things like that. But how is it done?
12:53 Speaker 00 Is by using differential equation, things like that. So when we tried to connect the dots, it came more and more obvious that we needed math. And actually, what we are doing was something that was happiness. And then we now jump to machine translation because we're like, okay. Currently, what can we start from? We see a lot of things. He asked me. How, why do you always speak in French when your mom speaks in phone? In the other language, I'm like, I don't really know phone because in Beijing you are trained and brainwashed to speak French, otherwise you are considered as someone who is from village or who is not civilized. So that is something that not so many people by then have explored because the whole term was around
13:39 Speaker 00 English and high resource languages. So we were like, okay. Let us then get into a space where we actually do something that not everybody is doing. That's like to bring something of, some sort of novelty, novelty. And that's where it started from. Then we started reading, reading, reading, started like taking classes and see, okay, how does, for instance, this convolutional neural network, how does it work? Our machine translation, how is the, the soft mass activation, things like that, those are all based on mathematic formulas, exponential function, logarithmic function, things like that. So we tried then to see, okay, like theoretically and machine learning, how is that connected?
14:23 Speaker 00 I mean, I won't say we've done it, but I'm saying that has been the click, the thing that made the change. And we are still learning, but so far has been so very helpful. I hope I helped like bringing down the gap between mathematics and machine learning right now.
14:39 Speaker 04 So your next step after Russia was in Germany?
14:45 Speaker 00 Yes.
14:46 Speaker 04 How did that happen?
14:47 Speaker 00 Okay, so I wanted to come here in Canada because of the pandemic. It was something already impossible. And also coming here as an international student in the, in the North Americas is something like that, it's very, very, very expensive, right? So we were like, okay, given the fact we can, I can have visa. For Canada, it would be super complicated, also from Russia. And the fact that education over there will be like very, very expressive. Where in the neighborhood is the best fit? And it's Germany. I've always had a love for Germany, sorry. My dad used to work with German people and that's, that's when I, I started loving Germany.
15:30 Speaker 00 All of them were like, oh, this guy or this kid here acts like German people. Also because of my head, because they say German people are very tall or things like that. But um, I said that when I went to Germany, I also found myself that I have a lot of common behavior or with the culture in general. So given that fact that I love Germany, that my dad had worked with people from there and everything. I was like, okay. And I also love the German language, which unfortunately, because I've been moving around a lot, I have not been able to practice. But it's a language that I love. So I was like, okay, let me go there and then like try that out. And then the next step could be Canada. That's actually how it went.
16:15 Speaker 04 That's what brought you to Canada?
16:17 Speaker 00 I mean, Canada was not predicted this soon. I predicted Canada potentially when I'll be finishing my master, but it happened just a few months after I got into my master in Germany.
16:45 Speaker 02 The change log is deep discussions in and around the world of software. And it's been going for over a decade. We interview hackers like Chris Anderson from 3D Robotics.
16:55 Speaker 05 At the time drones were like predators and global hawks and military industrial. They were classified and super, you know, 10 billion dollar things. And we had just built a drone with Lego pieces around the dining room table programmed by a nine year old. And it's like, okay, that should not be possible. You know, it's, when, when a nine year old can do something that is classified, that literally export controlled as ammunition with Lego, with toy pieces. It was something important in this world has changed.
17:24 Speaker 02 Leaders like Devin Zougal from GitHub.
17:27 Speaker 06 In the like 10 to 15 year range or 20 year range, what I would really like is for if you have like three 12 year olds hanging out and one of them's like, I want to be a firefighter. Another one's like, I want to be a lawyer. I want one of them to say, I want to be an open source developer.
17:42 Speaker 02 And innovators like Amel Hussein.
17:44 Speaker 07 I've yet to kind of see applications at scale that don't use multiple languages, that don't you have just arcane stories behind why this weirdo thing exists. You know, like, all right, when you open this file, you're going to have to turn around three times and tap your nose once.
18:01 Speaker 02 There be dragons.
18:02 Speaker 07 Just the most hilarious stories, you know, but applications are living, breathing, they have craft, that's normal. So I want to normalize weirdness because that's just how applications evolve over time.
18:17 Speaker 02 Welcome to the change log. Please listen to an episode from our catalog that interests you and subscribe today. We'd love to have you with us.
18:50 Speaker 04 And so, Canada, tell us about the interesting things that you're working on and with whom?
