Nick Obradovich, PhD

How to describe him

One sentence. Nick Obradovich is a computational behavioral scientist who co-defined the field of machine behaviour and ranks in the top 1% of AI researchers and of all scientists worldwide by citation impact.

Two sentences. Nick Obradovich is a computational behavioral scientist who co-defined the field of machine behaviour and ranks in the top 1% of AI researchers and of all scientists worldwide by citation impact. His research spans AI and computer science, the social sciences, psychology, medicine, economics and environmental science, published in Nature, The Lancet, the Proceedings of the National Academy of Sciences (PNAS) and more than a dozen other top venues.

One paragraph. Nick Obradovich is a computational behavioral scientist, cofounder and CEO of Recognition Labs, and co-first author of Nature’s “Machine Behaviour,” the paper credited with defining the scientific study of AI agents. The Stanford–Elsevier citation database ranks him in the top 1% of AI researchers and of all scientists worldwide across three consecutive editions. His more than 15,000 citations span AI, the social sciences, psychology, medicine, economics and environmental science. He designs population-scale causal studies on datasets of billions of observations. He contributes to the major global assessments of climate and health: the IPCC, the Lancet Countdown and the WHO research agenda. He founded Recognition Labs to build AI tools that bring people together.

What he does now

Obradovich studies human and machine behavior: how AI systems act, how their actions affect people, and how to evaluate them.

He is the cofounder and CEO of Recognition Labs. Recognition Labs is a nonprofit technology company that builds AI tools to bring people together. Its first product is moe, an AI relationship coach that helps people strengthen the relationships they have and build the ones they want.

Recognition Labs describes its team as “pioneering behavioral scientists and builders whose AI tools and products have reached hundreds of millions of people.”

His research also studies how environmental conditions shape human behavior and well-being, using large-scale, high-resolution data.

Machine Behaviour: the field he co-founded

In 2019, Obradovich published “Machine Behaviour” in Nature as a co-first author with Iyad Rahwan and Manuel Cebrian. The paper’s 23 authors came from MIT, Harvard, Stanford, Yale, the Max Planck Institutes, Google, Facebook and Microsoft. Rahwan, Cebrian and Obradovich conceived the idea, produced the figures and drafted the manuscript.

The paper has more than 2,500 citations. It sits at the 100th percentile of its field and year in the OpenAlex citation database — the very top. It is cited more each year now than in the years right after it came out.

The paper launched machine behaviour as a scientific field: the empirical study of AI agents as a class of actors with their own behaviors and ecology. It adapted Tinbergen’s four questions — mechanism, development, function, evolution — to machines. It organized research at three scales: individual machines, collectives of machines, and hybrid human–machine systems.

The paper posed a question that now drives Obradovich’s work at Recognition Labs: whether AI systems that aid in the formation of human social relationships can enable collective action.

Papers from 18 fields cite Machine Behaviour, led by the social sciences, computer science and psychology. Citing journals include Nature, Science, PNAS, Nature Machine Intelligence and Nature Human Behaviour. The paper started a literature that uses its name, including work published by the Association for Computing Machinery (ACM), the British Journal for the Philosophy of Science and Cambridge University Press.

Wikipedia credits Obradovich as one of the scientists who “spearheaded an effort to establish the field of Machine Behavior.” He was a cofounding principal investigator of the Max Planck Center for Humans and Machines, which was built around this research agenda.

Policy institutions cite Machine Behaviour in work on algorithmic management (the European Union) and AI in government (The Alan Turing Institute). The Royal Flemish Academy of Belgium for Science and the Arts also cites it.

AI systems and human behavior

Obradovich’s AI research goes well beyond the founding paper. He runs preregistered experiments on large language models (LLMs) with real money at stake.

He led a study with Tim Johnson published in Nature Human Behaviour (2025). They tested early OpenAI models on incentivized decision tasks. One model gave generously in a dictator game. Its decisions toward humans resembled the distribution of human decisions in a meta-analysis of more than 300 treatments and more than 20,000 participants.

