About

Small

Sune Lehmann is a Professor of Networks and Complexity Science at the Technical University of Denmark (DTU) and a Professor of Social Data Science at the University of Copenhagen. A physicist by training, his research uses complex systems, machine learning, and network science to study human behavior at scale – from how we move through cities to how information spreads online to how sequences of life events can be predicted from large-scale data. He is a member of the Royal Danish Academy of Sciences and Letters, a Chief Scientist in the Danish National Centre for AI in Society (CAISA), and a recipient of Denmark’s EliteForsk Senior Prize.

Medium

Sune Lehmann is a Professor of Networks and Complexity Science at DTU Compute, the Technical University of Denmark, and a Professor of Social Data Science at the Center for Social Data Science (SODAS), University of Copenhagen. A physicist by training – M.Sc. from the Niels Bohr Institute and a Ph.D. from DTU on the structure of complex networks – his work sits at the intersection of physics, computer science, and the social sciences, using massive datasets and methods from complex systems to study how people behave, move, connect, and influence one another.

After his doctorate, Lehmann spent three years in Boston as a postdoctoral fellow, splitting time between Albert-László Barabási’s Center for Complex Network Research at Northeastern University, the Center for Cancer Systems Biology at Harvard’s Dana-Farber Cancer Institute, and Harvard’s Institute for Quantitative Social Science. He returned to DTU in 2010 – promoted to Associate Professor in 2012 and Full Professor in 2019 – and has maintained near-yearly research visits to Northeastern ever since. Since 2020 he has also held a joint professorship at the University of Copenhagen.

Lehmann’s work spans community detection in complex networks (Nature, 2010), human mobility (Nature Human Behaviour, 2025; Nature, 2020; Nature Human Behaviour, 2018), the accelerating dynamics of collective attention (Nature Communications, 2019), complex contagion of information online (PLOS One, 2017; Best Paper Award, IC2S2 2018), and most recently the prediction of life trajectories from population-scale registry data using transformer models (life2vec; Nature Computational Science, 2024). From 2012 to 2017 he carried the operational leadership of the Copenhagen Networks Study – one of the most ambitious social-sensing experiments in Europe – alongside the scientific leadership of his Villum Young Investigator grant; the resulting dataset has underpinned dozens of publications across the group. With Laura Alessandretti, he co-leads the Social Complexity Lab at DTU, which received the Young Academy’s prize for Best Danish Research Environment in 2023.

He is a member of the Royal Danish Academy of Sciences and Letters (elected 2024) and the Danish Academy of Technical Sciences (2021), a recipient of the EliteForsk Senior Prize (2022) and the Columbus Prize (2021), and a Chief Scientist in the Danish National Centre for AI in Society (CAISA). He served on the Danish government’s COVID-19 modelling task force, the advisory board for the national Smittestop contact-tracing app, and the government’s expert group on tech giants; he is a guest editor for PNAS. He has supervised 13 Ph.D. students as primary supervisor, 14 postdocs, and over 170 Master’s projects, and has been nominated for best teacher at DTU four times.

Large

I’m a physicist who ended up studying people.

That sentence has been my elevator pitch for years, and it’s still the most honest thing I know how to say in one line. The longer version – what I’m a professor of, where, and what I work on – is below, but rather than tell it as a life story in chronological order, I’ve broken it into the parts of the job that actually take up the days. Read in any order. Skip what doesn’t interest you.

Where the work happens

I hold two professorships, both granted on the assumption that I would do more or less the same kind of work in two different places, and so far that has turned out to be roughly true. At DTU, I’m Professor of Networks and Complexity Science at DTU Compute – DTU’s home for applied mathematics and computer science – where I’ve been on the faculty since 2010 and a full professor since 2019. At the University of Copenhagen, I’m Professor of Data Science at the Center for Social Data Science (SODAS).

Since 2025 I have also been a Chief Scientist in the Danish National Centre for AI in Society (CAISA) – the national centre for research on what AI does to societies and what societies should do about AI – and since 2022 I’ve co-led the Networks and Graphs collaboratory at the Pioneer Center for AI. I was elected to the Royal Danish Academy of Sciences and Letters in 2024, joined the Danish Academy of Technical Sciences in 2021, and was a member of the Young Academy from 2012 to 2017.