18:58 Speaker 00 The Canadian episode is very interesting. I came here and I'm working currently with Atmila, like you said, which is one of the, if not the biggest research center in deep learning. Has always been my dream, the dream of Chris too. We have the person who is a currently our advisor, Yashua Benjo, who was one of the top figures that we're looking up to. I mean, you know, when you are kind of new in the field, you tend to look at someone be like, okay, he's achieved this and this, his way of doing or thinking, his material side and things like that. Things that actually like to say, okay, you, I have a role
19:40 Speaker 00 or a role model or a model on the field. So that was him and we really wanted to come here. Plus also, Mila has done very cool thing, very cool publications, things like that. Uh yeah, it happened and I came here in Montreal. I'm currently at Mira working on a drug discovery project. I said that I've also joined Rush, Rush Canada. And that one is more in my passion of working towards biotechnological challenges or pharma-related challenges using deep learning. So I've been doing both at the same time. I also had many other opportunities. For instance, I attended a scientist residence program,
20:22 Speaker 00 which was first dedicated especially for PhD students. Because when I applied, they were like, okay, cool. Your resume looked good, but we're only looking for PhD student. I was, I postdocs, I was like, okay. And then one day they just came down, okay, fine. Like, out of all the applications we see, you probably is the best fit. And then we jumped in and then that was also an amazing thing because it was more practical. It was shifted away from like academia and then it was like brought down into, brought into an industry. And I could actually see how people in industry do it there because it's not only about like creating a model to beat this baseline or this. It's also about the business. How can you bring it to the market?
21:06 Speaker 00 How can you monetize it, things like that? And finally, in January twenty twenty two, I joined Google and we'll be working on building NLP, uh, language technologies for more subservient languages, including mine, of course.
21:24 Chris Benson? I'm guessing that's how you met Daniel as well, working in that space. Cause I know
21:28 Speaker 00 Yeah.
21:29 Chris Benson? we've had some episodes on the show here about all of the work that's being done in Africa that so many people outside of Africa weren't aware of. Uh, it's really amazing work in the NLP space, but I want to hear about some of the work you're doing. But I just have to comment, you are living the dream, man. To have worked where you're working in both of those opportunities, especially is pretty cool. You want to dive into some of the the project work you're doing and and talk about kind of how you got into it, how you're approaching the problem and love to hear some of the detail.
21:59 Speaker 00 Sure. So here at Mila, I'm working as a drug discovery research intern. So it's also like putting together AI and also biology. I'm currently focused antimicrobial peptides. So the background is is that currently AMI antimicrobial resistance is actually been a very big, let me say, threat to the whole humanity. And like big companies like Roche, like the World Health Organization has given that red flag alarming it like if we don't do something to come up with new drugs, then potentially we could be dying from like very
22:43 Speaker 00 simple injury or things like that because batteries are becoming more and more drug resistant. So I'm currently working on a kind of active learning framework that will help like develop, generate good candidates, EMP candidate, EMPs. I cannot dive deep in because I mean, to explain why EMPs, but they are actually the top one candidate and all companies who are in the field are really focused on it. So there are some biological explanation about that, but I don't know yet. So coming to the machine part and active learning part, where actually we try to maximize the production of good candidates. Now, we are doing discovery,
23:24 Speaker 00 right? So we are not like in the mindset of rediscovering or recreating, regenerating what already exists. We want to do new things or like create or rediscover new things. So. That's where the part that we call diversity part comes in. So it's basically like taking, like setting up a lot of things. I've tried, I tried and brought like a lot of ideas, of course, with some other coworkers, my supervisor and co-supervisor. And we tried many bunch of things and then actually introduced a graph based like method that uses actually like committed component to enhance and like to produce more diversity points.
24:06 Speaker 00 It's basically the idea is that if I take a reference set that I know that I grew the MPs from the real world because the data set that we are using is from like a publicly available source that have been developed by biologists. So those sequences, the properties are well known. So if I take that and I compare it to what I have, does it like give me a good performance according to our metrics? One, two, is that statistically different from the initial set? So that's two things we wanted to be actually having the more or equal to the same properties, but at the same time to be different. Because you can imagine a protein, right?
24:45 Speaker 00 When you have like a sequence of the same, you won't do as better as multi different sequences. It's also like a human being, when you are specializing in one, only one thing. Or let me say you are only good at math, but in life they are not, there is not only math, there are also, there's also biology, sports, there. So you have to be like good like mixing. That was the main idea and that's what I have been working on. And I have good results, successful result. And I've written a manuscript for it, like looking forward to potentially send it to subscription, submission to a journal. And like I used to tell my partner, uh, these people will save the world.
