I’m not afraid of the machines taking over. I worry that we might fail to harness their capabilities in time

A maths champion turned internationally renowned researcher, Lenka Zdeborová, currently living and working in Switzerland, fuses mathematics with physics and computer science to give us humans a better understanding of intelligent machines.

Read the story in Czech translation here

Lenka Zdeborová’s work traverses the boundaries between physics, mathematics, and computer science – there and back again. The researcher connects and advances these three fields using knowledge and experience gained in each. Nonetheless, she still says of herself: “I’m a theoretical physicist. What interests me is how things work.”

She studies machine learning and neural networks. Along with colleagues from all over the world, she is trying to gain a deeper understanding of the functioning of algorithms that autonomously learn from available data. Or, more precisely, to understand under what circumstances they work – and under what circumstances they do not.

Tools that use neural networks and machine learning are already at our fingertips – in our phones. These include, for instance, the translation services Google Translator and DeepL. Over recent years, these services have improved to the point that their translations (after some slight revision) are fairly reliable for everyday communication, whereas just ten or fifteen years ago they were only reliable as dispensers of absurdist language jokes and nonsense poetry.

Before it stayed up, a Roman bridge might collapse three times. It’s much the same for us.

“It’s amazing. Yet programming these things is not that hard: we’re teaching our uni students how to do it. The problem is that we still don’t really know when machine learning will work and when it won’t. We still can’t say under what circumstances a neural network will actually start learning on its own. It’s done through trial and error,” says Lenka Zdeborová, pulling back the curtain on the practical issues, before offering an analogy with ancient engineers.

“When the ancient Romans were building a bridge, they relied on hit and miss. Three times, perhaps, it’d collapse on them, and the fourth time it’d stay up. But they didn’t know why. As a species, we’ve gradually learned a lot from these trials and errors and, thanks to mathematics and statics, we now know exactly how to build a bridge so that it doesn’t collapse.”

With respect to machine learning and neural networks, says Lenka Zdeborová, today we are at same point where ancient Romans were in terms of bridge building.

“The engineers are doing the programming, and they’re getting a lot of the stuff right, but we’re still unable to predict how a machine will actually behave, whether it will learn to solve very complex problems,” she says. “Once we’ve understood how exactly machine learning works, we’ll be able to improve it, perfect it, so that it can be used for decision-making in much more sensitive matters, such as who to hire or not to hire for a job, or to enrol at a university, on whom it makes sense to operate and on whom it doesn’t. Or in legal cases.”

That is a possible future. And it is also the main research topic for the team around Lenka Zdeborová. Machines will deal with increasingly complex tasks for us. The important thing, however, is not just the outcome itself, but also our ability to describe the way it was produced.

Progress in machine learning also raises some fascinating ethical issues. But we will get to those in a while.

Spoiled for a choice of ideas

Lenka Zdeborová and her family – her partner and two daughters – live in Switzerland. Both she and her husband, who is involved in the same field, have obtained professorships at the Lausanne Institute of Technology.

“I’m not sure whether us both doing the same thing is an advantage or a disadvantage in getting a job. We understand and complement each other both at home and at work. But getting an offer for two identically oriented professorships? Well, that isn’t so easy.” Nonetheless, such an offer did come from Lausanne, and the researcher couple left France and moved to the lakeside campus below the Alps.

How do they complement each other as research partners? “That changes over time, but when we started out, my husband always had a whole lot of ideas. An awful lot. But he wasn’t sure which to tackle first. In contrast, I was the one who’d fish in his torrent of ideas and haul out those we’d pursue. And it worked,” Lenka Zdeborová smiles.

A maths-contest hot dog

She graduated from the Faculty of Mathematics and Physics at Charles University Prague, where she obtained her PhD. Her second doctorate is from Paris-Sud University. She received her habilitation in Paris, too, at the École normale supérieure. Her next career step was a research fellowship in Los Alamos, New Mexico, a place particularly well-known for the development of the nuclear bomb. It is the home of the National Laboratory, ranking among leading technological institutes in the US.