In September 2025 I became Efor of Valkendorfs Kollegium – the academic overseer of one of the four historic student “kollegier” (student dorms / colleges-in-the-UK tradition) of the University of Copenhagen. Valkendorfs, founded in 1589 by royal treasurer Christoffer Valkendorf, is the oldest student kollegium in the Nordic region; past residents include famous Danes, such as Grundtvig, Ewald, and Herman Bang. The role is somewhat ceremonial today, but because Valkendorf is run by its inhabitants, I do get to help out once in a while when they need an old fart to weigh in on complex matters. And the job puts me in touch with a group of amazing young university students who teach me as much as I teach them.

What I study

My research is about finding patterns in large-scale data about human behavior. I started in network science – the mathematics of connections – and gradually moved toward using networks as a lens for understanding social systems. The list of topics I’ve worked on is, on the face of it, a bit promiscuous: networks, mobility, sleep, attention, contagion, academic performance, life trajectories, the science of science. But they share a spine. They are all attempts to find structure in the messy traces that people leave behind when they live their lives – and to do that without losing sight of the fact that the traces are, ultimately, people.

Networks and communities

The central question in network science is how to characterize meaningful structure in complex systems – the groups, clusters, and communities that organize a network. My early work was on modularity optimization and biclique communities, but the paper that broke through was the 2010 Nature paper with Yong-Yeol Ahn and Jim Bagrow introducing the concept of link communities. The key insight is simple. In real networks, communities overlap pervasively: every person belongs to multiple groups – family, colleagues, college friends, neighbors. Traditional methods force each node into a single cluster and miss this. Our trick was to assign the links, not the nodes, to communities. Your connection to your sister belongs to the “family” cluster; your connection to your colleague belongs to “work.” But you belong to both. The algorithm has since been used across biology, social science, and information science. I think it worked because it captures something true about how the social world is organized.

Human mobility

With Laura Alessandretti, Piotr Sapieżyński, Andrea Baronchelli, Marta González, Ulf Aslak, Louis Boucherie, and others, I’ve spent years studying how people move through space. Our 2018 paper in Nature Human Behaviour showed that people maintain a roughly conserved number of regularly visited locations – about 25 – at any given time. When you adopt a new restaurant, gym, or coffee shop, you tend to drop an old one. It’s like a Dunbar’s number for places. The 2020 Nature paper extended this, showing that mobility is organized as a nested hierarchy of spatial scales, from your daily neighborhood routines up through city-level patterns to long-distance travel. Earlier work in the group also pioneered methods for tracking mobility from WiFi signals rather than GPS – showing that you can reconstruct someone’s movements with surprising accuracy just from the access points their phone scans for. The privacy implications of that finding are still working themselves out.

Collective attention

With Philipp Lorenz-Spreen, Bjarke Mønsted, and Philipp Hövel, I studied how the dynamics of public attention have changed over time. Our 2019 Nature Communications paper documented something that many people feel intuitively but that hadn’t been rigorously measured: across multiple cultural domains – Twitter hashtags, Google Books, movie ticket sales, Reddit posts – the window of collective attention has been shrinking. Topics rise faster, peak higher, and disappear sooner. The nine-day wonder that Chaucer wrote about is now a six-hour wonder. The work ended up as part of the discussion around Johann Hari’s Stolen Focus, an experience that taught me a lot about the difference between careful academic discourse and the rough-and-tumble of public debate on Twitter. I still think the finding is important – not because it tells us anything about individual attention spans, but because it tells us something about how our collective focus is structured by the information systems we’ve built.

Complex contagion

How does information actually spread through social networks? The simple model is that it spreads like a disease: one exposure is enough. But social influence is often more complex than that. You might need to hear about something from multiple independent sources before you act on it. With Bjarke Mønsted, Piotr Sapieżyński, and Emilio Ferrara, I ran one of the first carefully controlled experiments on this question, using Twitter bots to expose real users to URLs in a designed setup. We found clear evidence of complex contagion: people were significantly more likely to share content when they’d been exposed to it by multiple sources. The paper (PLOS One, 2017) won the Best Paper Award at IC2S2 2018 and remains one of the cleaner demonstrations of social reinforcement in information spreading. I also co-edited the Springer volume Complex Spreading Phenomena in Social Systems (2018) with Yong-Yeol Ahn.