25:29 Speaker 00 But it's just a joke because I know all the talented people who are also working on it in the group. And we had also another extension to that, which is called the Gflow Net, which is also a powerful tool. Uh, which is one of the main attention of Yushua and Joe right now. And we potentially like develop it and then increase it to the same task to see how it performs. And brush is more pharmacological, like we have already the drugs and then like the pipeline, the way the drugs are created and everything. Uh, let me say then that we have feedback from users, they found this or this instead of feeling this or that symptoms or or side effects. So what could be the root cause?
26:13 Speaker 00 What could be the reasons why they are doing it? Is it because that there's been a slight mistake during the production or maybe the manufacturing process, things like that? So collecting all those information all together and then like making sense of it to improve actually the whole setup. Yeah, that's it. So at Google now, I recently just started. And I know that the main focus would be on building multilingual NLP uh models for African languages or Sub-Saharan languages. And I'll be with amazing people with whom I've been like working on a volunteering basis on the on many products, like cross-language,
26:52 Speaker 00 question and answering projects, uh, any our project, things like that. So yeah, that's like briefly uh what I'm working on right now. I hope it makes sense.
27:02 Chris Benson? It does. I have a question.
27:04 Speaker 00 Yes.
27:04 Chris Benson? I know that that you're doing NLP in both of both the Google work and the work for the drug discovery. Uh, I know you talk about there being generative models.
27:14 Speaker 00 Yes.
27:14 Chris Benson? When you discuss it, can you tell us a little bit like if somebody out there is listening, there are a lot of folks are tuning in to learn, to listen, uh, to get new ideas. Why approach it that way? What is it that is that that made you say this is the way I want to solve this problem?
27:29 Speaker 00 So I guess like I'm then targeting the project here at Miller, right? Technically generative models because we want to create new things. I don't know if you see the the idea, like if we are like pretending to do discovery is because we are creating new chemical compounds and new structures, new materials. So you can do that by using a predictive model, for instance, where you already have like set of things. And then like you just predict some value on it. Of course, we are using predictive models where our classification of the creation test.
28:09 Speaker 00 In the process, but it doesn't really add anything new, like the new thing that you bring in does not really come from applying or from building a prediction model. It comes from like a generative one because it learns from the data distribution of an existing data set. And then try to generate things that are similar but not equal. That's from where I guess that the discovery comes from.
28:37 Speaker 04 So you mentioned all those interesting things that you are about to do at Google with the Sub-Saharan languages and you're also working on that at Masahane.
28:46 Speaker 00 Masahane, yes.
28:47 Speaker 04 Masahane.
28:48 Speaker 00 Yeah.
28:49 Speaker 04 What do you see as the future of of those developments?
28:53 Speaker 00 Okay, so it is pretty straightforward, giving access to language technology to everybody. I started, I mean, one of the motivation that made me start this is like trying to improve the communication between my mom and I. And I can say that actually I did not like envision it to be way more important because after we launched, like there was also a keyboard that we launched like of native language of language in Benin and in many other countries. And with the translator, a lot of people use it. So a lot of um artists are using, people who are to create music in phone in the native or local language are using it.
29:35 Speaker 00 So it's helped create more content, it helps create more data, it helps create, it helps actually to improve and increase the the representation of the language on the internet. Because there's nothing frustrating more than, oh, you come to from a country or things like that, or even. Let's even say there are sometimes people asking where I'm from, I see I'm from Benin, but I don't know, it's frustrating. And I had to locate it, uh like, uh with respect to Nigeria, I have to say, oh, it's left to Nigeria. Oh, okay, because people know Nigeria. Why? Because it's big. Why? Because there are a lot of people coming from there. I mean, a lot of things, why Nigeria is not right.
30:13 Speaker 00 So same thing applies here for the language. We have people like me who are abroad, who actually can't actually speak the language, but the parent back at home are only speaking those languages. So hard to improve the communication between those people and the parents. Uh also like language resources, like when you go on the internet or Facebook, things like that, the translation of those languages are pretty often not good at all. And it's also misled. You can see also sometimes a lot of like hate speech going around. So actually, building or working on such kind of model will definitely benefit the marginalized and underrepresented group. And will also actually increase their representation of the language.