Until recently, she and her family lived in France, where she was employed at the Centre national de la recherche scientifique (CNRS), the French equivalent to the Czech Academy of Sciences. In 2016, the European Research Council awarded her a prestigious ERC grant. Two years ago, the family moved to Switzerland.

It was her elementary-school teachers in Přeštice, a town near the city of Pilsen, who discovered her talent for mathematics. While in the second stage of elementary school, she quickly became a champion at mathematics and later also physics competitions. Following her success in regional and national contests, her career as a maths competitor peaked when she won a bronze medal at an international event in Padua, Italy.

“I still remember the ride from the first district tournament, which I won when I was in year 5. My teacher bought me a hot dog – the very first hot dog I’d eaten,” she smiles, reminiscing about her childhood. She was drawn in by the world of mathematics and physics as young pupil and her fascination endured at grammar school, too, so her choice of university was a foregone conclusion.

The more difficult it is, the more…

She often says that the more difficult an assignment was, the more interest it held for her. “The harder the problem, the better. As a kid I was hugely motivated to find solutions to the most complex maths problems. But those were maths competition problems that had model solutions. This changes as you grow up and become a junior researcher. Now you’re faced with questions that nobody can answer, and going for the most complex isn’t always the best idea.”

Her attitude changed when she was in search of a topic for her MA thesis at the Faculty of Mathematics and Physics. She had to select a research project. “I found a friendly supervisor with interesting topics. He offered me three to choose from, ranking them by complexity.”

She automatically picked the most complex, of course.

“But then we talked it over. He told me he’d given it some thought and that they had a grant for the first, the easiest, topic and that they were part of a European network which was working on it. And that he could use a motivated student.”

“My mind was going: ‘Well, what do I know, he might be right…’ Eventually I said yes. And it was one of the best decisions I’ve ever made,” she adds, and for a moment our talk turns into a meditation on life’s crossroads, where our entire destiny is decided by choosing which path to take.

The issue in question was so-called spin glass. “Everything became incredibly dynamic. Before I knew it, I was leaving for my first big conference. By plane, no less! It was the first time I’d ever been on one. And I met pretty much the whole community there, including the Nobel laureate Giorgio Parisi.”

And including her future partner, the French researcher Florent Krzakala, with whom she gradually intertwined both career and life.

Mothballs and frustration

She began maturing as a researcher. Also, she gradually came to realise how important it is to balance the intricacy and attraction of a given problem against the available knowledge, resources, and time she can use to solve it.

She also considers asking the right question and choosing a problem well to be the cornerstone on which to build resilience against frustration.

“In research, you have to get used to the fact that there are questions you can’t answer. Or that you can’t do it in that particular moment with the resources you happen to have,” she explains. “I must be pragmatic. I must know what to focus on and quickly decide what to mothball – and perhaps take up again later. In a month. A year. Five years. A researcher has to be resilient against frustration and failure. But then you can’t put everything in mothballs, either,” she laughs.

Put simply: not giving up too early, but still being able to give up in time. “If I wallowed in frustration that things weren’t working out, that there was no point, and that I wasn’t making any headway, I wouldn’t be able to do research at all.”

More importantly still, she regards this particular ability as a prerequisite for rising to the highest ranks of academia. “As long as you’re a junior researcher, there’s always someone above you who bears more responsibility than you do. Gradually you mature to become more autonomous in your work, and especially in your decision-making. And once you’re aspiring for a professorship, you have to be able to decide – soundly, if possible, and without help – which problems to tackle as a research team. What lies within your power and what doesn’t.”

American questions, European answers

It was during her fellowship in Los Alamos that she honed the art of posing the right questions. “In Europe, the emphasis is on how to answer questions, rather than how to ask them. In America, you learn what question to ask in the first place,” she describes the different attitudes. “I think, however, that here in Europe we have more capacity to answer the right question comprehensively.”

So, it’s off to the New World for questions and then straight back to the Old World in search for answers? “That’s right, more or less. At least for me, that is. European-style education also stresses ways of making sure your solution is scientifically sound.”