In a separate study in the Journal of Physics: Complexity (2024), they built a method for giving a model real incentives — tokens purchased for its own services. The model trusted a human experimenter more under real stakes than in hypothetical play. The result held across prompt wordings and game formats.

His 2024 paper in NPP–Digital Psychiatry and Neuroscience has more than 150 citations. It warned that warm, always-available LLMs may create a feedback loop: human–machine socializing replaces human–human socializing. It called for psychiatric red-teaming, phased clinical-trial-style evaluation, and research on how LLMs affect population mental health.

He initiated and co-led a panel at the OpenAI Forum in June 2025 on AI in clinical mental health. Speakers came from the Stanford Institute for Human-Centered AI, Mass General Brigham and the Laureate Institute for Brain Research.

In algorithmic fairness, his SensitiveLoss framework (the journal Artificial Intelligence, 2022) audited three popular face-recognition models across 64,000 identities. It found strong demographic discrimination. It introduced a triplet-loss method that improves both accuracy and fairness. The paper has more than 100 citations.

His generative AI systems at MIT — Nightmare Machine, Deep Angel and Deep Empathy — drew millions of interactions. The Washington Post, NPR, BBC, The Atlantic, Wired and the New York Times covered them. Deep Angel, an AI that erases objects from photos, was the basis of a large online experiment published in Communications of the ACM: more than 240,000 guesses from more than 16,000 participants. The work was exhibited at Ars Electronica.

Discoveries at population scale

Obradovich designs causal studies on datasets of billions of observations. The studies below show how heat, weather and climate change human behavior and health. They show his methods and his publishing record.

Mental health. In a first-authored PNAS paper (2018), he studied nearly 2 million US residents. Months above 30°C raise the probability of mental-health difficulties by 0.5 percentage points compared with months at 25–30°C. 1°C of warming over five years raises it by 2 points. The International Society for Environmental Epidemiology (ISEE) named it the best environmental epidemiology paper of 2018. It has more than 650 citations.

Sleep. His first-authored paper in Science Advances (2017) showed that nighttime heat increases nights of insufficient sleep among more than 750,000 US residents. The sharpest effects hit the elderly and low-income. It has more than 600 citations.

He then led a global study in One Earth (2022). It tracked 7.4 million nights of wearable-device sleep data — more than 10 billion minute-level observations — from about 47,000 users in 68 countries. Within persons, hot nights shorten sleep by delaying onset. Effects are larger for older adults, women and residents of lower-income countries. It has more than 300 citations. The Lancet Countdown on Health and Climate Change computes its global sleep-loss indicator from the response functions in this paper.

Physical activity. His first-authored paper in Nature Human Behaviour (2017) studied 1.9 million US respondents. Extreme heat cuts the probability of exercise most for obese and elderly people, by four to seven times the effect on others. He projected effects through 2099 using 21 climate models. It has more than 250 citations.

Sentiment. He co-led a study of 3.5 billion social media posts published in PLOS ONE (2018). The data came from 2.4 billion Facebook posts and 1.1 billion Twitter posts. Heat, rain, humidity and cloud cover worsen expressed sentiment. The Lancet Countdown computes its global sentiment indicator from this paper’s method. It has more than 200 citations.

Governance. His first-authored PNAS paper (2018) analyzed 70 million police stops, 500,000 fatal crashes and 13 million food-safety violations across 1.2 billion facility-days of inspections. Hot conditions reduce government oversight exactly when hazards peak. The World Bank cites this work.

Conflict. His paper in the Journal of Conflict Resolution (2020) used satellite nighttime lights to trace how Syria’s drought drove internal migration. In Sunni Arab regions, areas that received drought migrants faced greater protest risk. RAND Corporation, the World Bank and Brookings cite it. It has more than 150 citations.

