Karim Beguir, co-founder of InstaDeep: People are too gloomy about AI

0:00Ed Butler:This BBC podcast is supported by ads outside the UK.

0:30Ed Butler:the highest B2B return on ad spend of major ad networks. Spend$250 on your first campaign on LinkedIn ads and get$250 credit for the next one. Just go to linkedin.com slash broadcast. That's linkedin.com slash broadcast. Terms and conditions apply. This message comes from Schwab. At Schwab, how you invest is your choice, not theirs. That's why when it comes to managing your wealth, Schwab gives you more choices. You can invest and trade on your own. Plus, get advice and more comprehensive wealth solutions to help meet your unique needs. With award-winning service, low costs and transparent advice, you can manage your wealth your way at Schwab.

1:10Ed Butler:Visit schwab.com to learn more. Hi there, I'm BBC presenter Ed Butler and this is the interview from the BBC World Service, the best conversations coming out of the BBC, people shaping our world from all over the world. If you're not a little bit afraid, then you're not paying attention. We have never seen a people so united. Do not make that boat crossing. Do not make that journey. Being born in America, feeling American, having people treat me like I'm not. We're more popular than populism. For this interview, I'm with Karim Begheer. He's the co-founder of InstaDeep, Africa's biggest AI firm.

1:53Ed Butler:In 2014, he launched the company with just two laptops and$2 ,000. Three years later, he was meeting the tech titan Mark Zuckerberg of Meta. By the time the pandemic hit, his technology was being used to track dangerous Covid variants, supporting the development of targeted vaccines. As you'll hear, Karim's belief in giving something back to society underpins his work. whether it's sharing his groundbreaking research or launching a project to bring the opportunities of AI to the developing world.

2:27Karim Beguir:It's about breaking that sort of like glass wall, which is in people's minds that I cannot be successful if I am in Africa. I don't have the same opportunity as in Silicon Valley. Of course, it's still true, but it's a lot less true than before because of the Internet. But also now, because of AI, you could be in a remote tone in Africa, but you would have essentially a team of geniuses working for you. Because modern AI, the IQ of the very best systems, is now reaching 150 IQ. So we're getting into the genius level category. And so in the past, you wouldn't have been able to build a team, a talented team that could compete.

3:11Karim Beguir:Today you can.

3:15Ed Butler:Welcome to the interview from the BBC World Service with Karim Begir.

3:20Karim Beguir:My father had just finished his medical studies and became a doctor in France. And so the whole family moved when I was one year old to the Tunisian desert, literally Tatooine, which is famous for Planet Tatooine of George Lucas and Star Wars. And this was literally the last city before the desert, as remote as it gets, even in Tunisia. And my dad could have made easier choices, either stay in France or like be on a coastal city or the capital in Tunisia. But he was from Tatooine and he wanted to give back. And so I grew up with essentially like dad being like the doctor of the city, having like an exciting childhood in many ways, like playing with the children.

4:06Karim Beguir:I felt I had a very privileged childhood.

4:09Ed Butler:And that idea of giving back was something that stayed with you?

4:14Karim Beguir:Absolutely. I mean, very often we say that we learn from our parents, not from what they say, but what they do. And really, like what I have learned from my dad was really like giving back, putting the human before anything else. And really, like he left a very profound moral heritage, not only in the city, but on me. and later on this guided some of the choices I made.

4:41Ed Butler:You studied mathematics, applied maths at universities in both France and in the United States. You then made a career in finance, right? You worked in banks.

4:52Karim Beguir:I studied all across Tunisia, made it to Paris, Ecole Polytechnique and later the US. And I was passionate about applied mathematics. For me, this notion of applying knowledge to the real world and having a positive impact was key. And at the time, if you were into applied math, advanced modeling, etc, the most interesting jobs were in investment banks. This was just after the tech bubble had burst. And so I did that before moving to my true passions. Tell me about your true passions. Back when I was in New York, I had studied neural networks. I had found them intriguing.

5:31Ed Butler:Describe what a neural network is.