One funeral at a time

She has comprehensive experience with the system of research in France, and now in Switzerland. How does Czech research compare within the European context?

“Some Czech research institutions are firmly rooted in the twenty-first century, in up-to-date European research. But generally speaking, I’d say that many of the differences are still underlain by the legacy of communism,” she reflects.

“For decades, posts and privileges depended on one’s Party membership, conformity, and compliance with the rules of totalitarian politics. And I still see some degree of inertia in this respect. Some institutions have been held back by this until quite recently. And they’re moving forward one person at a time. One funeral at a time, as the famous physicist Max Planck once joked.”

That sounds cynical. “But there are places where it’s literally true. It isn’t until someone passes on that their place can be taken up by someone younger, more open, with foreign experience, and so on.”

Not that she believes people in leadership positions are incompetent, like the proverbial dogs in the manger. “These may be researchers with incredible achievements, groundbreaking discoveries, to their name, people who, in their time, were at the cutting edge, but experience teaches us that even such researchers, once they’re nearing the end of their careers, pull those around them backwards, rather than drive them on.”

Her interaction with Czech science and research was very limited for several years, but recently she has been networking and reconnecting with fellow Czech researchers and institutions, and strengthening their relationships.

“This is mainly due to the active and enthusiastic people at Czexpats in Science, who connect us, hold meetings, conferences – theirs is a wonderful initiative. And I’m also obliged to the Neuron Endowment Fund. I’m grateful for the prize they awarded me last year, but mainly I appreciate them helping me find my way back to the Czech Republic. I’ve met many interesting people through Neuron, I’m feeling part of the Czech research scene again, at least a little.”

Lest humans interfere…

But let us come back to algorithms. To learning machines. The increasing ‘autonomy’ of some systems goes hand-in-hand with ever more heated discussions on related ethical issues. The most glaring (at least in the public sphere) is the debate about the future of autonomous cars.

This issue has spawned scientific papers as well as entire books. What we have to do as a society is to equip such vehicles in advance with the ability to decide how to act in the (admittedly extreme) scenario when a car, let’s say with children on board, is in danger of colliding with a lorry, but when any evasive action would put pedestrians in mortal danger. And many similar such moral conundrums.

She herself is rooting for autonomous cars. “They have incredible potential. I’m myself a driver, but a reluctant one; I know I’m easily distracted, I’m afraid I’ll do something stupid – and so might the driver coming the other way. It’s awfully dangerous,” she observes. “When you think about it in reverse, you see how convenient it’d be to let the machine do the driving.”

“If a person causes an accident, they may die themselves or cause injury to others, or it may all end well and they ‘just’ lose their licence. But what will they and the other drivers have learned? Hardly anything. In contrast, it’s precisely these kinds of mistakes that an autonomous system learns from, eliminating them in the process. I believe that once we’ve handed the keys to autonomous cars, traffic accidents will be a thing of the past. If vehicles drive the way they’re supposed to and all the kinks and errors have been ironed out, there’s no reason for accidents to occur.”

Will they skip or not?

“Yes, some issues will have to be addressed in advance. But setting those to one side, machines will learn by themselves from the data, from the huge amounts of data that humans definitely don’t have to hand, and which, unlike a machine, they can’t really use to inform their decisions. In the US, for instance, they’ve already begun discussing the use of an automated system in courts for deciding whether defendants will be granted bail,” she suggests by way of example.

The algorithms evaluate past cases and situations, so that any decision concerning bail (based on data from previous years) minimises the risk that a suspect released on bail commits another offence or absconds.

“In much the same way, machine learning is already routinely deciding for us when we apply for a loan, for instance, or take out an insurance policy. You fill out a questionnaire which the machines link to all the data they can find about you. And they cross-check all the risks, advantages, and disadvantages. It’s actually very similar to the way Netflix makes suggestions for TV shows for you to check out, based on what you watched last time.”

