It would be such a beautiful story: The man from the Tunisian desert town in whom a hidden, world-changing power slumbers like Luke Skywalker, the hero of the first “Star Wars” films.
La porte du désert, the gateway to the desert, is what the French called Tataouine, and yes, it’s the city where George Lucas filmed parts of the film series. Which is why the desert planet in the film is also called Tatooine. Unfortunately, the story is not entirely true. Karim Beguir grew up in the desert town of Tataouine. But he is half French and studied in France and the USA. Started a normal career in big companies. But then it broke through, the hidden power. In 2014, Beguir decided to give space to his real passion: applied mathematics, artificial intelligence. And he made it his mission to leverage the potential that is in the people of the African continent.
“The two of us started in Tunis with nothing more than $2,000 and two laptops,” says Beguir. “When we said we wanted to found a company here that would do just as exciting things as the Google subsidiary Deepmind, a lot of people just didn’t believe it.” With his skills, Beguir could easily have gone to Silicon Valley, but in addition to his passion for artificial intelligence, he also felt an obligation to take care of the people who usually don’t get the opportunities he does. So Tunis. And it still worked.
Today, his company, Instadeep, is on par with Deepmind, the London company best known for having computers teach themselves the infinitely complicated board game Go. Instadeep is now even cooperating with Deepmind.
What happened like this:
“We met in South Africa,” says Beguir, “the Deepmind people had a problem that they couldn’t solve. We looked at it and six months later we wrote a paper about it and introduced it to Deepmind.”
Apparently they were very impressed. The two companies jointly published a paper that was well received in the research community. “For the first time ever, achievements of a start-up founded in Africa have been recognized at leading AI conferences worldwide,” says Beguir proudly. Today, deepmind parent Google is one of Instadeep’s investors. Beguir, who appears calm and considered during the video call, but also focused, has taken the company a long way in just a few years. Instadeep raised a total of $100 million in the most recent round of investors this January. The company’s headquarters have meanwhile been relocated to London, there are still offices in Tunis, but also in Lagos, Nigeria, in Cape Town, Dubai, Paris – and soon also in Berlin.
Two important customers in Germany
This not only has to do with the fact that Instadeep has two important customers in Germany. Deutsche Bahn is working with the AI experts to use their network more flexibly. And the biotech star Biontech, which rose to prominence in the Corona crisis, has developed a process together with Instadeep to discover new, potentially dangerous variants of the corona virus at an early stage. “Nobody had done that back then,” says Beguir. From the gene sequence of a virus alone, the system recognizes with a high hit rate whether it is a potentially dangerous variant.
The fact that the currently predominant omicron variant is a high-risk variant was discovered on the same day that the gene sequence was published.
The long-term partnership with the Mainz-based company should also help to develop new immunotherapies. Sequencing the human genome is getting cheaper and cheaper, says Beguir. From the data obtained in this way, many insights could be gained in the future. In an AI laboratory, things like protein design, but also the optimization of operational processes should be advanced. Deutsche Bahn and Biontech are also among the investors in Instadeep.
In contrast to Deepmind, which tends to pursue research projects, Instadeep is about helping industrial customers to solve specific problems. This is how it looks at Deutsche Bahn: Their approximately 33,400-kilometer route network is branched and complex. New routes are very difficult to implement and they would also be very expensive. So the focus must be on using the routes that already exist more efficiently. But, says Karim Beguir, the project is also about getting decision support if, for example, a route is blocked by a defective train. “There has never been an AI timetable system for a route network of this size,” says Beguir.
Supercomputers have to help
The role of Instadeep is divided into two: On the one hand, the team has a lot of experience with supercomputers – and this particularly fast hardware is also needed to quickly solve complex problems such as that of the rail network. The other important part is the applied mathematics, which are algorithms developed using machine learning.
“It’s a bit like Deepmind did with chess and go,” says Beguir. “The beauty of it was that the system never saw a game played by a grandmaster, rather it learned by playing against itself.”
Instadeep now wants to do the same with the railways.
The experts simply feed the system with data from real operations and let it learn by simulating the processes. Once the system has learned how the system works, traffic “can be diverted much more efficiently within minutes”. Deutsche Bahn is getting a supercomputer from Instadeep and has entered into a long-term partnership with the start-up.
Fear of AI
When it comes to rail transport, few people will probably object if artificial intelligence is used to organize it more efficiently. But in other areas there is definitely a fear that the automated systems could put minorities at a disadvantage, for example. Many are also bothered by the fact that ultimately one does not know exactly how the findings of the self-learning technology actually come about.
Beguir finds this not only understandable, but also correct that critical questions are asked. Ultimately, the influence of artificial intelligence will be even greater than that of the Internet, both economically and socially.
“So we at Instadeep work by proposing a potentially good solution. But the application should be done with a different system.” In the case of the railway, for example, the system does not control any train independently. Control is taken over by a completely independent system that does not work with AI and is also monitored by human experts.
“We can find solutions to problems where the amount of data is larger than there are atoms in the universe, but the proposed solution should always be reconsidered and accepted with reservations.”