5:33Karim Beguir:So neural network is really like taking a signal, whether it's data, an image or the like, and doing like multiple simple operations on it, multiplications, additions, and trying to get the system to learn. For example, like if this image is an image of a cat, it should tell you it's a cat and not a dog. And so neural networks are a way to go directly from the data to the final signal that you care about. And in principle, it works, except at the time they were very small. Compute capability was not that strong. So it would work for relatively simple stuff. But much later on, I came up with a research paper.

6:18Karim Beguir:And it was talking about this amazing new technology called deep learning. And I was intrigued, like, what is deep learning? So I start reading the research paper and I'm like, well, these are neural networks. I know this stuff.

6:30Ed Butler:And that's what we decided to do. You decided to do this in 2014, right, with the launch of InstaDeep. Now, this was in Tunisia. I've seen you've written that it started with two laptops,$2 ,000 and a lot of enthusiasm. That doesn't sound like very much given the challenge you were taking on. Exactly.

6:53Karim Beguir:And I have to add, without a clear business plan, it's not like, hey, we had this amazing strategy. We were going to do AI at the level of the very best in the industry. And suddenly we did it. No, it was I had this hunch that doing something entrepreneurial, connecting Africa and Europe, in particular, North Africa. This was something that I was highly interested in. But really, like the trigger for me was meeting my co-founder, Zora Slim. Zora was the kind of person that you would feel super comfortable working with, no matter the time of the day and night. I remember us working until 3 a.m.

7:33Karim Beguir:trying to get the website of that project, which we were called L 'Academy. So startups are about people. They're not about like having this magical idea or business plan or whatever. It's really about, it's a human adventure. this will be so tough that it's all about like are you ready for it do you have the right partners is the relationship you have with your co-founder really like rock solid because it will get tested and that's how we created insta deep with like very limited ambitions but that willingness to to do stuff experiment and it was a lot of fun in the early days but you were in tunisia which

8:13Ed Butler:let's be honest, was not a massively rich or wealthy business landscape in itself, was it? And you're trying something which is technology which nobody's heard of. Artificial intelligence was hardly a word on people's lips in 2014, was it? So how did this evolve? What were you actually selling?

8:31Karim Beguir:So software systems, websites, apps. The first years were tough. We took zero salary, four years, me and Zora. The first AI product we sold was in 2016 and was quite advanced for its time. So as a visual AI system to recognize objects and give them an estimate of price, for example, like luxury objects, watches, bags, etc. But in the early days, it was simply a tech company. And then we started digging into AI.

9:00Ed Butler:In 2018, you raised your first Series A funding, This is$8 million or so, which I guess must have felt like a leap forward. What felt like the first sort of major, really groundbreaking technological achievement of InstaDeep?

9:19Karim Beguir:The first breakthrough from a sort of like product and visibility, this came in 2017. Like the context was, we went to the Mobile World Congress in Barcelona. we showed our app and people were blown away you guys are from north africa from tunisia and you built this and we were like yeah yeah and a journalist from pc mag at the time came and saw this and he was very impressed and we ended up in the 20 most intriguing startups of the mobile world congress and this led ultimately to actually me receiving an email from the leadership of Facebook Meta saying, hey, actually, Mark Zuckerberg would like to meet you.

10:02Karim Beguir:I was like, this must be a fake. I called Zora. I said, can you check this? And she was like, no, no. The email is indeed from Facebook, you know, at the time. So Meta. And you should go. And that's how suddenly I land in Silicon Valley. Meet with Mark. We were like, this was a roundtable with 12 entrepreneurs. And I was like, this is actually possible. like Silicon Valley is not this distant dream that you cannot reach. We have a shot at this, ultimately leading the year after to the Series A and our first breakthrough, which was a research breakthrough. I had been tremendously inspired by AlphaGo and AlphaZero from Google DeepMind, this notion that a system could learn even without data, just by trial and error.

10:52Ed Butler:So this is a machine literally teaching itself. Exactly.

10:55Karim Beguir:It's like in this particular case, think about the game of chess and go, essentially like play multiple, multiple games. And starting from zero, the system in a few hours, four or five hours would become essentially the very best player in history if you throw a compute at it. And I was fascinated by this. And I was like, okay, this is cool. But what about applying it with modifications, not in the game setting, but in real industrial situations, improving the efficiency of container loading, improving the efficiency of logistics, fleet management and the like. And that was our first paper, which we decided to publish.