She goes back to the issue raised about trust in machines. “In healthcare, for instance, it’s still up to the doctor to decide whether to operate on someone. Or not to, because the benefits of surgery don’t outweigh the potential risks. Very soon, such assessments will be handled far more competently by a machine, but first we must learn to trust it. And we also have to make sure we know who’ll be held responsible in case of failure,” the researcher explains.

I’m not afraid of a robot revolt

From her words, one might infer that she is not afraid of the machines taking over. “Absolutely not. Quite the contrary, what I worry about is that we’ll fail to make timely use of the capabilities AI is giving us or will give us in the near future. All around we keep seeing examples where we have the data, have the algorithms that can correctly evaluate it, but we don’t act on it. We saw that during the pandemic. And we’re also seeing that with global climate change, for instance.”

We know this Titanic of ours is on a collision course with an iceberg… We have the facts, the data, and we know what we should be doing. But steering the Titanic out of harm’s way is taking an age (although, in a climate crisis, the danger comes from the iceberg having time to melt).

“We have solutions to lots of problems, but we don’t know how to implement them in a socially acceptable way. Take vaccines during COVID: in terms of science, undeniable protection against severe complications. But what’s the vaccination rate?” she asks rhetorically.

Dynamics of an epidemic vs. the unpredictability of war

She herself confronted difficulties when trying to put an existing solution into practice during COVID. As early as 2015, she had co-authored a paper proposing an algorithm suitable for tracing systems (during a then-hypothetical health crisis).

“When the epidemic started, we thought we might try and apply our algorithm to mobile phones. We weren’t alone in this, there were other efforts like ours. The issue certainly wasn’t new, but the urgency was… Before, we’d been exploring a hypothetical scenario, but suddenly a practical situation arose that cried out for implementation. But we ran up against the brick wall of protecting privacy.”

Although aware of the high value our society attaches to privacy, she was nonetheless surprised by how quickly any idea of implementation was scrapped.

“At least we could’ve had it up and ready. At the beginning, there was no way of knowing how much trouble we were in for. And it’s still possible – and I don’t mean to jinx us – that a much worse epidemic is heading our way, and only then will these walls start to break down. So, we have a post-doc who’s continuing the research into tracing algorithms. Something in reserve, just in case,” she promises.

An epidemic has many variables that forecasters must account for, but we can always apply the data and the scientific method. She agrees with me that Russian aggression, or more precisely its war on Ukraine, is much less predictable in this regard, driven as it is by a dictator with psychopathic traits.

“Alas, I’ve no special algorithm up my sleeve that can predict what will happen next. We don’t know what Putin will use, how he’ll act,” she shrugs.

“Yet in a war there’s at least a culprit you can point your finger at. But how about the climate crisis? The responsibility lies with all of us, with everyone on the planet. In theory, we know the solutions, but then a huge mass of people will have to do without their comforts, without their profits. That frustrates me all the more, since we have the situation in our grasp, but there’s no actual solution in sight.”

Being a mother of eight-year-old and eleven-year-old daughters, she often asks herself what we are leaving to our children.

“It really weighs on my mind. We live below the Alps. We go up to a glacier in the summer, one that’s conveniently accessible, and it’s great because you can hike there with kids without any special gear. It’s two glaciers, in fact, that were always joined together. But this year, for the first time, that join melted. You could walk between them. Walk on ground that no human foot has stepped on for tens of thousands of years, until now,” she says sadly.

“In a valley beyond the lake we can also see the Mer de Glace glacier, which translates as ‘Sea of Ice’… A century ago, a railway was built that took you to a place where there used to be the mouth of an ice cave you could enter. But the glacier has melted so much that the cave mouth is now hundreds of meters lower, so that you have to climb down from the train stop to reach it. Going down, you can see dates on the rock marking how far the glacier had sunk in a given year. It’s heartbreaking to see the melt in real time.”

She often discusses climate issues with fellow researchers at the university. “We have lots of sustainability projects. We have funds for research. We have ideas and possible solutions, but it’s a little depressing that the solutions aren’t and probably won’t become politically, economically, and socially feasible.”

Liberty, equality…

We have already commented on one of the many negatives of the communist legacy, but Lenka Zdeborová points out one positive aspect, too.