30:58 Speaker 00 We could go on the internet like in the next few years, I hope I'll be able like to have a voice assistant, for instance, for my kids, I hope my children, which we'll be able to speak in front with them, because that's also part of the culture, right? That's also part of the identity. Yeah, that's also another interesting part, like working on those languages helps to the restoration of the cultural identity of things like that, like things actually are intersect to the, to the nature of things like that. So it's, it's way more above like just trying to build models. No, it has a very deep root, which like in preservation of those languages,
31:41 Speaker 00 we've seen like a lot of uh groups of people who actually got instented because actually they stopped like being in in numbers. So we don't want it to happen with our other languages and everything. So I think I touched a lot of points. That's mainly how vast the problem is and how I think that working on those kind of language could actually help solving those problems and bringing language technologies to everyone.
32:09 Chris Benson? It strikes me that though you're using similar toolsets, you're using deep learning in both problems. They're somewhat different, one being drug discovery and one being under-resourced languages.
32:21 Speaker 00 Yeah.
32:21 Chris Benson? But I also notice that you seem to gravitate toward doing kind of improving the state of the world. Things that will help people and and help do that. Is that an active part of your decision making as you're looking for projects to work on? To try to make the world a better place because that's the common thread between two otherwise apparently unrelated projects that you're working on.
32:45 Speaker 00 Yeah, I did not mention it because I I mean, I like to look like a superhero.
32:53 Chris Benson? I love the honesty in that.
32:55 Speaker 00 But I don't like to show off too much either. Like there should be a very like good trade off between both. But yeah, you said it. I've always been in the mindset that of course we are all mortal. So we came and we are going to die one day. I want actually to leave things that I can call a legacy so that when I leave, people will still pointing at those things and remembering about me. We have like good mathematicians nowadays who are still like their theories or their formulas are still like helping into solving the world's problem of biology. Scientists Einstein is is there like since a while, but his name is as if he's not there because they keep calling him like he's still useful actually to the world.
33:40 Speaker 00 So yeah, I always have that mindset of whatever I'm working on, make it like somehow to the best of my capacities. Improve humanity, impact people. I know like a lot of friends, a lot of, let me say, young people back in Africa. If you ask them, oh, what are you actually interested? I mean, who is your? Role model, who is the person you look up to? I mean, people who are actually in computer science or who have interest, they will point out that big gates, they will point out that people like that. For instance, even myself, I point out as Joshua, who is a Canadian.
34:20 Speaker 00 But I would like to be a role model for other people back in Africa, like from the next generation. And for doing that, I would definitely have to bring something new, right? Definitely had to have in some way. So, yeah, I always have it in my mind. And the other part of why I'm working also on health, or I'm really willing to do so, is because of the access to healthcare, which is actually a very, very, very big problem. Access to healthcare should be like a primary need, right? But in Africa currently is a luxury to have access whenever you want to healthcare.
35:01 Speaker 00 People die from me. Um, I had a sister that I never knew because she actually could not come because of a problem and she came before her term. And because of the lack of cares and things like that, she could not. Many people live the same situation. Or people actually who could not have access to surgery because they don't have money or because there is lack of electricity or things like that. So there's access to health cares is still a very big problem and a very big field that we scientists need to work on. So I would also like to bring something new into that, maybe through my research in A I,
35:41 Speaker 00 doing drug discoveries or things like that, actually contribute actually to the, to the development, to the emancipation of the continent and not just like focusing on myself on the fact that, oh, I'm working at good places, there are people, places that everyone would like to be at. Maybe also maybe making good money of doing this or this versa.
36:01 Chris Benson? You've gone from medicine to, you realize that you have invented the notion of the A I superhero, which I actually love that idea. It's a beautiful idea. It's great.
36:12 Speaker 00 Yeah, I would say, yes, it's actually your point, the fact that I wanted to be not the superhero, but someone who helps, someone who brought to the world, to the continent, to his country when he was alive. That one brought with AI because I saw it as a tool to do those kind of things.
36:32 Chris Benson? My advice, my friend, is to own it. And the reason I say that is kids will aspire to it. You'll bring the next generation along that way. So what would you, what are you thinking about as you look toward the future? You've kind of hit these two spectacular areas as projects. Uh, drug discovery, which uh will presumably have a high impact on medicine in Africa and elsewhere.
36:57 Speaker 00 Yeah.
36:57 Chris Benson? Where you can really impact people's lives that way. And you've also simultaneously addressed a whole different problem that you're addressing in terms of underserved language, uh, under-resourced language to help people communicate better. You've kind of tackled two big things. What are you thinking about in the future? Are there certain aspirations that you would like to say, I would like a swing at that and see what I can do to make it? Is there, is there anything that you can envision yourself after you complete these projects that you might want to go do?