Sleep and daily rhythms

An unexpected research direction that grew out of the Copenhagen Networks Study was sleep. With Sigga Svala Jónasdóttir and James Bagrow, I’ve studied how sleep patterns vary across gender, age, and context, using data first from instrumented smartphones and later from globally distributed wearable devices. We found, among other things, systematic gender differences in sleep variability across the adult lifespan, and that people naturally compensate for sleep debt while traveling – a hopeful finding for anyone who has ever slept beautifully in a hotel room. This is one of those threads I didn’t plan to pull on; once we had the data, it pulled itself.

Academic performance, or: the right data beats big data

With Andreas Bjerre-Nielsen, Valentin Kassarnig, and David Dreyer Lassen, I pursued a three-paper arc on whether the rich behavioural data we’d collected through the Copenhagen Networks Study could improve predictions of academic performance over what you could get from a transcript. The punchline – which genuinely surprised us – was no. When CNS features were lifted into Statistics Denmark and pitted head-to-head against registry data, the rich behavioural data added essentially nothing on top of past grades. We published the conclusion in PNAS in 2021 as Task-specific information outperforms surveillance-style big data in predictive analytics. The implication, which I find deeply important, is that for many practical applications you should think very carefully about whether you actually need to collect all that behavioural data. The right data often beats big data. It is an argument for privacy, for parsimony, and for thinking carefully about what information is actually relevant to the question you are trying to answer.

Life trajectories and life2vec

My most recent major project, with a key paper led by Germans Savcisens and published in Nature Computational Science in 2024, asked a provocative question: can you treat the sequence of events in a person’s life – education, employment, health, income, relationships – as something like a sentence in a language, and use the same techniques that power large language models to learn its structure? The answer turned out to be yes. We trained a transformer on life-event sequences from Danish national registries, and it learned meaningful representations of human lives, capturing relationships between education, health, social position, and many other dimensions. The model could predict future outcomes, including mortality risk. The project generated an enormous wave of international media coverage – Science, The Washington Post, The Economist, AFP syndication across dozens of countries – and much of that coverage sensationalized the work as a “death predictor,” which is not what it is. We built a clarification page (life2vec.dk) to set the record straight. The work is about the deep structure of human experience, not a tool for deployment. There is a fine line between understanding and surveillance, and I care a great deal about staying on the right side of it.

Scientometrics and the science of science

A long-running thread in my work has been studying science itself. My 2006 Nature correspondence with Andrew Jackson and Benny Lautrup on measures of scientific quality was an early contribution to the bibliometrics debate. The 2018 PNAS paper on the “chaperone effect” (with Vedran Sekara, Pierre Deville, Sebastian Ahnert, Barabási, and Roberta Sinatra) showed that first-time authors publishing in prestigious journals typically do so alongside an experienced co-author who has published there before – a kind of mentorship structure embedded in the publication record itself. It is the kind of result that makes you look at your own coauthor list a little differently.

The Copenhagen Networks Study

From 2012 to 2017 I led the Copenhagen Networks Study, the smartphone-instrumented cohort study that defined the lab’s first decade. CNS sat at the intersection of two grants – the Villum Young Investigator grant where I was sole PI (2012–2017, DKK 6.98M) and the UCPH Programme of Excellence grant Social Fabric where David Dreyer Lassen was PI of a twelve-investigator interdisciplinary consortium (2013–2017, DKK 16M) – and ran the same instrumented population through both. By my own account, the Villum YIP was the grant that made my career. CNS is what it bought.

What was distinctive about that period was less either of those leadership loads on its own and more that they were carried at once. On the Villum side, the responsibility was the conventional one for a young-investigator grant: setting the scientific agenda, recruiting and directing the research team, owning the research output. On the Social Fabric side, where David was PI and I was the only DTU member among the principal co-applicants, I carried something operationally larger: the running of SensibleDTU itself. Three things sat on my desk that the science papers don’t really capture. Engineering: building and operating the multi-channel data pipeline that captured face-to-face proximity, calls, SMS, GPS, and surveys at scale and high resolution. Logistics: procuring, custom-flashing, distributing, supporting, and replacing phones for an entire DTU freshman class. People: hiring, paying, and sustaining a large operational team across years – the kind of staffing structure that conventional research grants don’t really plan for.