“One of the few things that, in retrospect, strikes me as good was a kind of egalitarianism at schools. Unfortunately, we’re quickly catching up with the West when it comes to the gap opening between different strata of society, starting already in kindergarten,” she opines.

“Children from socially disadvantaged groups usually don’t get to do research, even though there are many among them who would otherwise be smart enough.”

Maths offers freedom, but the Czech Republic is lagging behind

When she recalls her childhood and her early days as a mathematician, we inevitably arrive at the issue of the position of women in society and science. As a girl with a knack for maths, she herself did not encounter obstacles. “In a maths competition it didn’t matter whether you were a girl or a boy, whether your dad was rich, whether he was a Party member or not, none of that would have solved the maths problems for you. And I don’t think that’s changed. In this sense, maths was – and is – free.” On the other hand, social conditions and attitudes may perhaps discourage some girls from casting their lot with maths or physics.

“In this respect, the Czech Republic is definitely lagging behind. In the US, equality is emphasised much more, so much so, at times, that you worry you might’ve said something that will come back to haunt you, that you might be accused of discrimination. But in France, for instance, these matters are arranged really well; there I always had the feeling that the balance works, that it was even natural somehow.”

Heartless mothers

In her opinion, the French approach to equality is reflected in, among other things, the attitude to and social conditions for women as mothers. “How you handle motherhood is your own personal business. Everyone can pick the model that suits them. There are plenty of options available. You make your choice and others respect it. Nobody looked down their nose at me for hiring a nanny for my girls at five months. And neither does anybody judge a woman for staying at home with the kids for as long as she wants,” she relates her personal experience.

The social climate in the Czech Republic was a shock for her. “I’d meet my ex-schoolmates, and when we started talking I was shocked by how they’d comment on me not staying at home until each of my daughters was three,” she recalls. “Asking me whether the big bucks I was getting at work were worth handing my kid over to a nanny and not seeing it grow up.”

She had little urge to explain that, quite the opposite, almost all the “big bucks” she was earning at work were being spent on the nanny. Yet her classmates saw her as a career woman, trading her children for gold.

“It was funny, actually. In France, researchers aren’t that well paid. Me and my partner sat down and calculated the cost of the nanny compared to possible welfare benefits and our income, and it turned out that if I’d stayed at home, we’d have been better off financially,” she describes.

Yet like many other women who love their job and do not want to miss the boat – something that, to a researcher, can happen relatively fast – she, in fact, invested her salary into the opportunity to continue her work.

“I didn’t want to break off my research mainly because I’d have missed the sense of personal fulfilment and lost touch with colleagues and students. And I didn’t want to muddle along with a baby in tow: looking after a small child is a full-time job. Instead, I preferred to make most of the time I did have with my baby girls,” she describes an experience similar to that of many other working mothers. Rather than being a frustrated mum available to her child around the clock, better to come home to it happy from work.

“We had this incredibly good and kind babysitter. She looked after my daughters for six years, she was their surrogate grandma, and she’s in our lives to this day. She was incredibly kind. Her cooking was so delicious that I envied my daughters, she was there for them more than I could ever have been, even if I’d looked after them day in, day out. In France this raised no eyebrows, it was just from Czechs that it drew a plenty of comments.”

Czechs are outstanding mushroom hunters 

There is, however, one Czech (or rather Eastern European) peculiarity she is proud of, and she zestfully pursued this pastime in France, Switzerland, and America, regardless of being labelled an eccentric. Mushrooming.

“I’d only meet Russians and other expats or migrants from Eastern Europe in the woods,” she chuckles. “If a French or a Swiss is picking mushrooms in the forest, it’s nothing but boletes or chanterelles. Everyone was astonished at what I dared basket. My kids love breaded parasol mushrooms and my partner has a soft spot for fried blushers.”

She is not planning to return to the Czech Republic (yet), but rather than losing touch with her original homeland, she is trying to strengthen her connections here as much as possible. And her daughters have been given Czech names: Alenka and Julinka.