11:34Karim Beguir:And this was an important decision. Like, should we try to hide and patent and stuff? I was like, no, the AI research community is very open. Let's publish. And that was probably like the best thing we had done or decided to do at the time.

11:48Ed Butler:Why was that the best decision? Your IP wasn't going to be stolen by someone else?

11:53Karim Beguir:That was the most surprising thing. Like, I did it for fundamental reasons. And then I realized, actually, it was a very good move because suddenly people got interested into what we are doing. We literally shown that we could do innovative research. That paper actually attracted the attention of many researchers in like the AI community, which led to hiring talents, led to much more visibility and even led to business. So incredibly, being open and sort of being good karma led to much more progress than I ever thought. That was an important lesson for us.

12:34Ed Butler:So you're working on supply chains effectively with big companies in some cases. Then there is COVID-19, the pandemic, and you develop this relationship with BioNTech. Tell me about that.

12:48Karim Beguir:so i actually had the opportunity to meet the founder of biotech go sahin the founder and ceo summer 2019 he was telling me about the research that we're doing in biology and i was telling him about the research we're doing in ai and notably in reinforcement learning those trial and error techniques and we were like maybe we should start a project together in late 2019 we were working already on personalized cancer vaccines. So methods like using RNA to develop like a vaccine, but just for a patient that has cancer. And then indeed COVID and the pandemic arrived.

13:31Ed Butler:And I guess then the race is on amongst every big pharma company in the world to produce the vaccine for COVID. And you became part of that process.

13:42Karim Beguir:So we were working with biotech. We were not directly part of the team that designed the vaccine. The COVID vaccine was developed using like classical biotechnology methods. But then you had this explosion of variants. You would have like tens of thousands of different variants sequence every week. And I was like, no human can process that much amount of data. So we need to find out something. Like if you would get the amino acid sequence of that particular variant, the question was then, could you guess just based on the sequence without any further testing? Would this be a dangerous immune escaping variant or would be like a classic one which would not warrant further attention?

14:32Karim Beguir:Fizzle out and disappear. Exactly. And this was an important question. Apparently, we were the first team or one of the very first teams in the world to deploy large language model technology, LLM technology. So transformers, attention based models to actually the COVID variants, helping us prescreen with AI the sequences, the variants of interest.

14:56Ed Butler:So this is an early warning system. It's an early warning system for future variants that could be dangerous. People may remember the Delta variant. It was very dangerous or it seemed to be in the first months, right?

15:09Karim Beguir:Yes, Omicron and the like.

15:10Ed Butler:Omicron as well. So these were the variants that were cropping up at that time. And you were able to kind of catch them or at least flag them as they were approaching, which allowed people who were developing vaccines, right, to adapt.

15:25Karim Beguir:Exactly. And the important thing was to catch early signal before anything else. The goal here is to use it as an early filter to sort of like guide you towards what needs to be tested. You still need to test. You still need to validate in the lab. But having that first filter is very useful. For example, Omicron, the moment we got the sequence, we immediately flagged that this was the most immune escaping variant we had ever seen so far. And so AI turned out to be actually a powerful tool in the fight against the pandemic. And even though it's a modest contribution in the end, nothing like what has been achieved with creating the vaccine, I felt this was exactly the kind of things that InstaDeep was about.

16:18Karim Beguir:It's literally build AI that benefits everyone. So we are very excited to have been able to achieve that.

16:32Ed Butler:The best B2B marketing gets wasted on the wrong people. So when you want to reach the right professionals, use LinkedIn ads. LinkedIn has grown to a network of over 1 billion professionals, including 130 million decision makers. And that's where it stands apart from other ad buys. You can target your buyers by job title, industry, company, role, seniority, skills, company revenue so you can stop wasting budget on the wrong audience. It's why LinkedIn ads generates the highest B2B return on ad spend of major ad networks. Spend$250 on your first campaign on LinkedIn ads and get$250 credit for the next one.