Some of it was, in retrospect, genuinely absurd. In 2013 a team of graduate students spent about nine hours rooting and flashing custom firmware onto more than a thousand Nexus 4 phones. At the time I described it as the team “wasting their years of education, doing sweat-shop level work at an undisclosed location in the greater Copenhagen area.” They saved a third of the outer boxes and built a Nexus-box installation in my office while I was away. Flashing the firmware onto a thousand phones in 2013 remains one of my fondest, and most exhausting, memories.

The dataset that came out of those years has been extraordinary. The PLOS One deployment paper (2014, with Stopczynski, Sekara, Sapieżyński, Cuttone, Madsen, and Larsen) is the canonical citation. The Scientific Data release paper (2019, with Sapieżyński, Stopczynski, and Lassen) is the artifact that made the data reusable. In between, the cohort underpinned the gatherings-and-cores paper in PNAS (2016, with Sekara and Stopczynski – cover of PNAS, with a generous Lambiotte commentary called “Rich Gets Simpler”); the WiFi-tracking paper in PLOS One (2015, with Sapieżyński, Stopczynski, Wind, and Leskovec – Google quietly changed its WiFi permission model within days of publication, and I claim no causal arrow but I do note the coincidence); the learning-analytics trilogy on academic performance; sleep and rhythms papers; epidemic-modelling work; and dozens of student projects. Ulf Aslak’s gatherings visualization built on the same data went on to win Science Magazine’s Data Stories competition in 2016.

The most honest description of CNS’s original ambition, which I’ve used in talks, is phishing expedition – if we collect all this data, surely something must be in it. It happens to be a description I can stand behind, because the alternative – to design the perfect dataset before you’ve seen what’s actually there – would have produced a much smaller version of the same insight, on a much longer timeline.

Most of the research above happens inside the Social Complexity Lab at DTU, which I co-lead with Laura Alessandretti, my former postdoc and now an Associate Professor at DTU Compute in her own right. At any given moment the lab is 10-15 people from a mix of countries. It’s also a mix of Ph.D. students, postdocs, and visitors. In June 2023, Det Unge Akademi (the Young Academy under the Royal Danish Academy of Sciences and Letters) named us the Best Danish Research Environment, out of twenty-eight nominated environments across the Danish universities. The jury cited the lab for normalising the discussion of failure, for balancing structured group days with research freedom, and for folding art into scientific communication. I keep the prize on a shelf where I can see it.

Some of the lab’s character is structural: Thursday office hours where anyone can show up with anything; Friday lunches where, by tradition, you describe one paper rejection from the past week. Some of it is just temperamental – I want harmony, and I say this without false modesty, because it’s pure self-interest. A lab that is genuinely kind to itself produces better science with less attrition. A lab that is performatively kind produces neither. The difference matters and I think we have, mostly, managed to be on the right side of it.

Teaching and mentoring

I’ve taught at DTU since 2010, typically reaching 300 to 350 students a year across two MSc-level courses. 02805 Social Graphs and Interactions (about 146 students a year, on average) teaches students to access, collect, and analyse user-generated data using network science and natural language processing; the running case study is a network of country-music artists scraped from Wikipedia, chosen because the data is genuinely messy and cleaning it is part of the lesson. 02806 Social Data Analysis and Visualization (about 187 students a year) is a practical course built around twenty-plus years of San Francisco crime records; students clean the data, merge across schemas, and end the semester with a published data story on GitHub Pages, dressed up in the conventions of long-form data journalism. Both courses are flipped-classroom, project-based, with short pre-recorded lectures and substantial student-led work. I was an early adopter of these methods at DTU. The stated philosophy: working with real data is a pain, and no one will be giving you a nice, cleaned dataset.

My advising portfolio is, by now, considerable: 13 Ph.D. students as primary supervisor, 12 more as co-supervisor, 14 postdocs, 170 MSc thesis projects, 14 BSc thesis projects, and 74 special courses. I’ve been nominated for DTU’s best-teacher prize four times (still haven’t won – but I keep trying) and nominated for best Ph.D. advisor once. Former group members hold positions at Google and Waymo, UNICEF, Northeastern, the IT University of Copenhagen, SODAS, DTU, and universities across Europe; several have started companies (Peergrade most prominent among them); many remain close collaborators and friends. Watching the people I’ve trained become better at this than I was at their age is genuinely one of the best parts of the job.