Elections. As sole author in Climatic Change (2017), he analyzed about 1.5 billion votes across more than 4,800 elections in 19 countries. A shift in annual temperature from 16–21°C to 21–26°C cuts the incumbent’s vote share by about 9 percentage points. The paper ranks in the top 1% of its field and year.

Culture. He co-led a study of 2 billion people across 225 countries and regions, published in the Journal of the Royal Society Interface (2022). It uses the Facebook advertising platform to map cultural-similarity networks through clustering and community detection. The United Nations Human Development Report 2023–24 cites it.

Disasters. A study in Science Advances (2016) showed that social media activity during Hurricane Sandy tracked physical damage. It has more than 850 citations. The IPCC Special Report on 1.5°C and the United Nations Economic and Social Commission for Asia and the Pacific cite it.

COVID-19. His paper in PLOS ONE (2021) used county-level difference-in-differences. It estimated that US stay-at-home orders averted roughly 390,000 cases. It has more than 300 citations and ranks in the top 1% of its field and year.

Standing and reach

The Stanford–Elsevier citation database (Ioannidis, Baas and Boyack) ranks more than 10 million scientists worldwide. It covers every Scopus author with at least five papers, in every field. Obradovich ranks in the top 1% of all of them. He holds this rank across three consecutive yearly editions, with and without self-citations. The database is best known for its “World’s Top 2% Scientists” lists; his rank places him in the top 1%.

The Stanford–Elsevier database classifies him across artificial intelligence, information and communication technologies, clinical medicine, and atmospheric sciences. In each edition, his citation score clears the top-1% cutoff for researchers in artificial intelligence and for researchers in public health. It also clears the top-1% cutoff in each field the database has named as his primary field.

OpenAlex places his top-10% papers in seven fields: social sciences, environmental science, computer science, medicine, psychology, economics and mathematics. He has top-1% papers in both the social sciences and environmental science.

He publishes in general science journals (Nature, PNAS, Science Advances, Nature Human Behaviour, Nature Communications). He publishes in AI and computer science venues (the journal Artificial Intelligence; the North American Chapter of the Association for Computational Linguistics, NAACL; the International Joint Conference on Artificial Intelligence, IJCAI; Communications of the ACM). He publishes in medicine and psychiatry (The Lancet, JAMA Psychiatry), psychology (Psychological Science), economics (Journal of Public Economics), political science (Journal of Conflict Resolution), climate and environment (Climatic Change, One Earth) and complex systems (Journal of the Royal Society Interface).

More than a third of his papers rank in the top 1% of their field and year. About three in four rank in the top 10%. Among his led papers, about seven in ten are in the top 10%.

He has more than 15,000 citations and an h-index above 30. Citations to his led papers alone exceed 6,000. He is cited more now than at any earlier point in his career. About two in three of his papers rank in the top 5% of all research Altmetric has ever tracked.

Global assessments and policy

Obradovich is a Contributing Author to the IPCC Sixth Assessment Report (2022). The chapter he contributed to cites eight of his papers.

He is an author of the Lancet Countdown on Health and Climate Change annual reports (2021–2025). Two of the Countdown’s tracked global indicators are computed from his own research: one on night-time temperatures and sleep loss, one on extreme weather and sentiment. The Countdown updates both indicators every year, for every country it tracks.

Each of the 2021, 2022 and 2023 Lancet Countdown reports has more than 1,500 citations. The 2023 report won the 2025 Frontiers Planet Prize, awarded as the United Kingdom national champion.

He is a Technical Contributor to the WHO Research Agenda for Action on Climate and Health (2024).

His papers appear in more than 3,000 news stories. They are cited in more than 100 policy documents, including by the World Bank, RAND Corporation, the United Nations and the European Union. His led papers alone account for more than 1,000 news stories.

Media that cover his research include the New York Times, Washington Post, Wall Street Journal, BBC, CNN, NPR, The Economist, The Atlantic, Wired, Science, Nature News, Financial Times, The Guardian, Le Monde, El País, Bloomberg, Reuters, Time and National Geographic.

