Matteo Manica, mathematician: "AI gives us superpowers; it's up to us to choose how to use them."

They say mathematics is the servant and handmaiden of all disciplines. And it's true. There's no science that can do without it. A wonderful discipline, mathematics can change the world, but on its own, without the other disciplines, it risks being abstract. With artificial intelligence, for example, it can be applied to scientific discoveries and industrial problems. That's what I do.

Matteo Manica is in love with his subject, mathematics. As long as it's applied. He's a Senior Research Scientist at IBM Research in Zurich, a hidden gem in the heart of Europe where Nobel Prize and Turing Award winners work. the most advanced technologies are born, from artificial intelligence to quantum computing.

Manica builds AI models to improve industrial processes, from generating new, more sustainable materials to modeling new chemical formulations.

He studied mathematical engineering at the Polytechnic University of Milan, and from the beginning chose to work in the space where theory meets application. "Mathematics is a necessary condition. It holds every discipline together, but on its own it's not enough. We need other tools, other languages ​​to make sense of it."

At IBM Research in Zurich in 2020, Manica and the team he works with created something unprecedented: a chemical laboratory entirely controlled by artificial intelligence. They trained natural language models—models like ChatGPT, Claude, Granite, and others—with millions of organic chemistry patents, building a system capable of autonomously creating new molecules.

It works like this: you draw a molecule on a screen, and specialized language models tell you how to build it, which reagents to use, and which steps to follow. Then the procedure is transferred to a robot. And that robot actually carries it out. No handwritten formulas, no equations. Just a model that can read and write the language of chemistry. We treat molecules as sequences. It's as if the model were translating a sentence from one language to another: from the final product to the starting ingredients.

Today, Manica creates models for producing more sustainable products. "We have collaborated and continue to work with large companies to develop technologies that support the creation of more sustainable materials and processes: food packaging, chemical formulations, and battery lifecycle modeling to improve durability and efficiency.

A 37-year-old from Novara, he studied science at a high school, then decided on a major. "I was torn between philosophy, mathematics, and medicine because my parents are doctors, which is why I took the test. Then, thanks to a professor's advice, I chose mathematical engineering at the Polytechnic University of Milan." He spent three years at a spinoff doing numerical simulations of systems. "I worked on models for volcanology, models for printing processes, and models for blood flow in arteries." Then he returned to school. He earned a doctorate at ETH Zurich, sponsored by IBM. "I did things similar to what I do today, but more applied to biology or computational biology." There, he began using statistical and machine learning models, designed for data-intensive contexts. "We took these models, which typically find applications in fields where there's a lot of easily available data, and tried to apply them to contexts, like biological ones, where you don't have ready-to-use data in the initial phases. The impact is incredible: they can describe very complex systems without necessarily having to impose rules or equations."

For a mathematician like him, accustomed to building models from formulas, it was a turning point. "Having studied mathematics, I've always created models from equations." But there he began to change his perspective. "We've moved from the old school, almost Newtonian, approach to science to a more data-driven approach."

He uses a simple example to explain: "The fall of a heavy object. You see it fall and at a certain point you say: it seems like there's a constant, which is the acceleration, and that depends on the mass. So you create your own little system that says: the applied force should be a constant, g, multiplied by a variable: the mass."

As long as the system is simple, it works. But as soon as complexity increases, the paradigm shifts, and you can't model everything.

Hence the shift: "We've gone from a deterministic approach, where you assume you know everything, to a system where you suspend judgment and say: I don't know exactly how many variables there are, I want the data to speak to me."

Most projects are conducted with an open source approach, which is what IBM believes in because collaborative development within the open source community yields the best results. "We do something very similar to what can happen in a university environment."

A wonderful place for scientists.

The first thing that makes it special is its location, because we're close to the ETH Zurich, which I think is among the best universities in Europe. Not only that, we also have the opportunity to interact with other top-level academic institutions. And then there's a super vibrant research community, featuring collaborations with companies like Microsoft, Apple, Google, Nvidia, Anthropic, and OpenAI.

And again: it's a historic institution. It was the first laboratory IBM opened outside the United States, in 1956, and when you enter the Think Lab, you breathe an air thick with excellence. You can meet experts in AI, cryptography, physics, quantum computing, and molecular simulations. Some Nobel Prize winners worked here. You go to the cafeteria and meet more than 300 scientists with whom you can exchange ideas, and from this continuous cross-fertilization, projects capable of changing the rules of the game can arise.

Matteo Manica sees his future here.

I hope to find new applications of mathematics in contexts that truly have an impact. The transition to more sustainable industries, for example, is one of the most important challenges. I'm thinking of consumer goods, the materials we use every day: if you manage to optimize even just one production process, the impact on the environment is enormous. These are invisible, yet profound changes. They may not make headlines like a new drug, but they can improve the world we live in. They do it quietly, for everyone.

A scientist and musician, Manica maintains that mathematics also plays a part in his passion for music. "I spent a lot of time playing; I still play the electric bass and double bass at a decent level. But I've also played the flute and alto saxophone. And music has a lot of mathematics behind it: the vibrating string, the equations that describe tuning, the evolution of harmony, Schoenberg's twelve-tone systems." It's always mathematics that sparks curiosity.

"It sparked my interest in things. And if mathematics is the servant and handmaiden of all the sciences, the bass plays the same role in music: without it, nothing works, and on its own, it means nothing."

What have you learned in your career that can be useful to all of us? "The biggest lesson is that math isn't difficult. Maybe we just need to rethink how we teach it. I don't have a recipe, but I know it's much easier than it seems. You just have to not be afraid. It takes patience. It takes practice, like everything. There's nothing special about what I've done. You make a lot of mistakes, but if you persist, if you don't let the complexity paralyze you, one step at a time you can get anywhere."

Even getting into a research institute like IBM Research isn't difficult for Manica. "Honors, awards, or what's on a resume don't matter. When we hire someone, all of that is completely secondary. We sit down in front of a whiteboard or a screen, write down a problem, and try to solve it together. That's when you understand how a person thinks."

Will artificial intelligence change the way we do science? Will it even lead to writing a scientific paper and making a discovery? "It will change the way we do science, yes. Many things today are already discovered or suggested by the models themselves. But we need to define what it means to discover. I'm not anxious about this. Maybe AI will write a 10-page scientific paper and do it better than us, but without our three bullet points describing an idea, it can't do anything. It will have a huge impact on many areas of industry and society. We need to use it correctly. At IBM, we use models sized for a specific problem; they aren't excessively expensive or power-hungry, and they're specialized. Most of them can practically be used on your computer. AI will give us researchers superpowers. We'll be like augmented researchers..." And perhaps, in the silence of the laboratories, we'll start asking ourselves new questions.