Science as public conversation

I think scientists have a real responsibility to talk in public about their work, and I have been trying to discharge that responsibility for most of my career. The first wave I lived through was TwitterMood in 2010 – a sentiment-analysis paper that ended up on CBS Evening News with Katie Couric, in The New York Times multiple times, on the BBC, in TIME, the Wall Street Journal, Le Monde, Vanity Fair, NPR, and Scientific American. I was a third-year postdoc; it was the first time I understood, viscerally, that a single press cycle could rearrange your week. The blog at sunelehmann.com has been the through-line ever since – a place for longer-form thinking about filter bubbles, big data versus the right data, the ethics of prediction, and whatever else needed more room than a paper would give it.

During the pandemic I served on the Danish government’s COVID-19 modelling task force (2020–2022, Ekspertgruppen for matematisk modellering af covid-19), which produced infection forecasts and reopening-scenario projections through some of the more uncertain months of the pandemic; I also sat on the advisory board for the national Smittestop contact-tracing app, where I pushed hard, with limited tact, for a privacy-preserving decentralized architecture. From 2022 to 2024 I served on the Danish government’s external expert group on tech giants, chaired by Mikkel Flyverbom at CBS – we delivered a main report in 2023 with thirteen concrete recommendations on harmful design patterns, data harvesting, prediction limits, and the protection of children’s data, and a follow-up report in 2024 on tech giants as digital infrastructure.

The piece of public writing that has had the longest tail, though, came in spring 2023: a four-part Politiken kronik series I co-authored with Anders Søgaard, Rebecca Adler-Nissen, Ole Winther, and Michael Bang Petersen on the societal implications of GPT-4-era language models. Our framing – that society was unlikely to react fast enough on its own – was one of the key threads that fed into the formation of CAISA, the Danish national centre for AI in Society. I came in as an early addition to CAISA, and am now one of its Chief Scientists. Looking back at the arc, the order goes: a research interest in attention and digital traces → blogging about it → being invited onto a government task force → coauthoring an op-ed series with people who agreed something was missing → an early addition to a new national centre that does something about it. None of those steps were planned in that order. Most of the time, public engagement is what scientists do instead of more research. Sometimes, occasionally, it becomes its own research programme.

A few smaller channels round it out. From 2021 to 2022 I hosted Too Lazy to Read the Paper, a podcast in which the author of a recent paper in network or data science explains it to me, conversationally – twenty episodes across two seasons, with guests including Marta González, Roberta Sinatra, Dirk Brockmann, Renaud Lambiotte, Martin Rosvall, Dashun Wang, Tina Eliassi-Rad, Esteban Moro, Laura Alessandretti, and David Lazer. In 2023, my then-postdoc Jonas L. Juul and I ran Philosophy of the Predicted Human at UCPH, a single-topic invited talk series that brought nine speakers – César Hidalgo, Shannon Vallor, Joshua Blumenstock, Johan Ugander, Dean Eckles, Gillian Hadfield, David Lazer, Jon Kleinberg, Keyon Vafa – to push on what it means to live inside predictive infrastructures. The newsletter at sunelehmann.com, Bursty Transmissions, is the slow channel: longer pieces, on the bursty cadence the title promises.

The talk count has also crept up over the years. I have a folder with roughly 240 Keynote files in it; that’s one talk every two or three weeks across the active years, with peaks in 2019 and 2024. The talks I have enjoyed most have been the ones where the audience was not expecting to be in the room with a physicist – Science and Cocktails at Christiania in 2014, still my favourite talk I’ve ever given because the audience laughed at my jokes; a 2023 Bloom Festival conversation with writer Amalie Smith and stem-cell researcher Martin Røssel Larsen on what it means to be alive in an era of digital and medical breakthroughs; a 2025 evening with cartoonist Line Jensen at Tor Nørretranders’s Sommersal on Samsø, on friendship, drawing versus measuring.