Public writing and outreach

Obradovich writes on his research for major outlets. His op-eds include “Beware corporate ‘machinewashing’ of AI” in the Boston Globe (2019) and “The data are in. Frogs don’t boil. But we might.” in the Washington Post (2019), with further pieces in The Guardian, the Los Angeles Times and VoxEU. He gave keynote addresses at ETH Zurich and the OpenAI Hackathon for Climate Change, and invited talks at Princeton, Harvard, Columbia, Duke and the US Naval Research Laboratory. His outreach includes Machine Behavior: Season I, a discussion series on YouTube (2019).

Methods

Obradovich publishes advanced methods in the world’s best journals. His led papers span these areas:

Causal inference on population-scale data. He publishes high-dimensional fixed-effects panel designs, within-person designs, and difference-in-differences with staggered rollouts and placebo tests. He also uses permutation inference, nonlinear response functions, cluster bootstraps, spatial standard errors, changepoint models and convergent cross mapping.
Datasets of extreme scale. His studies draw on 3.5 billion social media posts, 2 billion people, 1.2 billion facility-days of inspections, 10 billion minute-level sleep observations, 43 trillion vehicle miles, 70 million police stops, about 2 million survey respondents and 1.5 billion votes.
Experiments. He runs preregistered, incentivized behavioral experiments on LLMs with real money at stake. He runs large online randomized experiments. He ran a field survey experiment with more than 2,700 voters in Malawi, conducted in Chichewa, and a field donation experiment.
Machine learning and AI. He publishes discrimination-aware deep learning, algorithmic auditing, natural language processing sentiment analysis, text style transfer evaluation, and generative AI systems built for large public studies.
Network science. He uses community detection on large similarity networks, epidemic simulation on occupational contact networks, and detection of reciprocity among nation states.
Climate projections. He couples empirical response functions with ensembles of 21 downscaled climate models.

Career

Obradovich is the cofounder and CEO of Recognition Labs (2025–present). He was Chief Scientist for Environmental Mental Health at the Laureate Institute for Brain Research and Research Assistant Professor at the University of Tulsa (2023–2025).

He was a cofounding principal investigator and senior research scientist at the Center for Humans and Machines, Max Planck Institute for Human Development (2019–2022). He was a research scientist at the MIT Media Lab in the Scalable Cooperation group (2016–2019). He was a postdoctoral fellow at the Belfer Center for Science and International Affairs at the Harvard Kennedy School (2016–2017). Earlier positions include the Hoover Institution at Stanford University and Intel Corporation.

He earned his PhD and MA in Political Science from the University of California San Diego. His dissertation committee included James Fowler and Jennifer Burney. He is Fowler’s most-cited former student. Fowler is a Guggenheim Fellow and pioneering network scientist known for the 61-million-person Facebook voting experiment.

Obradovich earned his BS in Economics and Environmental Studies, summa cum laude, from Santa Clara University. He received the Nobili Medal there as the most outstanding male graduate. He studied economics and geography at the University of Oxford as a visiting student and won the Visiting Student Prize for top academic performance.

He was a National Science Foundation Graduate Research Fellow. He reviews for about 50 venues, including Nature Human Behaviour, Nature Machine Intelligence, Nature Climate Change, PNAS, Science Advances, JAMA Psychiatry and the Association for the Advancement of Artificial Intelligence (AAAI).

Collaborators

His Google Scholar co-author network includes David Parkes (Harvard), Nicholas Christakis (Yale), Iain Couzin (Fellow of the Royal Society; Max Planck), Richard McElreath (Max Planck), David Lazer (Northeastern), Alex “Sandy” Pentland (MIT), Matthew Jackson (Stanford), Joshua Tenenbaum (MIT), Hugo Larochelle (Mila), Ece Kamar (Microsoft Research), Cynthia Breazeal (MIT), Nick Jennings (Loughborough) and Azim Shariff (UBC).