37:27 Speaker 00 I would say education. That's also one of my dream to be able to come up with a research institute like Mila or in any other places in Africa. You know, most of the fundings, most of the researchers, even like during this pandemic, most of the vaccine, things like that did not come from Africa. We just like waited and waited. We waited and like for people to have solutions and, and actually have been even suffering to have done vaccines. And at that time, we think people are going stick, but they are not. They worked on it. They have to care about the people first. I feel like AI can help like.
38:09 Speaker 00 Not only economically improve or develop Africa, the continent, but also improve the access to education. Because when today you're talking about AI to some, to people or to people, uh, to young African men or women, when they don't know, like, how, what is it actually, they'll feel like you are a god, like you can do anything. People usually come to me and be like, oh, can you create this model that does this and that? But I'll ask them, can you actually do that by yourself? Because if you can't actually teach a computer in a very succinct way, or if you actually can't do it by yourself, so how you're going to teach it to a computer?
38:51 Speaker 00 Things like that. So they still done lack of education. There are not so many universities, for institutes, programs that are actually training the people, the young generation people to come to actually acquire that knowledge. That powerful side of what AI can, can, can, can, can do for Africa. And for that, like I was saying, one of my dream is to create a research, or to have a research institute in Africa, like the Mela, where actually world renowned scientists could actually come, where actually a lot of research would be ongoing, whether it's on vaccine, whether it's like on promoting the languages,
39:36 Speaker 00 working on tackling healthcare problem of the continent, because we have problems over there that for us, if you're there, they not have in Canada or in the US or in France or in Germany. So every continent, every country has its own reality. So one institute that will actually attract investors, that will attract world, uh, world scientists to Africa. And that will actually make it a bigger hub of AI. I mean, not bigger, I mean, I don't know how big it's going to be, but a reference at least in Africa. If I, I succeed in making it at least a reference in Africa, that would have already been great.
40:16 Speaker 00 So I talked about education, I talked about healthcare, and like I said earlier, there's also access to language technologies. I hope I'll be still alive to witness there, our languages like incorporated into.
40:31 Speaker 00 Cell phones. So like the keyboard we develop, people are already using, but the model that we develop, people are not, they have to go to a website to be able to use it. But what if it was integrated into an app? That's something I hopefully wish achieve by this year with my friend Chris. But if there was an app integrated in the phone, so that they could just easily use it like they use Google Translate or things like that. And those language technologies would come from AI and from many other groups, many other sectors, activities, like electronic expenses, that should be interdisciplinary collaboration. But still, AI has its own role into creating those technologies so
41:12 Speaker 00 that other field could use it. So those are the potential trade fields and the true ways I feel like AI can impact people's life in Africa. And I also share one of my dream of making one of them like happening.
41:27 Speaker 04 That is so inspiring and that is so many things and yet it feels like something that is on your shortlist, you know, that's kind of like, Bonaventure, what is your next five-year plan? And it's like, gonna do that. And then, you know, in five years, we'll have you here again. So what's next? With all the wonderful things that you're, you have achieved and you will be achieving.
41:48 Speaker 00 I hope so.
41:49 Chris Benson? Well, Bonaventure, thank you so much for coming on as a guest today. And it's very inspiring. I'm looking forward to releasing the episode. I really actually hope some of the parents out there will share it with their kids. And Natalie, thank you very, very much for coming on and co-hosting with me today.
42:05 Speaker 04 That was such an interesting conversation.
42:07 Chris Benson? It really was. It was fantastic. Thank you very much, both.
42:10 Speaker 04 Thank you.
42:11 Speaker 00 Thank you so much for having me.
42:16 Speaker 02 All right. That's Practical AI for this week. Thanks for listening. If this is your first time with us, subscribe now at practicalai.fm or simply search for Practical AI in your favorite podcast app. We're in there. And if you're a long time listener, do us a solid by recommending the show to a friend. Word of mouth is still the number one way people find new podcasts they love. Special thanks to our partners for supporting our work, Fastly, Launch Darkly, and Linode. We appreciate it. And to the mysterious Breakmaster Cylinder for cranking out new beats for us all the time. That's all for now. We'll talk to you again next week.
Previous Democratizing ML for speech Next Exploring deep reinforcement learning Page updated October 26, 2026 · transcript and summary are machine-generated