Service, recognition, funding

The recognitions I’m most quietly proud of are the ones the lab earned together – most especially the Young Academy’s prize for Best Danish Research Environment in 2023, awarded to the Social Complexity Lab as a whole. Among the individual prizes, the EliteForsk Senior Prize in 2022 is the flagship one in Denmark – five senior researchers per year, total DKK 1.2M (DKK 200,000 honorarium plus DKK 1M for research activities), presented by Crown Princess Mary at Moltke’s Palace in a ceremony that briefly went off-script and required, charmingly, the Crown Princess herself to sort out the on-stage logistics. Earlier prizes include the Columbus Prize (2021, co-recipient, for the HOPE project on Denmark’s COVID-19 response, with Michael Bang Petersen leading from Aarhus and Rebecca Adler-Nissen and me as Copenhagen co-leads), the IC2S2 Best Paper Award (2018, for the complex-contagion work), the Reinholdt W. Jorck Prize (2011, presented at Christiansborg Palace in the Supreme Court chamber), and the Copenhagen Kongres & Event Award (2013, for chairing NetSci in Copenhagen with 400-plus participants).

The institutional memberships have also arrived over time: elected to the Royal Danish Academy of Sciences and Letters in 2024 (Natural Sciences Class); member of the Danish Academy of Technical Sciences since 2021; alumnus of the Young Academy after a standard five-year tenure (2012–2017). I’m a guest editor for PNAS (invited each time by a member of the National Academy) and an Academic Editor for PLOS One, Applied Network Sciences, and Frontiers of Physics. I sit on the executive board of the Network Science Society and review grants for the NSF, the European Commission, ANR, and the Volkswagen Foundation.

Across my career I’ve secured or co-secured something like DKK 103 million in research funding as PI or co-PI. The two grants that mattered most were the Villum Young Investigator Programme in 2012 – the one that made SensibleDTU and the body of work on smartphone-instrumented human dynamics that defined the next decade possible – and the Villum Foundation’s Synergy Grant in 2020 (DKK 19.7M, PI), which is building national-scale social-network infrastructure for Danish research. Add in the Sapere Aude Research Leader grant in 2015 (DKK 6M, sole PI), the UCPH Programme of Excellence grant Social Fabric (2013–2017, DKK 16M, co-PI with David Dreyer Lassen as PI), the Carlsberg Semper Ardens grant for HOPE (2020–2023, DKK 30.3M, co-PI with Bang Petersen as PI), the ERC Advanced grant DISTRACT (2020–2024, ~DKK 18M, co-PI with Morten Axel Pedersen as PI), and a DFF-Project 2 with Morten Mørup at DTU (2020–2024, ~DKK 6M, co-PI), and you’ve got the spine of how the work has been paid for.

The bigger picture

What drives me, when I can articulate it, is the conviction that computational approaches can help us understand social systems in genuinely new ways – but only if we do it with care. The data we work with represents real people living real lives. The patterns we find can inform policy, improve public health, and deepen our understanding of ourselves; they can also be used for surveillance, manipulation, and control. The difference is in the intention, the methods, and the institutional safeguards we build around the work. There is a fine line between understanding and surveillance, and I care a great deal about staying on the right side of it.

Richard Feynman once responded to Whitman’s poem about the learned astronomer by arguing that understanding nature scientifically doesn’t diminish its beauty – it adds new layers of wonder. I feel the same way about human behavior. The more I learn about the hidden patterns in how we move, connect, and influence each other, the more remarkable the whole thing seems.

Notes

A small handful of external references for the threads above:

Papers cited throughout are listed at /featured-publications and on the longer publication list linked from /cv.
life2vec clarification page (the team’s response to the “death predictor” headlines): life2vec.dk.
Too Lazy to Read the Paper podcast: toolazy.buzzsprout.com.
Bursty Transmissions newsletter (the slow channel): /subscribe.
The 2023 Politiken kronik series with Anders Søgaard, Rebecca Adler-Nissen, Ole Winther, and Michael Bang Petersen is paywalled at Politiken; the University of Copenhagen’s SAMF mirrors are on samf.ku.dk.
The Danish government’s Expert Group on Tech Giants delivered its main report in 2023 and a follow-up on tech giants as digital infrastructure in December 2024, both on em.dk.
Johann Hari’s Stolen Focus (Bloomsbury, 2022) takes up the collective-attention work in chapter 3.