–

Title: From Online Traces to Material Imprints: Articulating a Vision for Planetary Social Computing

Setting: Adel Daoud · ICSC 2026 · 3 September 2026 Chair: Charles Rahal · Nuffield College, University of Oxford

[00:00] The question: planetary social computing

When I was here, I believe 10 months ago, in this exact room, I was presenting our research on planetary causal inference, an agenda that we’re leading in terms of methods development for social scientists. And I then met with Charlie and his team, and they kindly invited me to this conference and said, “What about you come and talk about social computing?” And at that time, I was thinking, “What is social computing?” As a computational social scientist, I’ve been focusing on social data science, broadly speaking.

And so I started to read and think about ways of connecting this field that has been doing amazingly well in terms of the intellectual depth that we require to bring data science, social data science, computational social science onto a different theoretical footing. And so in the past couple of weeks, very much the past couple of days, I’ve been reading deeply and thinking about how to connect the work that we’ve been doing in my lab to this field. And so, hence the idea of planetary social computing.

[01:05] Collective intelligence and material imprints

So the argument in brief is essentially the following. The idea that I’m going to propose to you is: in order to better understand, explain and predict collective intelligence at a broader scale, at the scale of humanity, we really need to think about this phenomenon of planetary social computing. I’m going to start off with an introductory definition and then walk you through what I mean by this. I hope to co-develop this idea further as a work in progress. So planetary social computing—you could think about it as that human process of transforming the Earth to encode past progress.

I’m going to show you exactly what I mean by that. And that past progress is a way of also helping to ensure that the future goes in accordance with what these humans are thinking, predicting, wanting and organizing. By focusing on these material imprints that occur on the face of the Earth, we can then make better predictions about the future. That’s the basic proposition. Starting off with the definition of social computing, I know that this is an active work in progress. What is social computing?

One definition that I’ve resonated with throughout these readings is what James Evans presents in his paper, “Social Computing Unhinged,” in the Journal of Social Computing, where he thinks about this co-development of the social and the computing as one entity that enables collective intelligence. He thinks about it as a collective computer, in a way. And the new objective of the social sciences is to understand these social systems and how they compute to solve their problems. That’s the overarching definition that he proposes there.

[02:51] From bits to atoms

Now, this definition invites us to think about the online world. Most of the examples that James talks about are internet collective behavior in the digital world, and that’s natural because that’s also where most of the data is. What I’m going to ask is to start thinking about how that online world is even possible. How are these computers even possible? How do we think about the material foundation of those social computers in terms of the offline world? And that offline world really brings us to move from bits to atoms.

How do we think about computing in the world of atoms? That’s basically the idea. And so if we think about the concept of a computer, even that— if you think about how that entity is possible, this computer that I have in front of me, that you have in front of you, has a material foundation.

[03:46] The global networks behind a computer

Think about this stylized network, where the raw materials are excavated in South America, in Central Africa, brought to Asia by sub-subcontractors, put together— this web of intertwined collective intelligence that relies on the material world to make computers even possible. It’s obvious in a way. But it requires us to start thinking about this computing happening not only in the online world, but really on the face of the planet. But even that, if you push a bit further, where do these raw materials come from?

How is it even possible to build these machines that go into the mines, that collect these minerals, that make that raw material possible in the first place? You keep on pushing and pushing, and you’ll see that we’re reaching this idea of economic complexity that César Hidalgo and others have worked on: the computer doesn’t exist in a vacuum. If you would bring a computer back to the Stone Age, it would be a computer, but it would be totally worthless because you can’t do anything with it.

It’s functional because it exists in that web of goods and services. Pushing and pushing and pushing, you’ll see that everything continuously happens on the face of the Earth. To make these materials possible, to make this collective intelligence possible, we need to see how humans are transforming the planet, to think about that collective intelligence that we’re seeing in the online world. This web of building.

[05:22] Natural and social planetary systems

To contrast this a bit more, to work with me in terms of the definition of what planetary social computing could be, we can think about planetary natural computing. What would that be as a contrast? If we look at the eons of planetary computing in the past, these are figures borrowed from Frank et al., from the International Journal of Astrobiology. That was fun, to go and read some astrobiology as a sociologist. But what they talk about here and show us is billions of years of evolution in terms of how the planet itself has formed.

And then in the Archean eon of Earth history, basically four billion years ago and onwards, there were no humans, obviously. Everything was naturally adapting in terms of the chemicals that were in the air. It wasn’t even possible for living beings as we know them today to exist. And then naturally that transition happened where oxygen came about, the carbon dioxide levels went down and methane stabilized. We saw the natural conditions for life to emerge. In the era or epoch when humans saw the face of the Earth, we started to transform the planet.

Transformed it in a way that we have created these artificial chemicals as well in the air. But this was, if you like, necessary in order to create computers in the first place. And so you can think about this transition into the future: what would the Earth have to be looking like in terms of the future collective computers that we will have to become? It will have to be a transformation of the planet, in coexistence, if you like, in co-synergy with what is happening at the systems level.

And so if you think about the components of what planetary social computing would require us to think about, there are components of the geosphere, the various types of emergence that happen at a network level, the semantic and the complex adaptive system. What you want is a system that also is self-adaptive to the environment itself in terms of human-environmental coexistence. And you can think about this as a multilayered planetary and dynamic network where you connect the environmental level to the technosphere. And this idea is that what we are today is just a point in where we could be in the future, but that could have to be transformative.

[07:49] A thought experiment: computing on Mars

And to hammer in this point even further, think about what would happen if we start again as humans: we colonize, terraform Mars. How would that look in order to make those computers possible, that AI possible? Well, what level of rapid technology development would the Martians have to develop in order to compete with Earth’s technology? And you could push this even further and think about interplanetary social computing in terms of the trade and the intertwinedness that two planets would have to have in order to make collective intelligence possible and fueled to a level beyond what we can see today.

And obviously keep on pushing. Why stay there? You can think about the interstellar or the intergalactic relationships, networks that you have to build in order to make this new era possible.

[08:40] Measuring Earth with satellite observations

Now, I’m going to bring you back to Earth. How do we do this? How do we go from the philosophical to the actual measurement of the empirical world? And what we’ve been doing in much of my lab is essentially focusing on Earth observations, computer vision as a source to study planetary social computing. And so I’m going to show you some material here. It doesn’t exclude any data sources. Other modalities obviously are very important. But as you’ll see, and what this table shows you, is that with Earth observations, you’ll be able to capture with a high level of fidelity over time and space what is happening on the face of the Earth.

So tabular, text data, network data—all fantastic data sources, but they are hard to acquire oftentimes. And that’s also why we have been focusing on the digital so much, because they are omnipresent and it’s easy to do. But Earth observations give you a gaze at the material world, how humans are transforming from bits to atoms. And that’s the running thread of this talk.

[09:37] Forests, cities, night lights and infrastructure

So what can we do with these material imprints, these changes? What can an image even provide us with? So look at this deforestation process happening in Brazil, for example. I’m not sure how it’s looking at the back of the lecture hall, but you see the deforestation happening over time here: 20, 30 years of data where these material imprints are visibly clear. Same coastal expansion in Dubai. If you look at the 1970s, nothing was existing there in the early days, but they pop up. They’re clearly visible from the sky, that humans are changing the face of the Earth.

Same urban development: Las Vegas, how it looked 20, 30 years ago, and what it is looking like right now. There’s a lot of information happening in these images. But also, if you look at what happens at night, I’m going to stop for a moment and ask: what city is this, and what is going on here? Anyone who would like to take a stab at this? And no Germans can answer. This is a hint. Any Germans who would like to? Yeah, please.

Berlin, exactly. What is going on here? East–West Berlin, exactly. That’s right, you’re hitting it. So what is happening here is essentially— that might be hard to see again, just because of the visual transformation from computers to your eyes— but you see that on the East Berlin side, light is a bit more orange compared to the West. And this is because of the institutional setup of the light bulbs that they used in East Berlin that were markedly different in terms of production.

This is still here 20, 30 years past when the Berlin Wall fell. Amazing how you can see the institutional traces imprinting on Earth. Similarly, if you think about war and destruction: Syria, you see the immense destruction that happens during daylight, but you can see visibly what happens when complete cities go black. And this is a way of thinking about war being a destructive planetary social computing process. You destroy the intelligent possibilities of what a society can do. This is immense destruction of that, but also the way that networks happen over time— this is South Africa during apartheid, how the white communities, when they build roads, in terms of the road networks that they create, very much done in a way such that they’re fueling the exchange of goods and services and communication of the whites, but excluding the black homelands, as shown here in the colored area.

And similarly, if you think about Africa as a continent, much of the work that we’ve been focusing on is Africa. We’re going planetary. It’s the way that these network connections are happening over time, an indicator of trade and other community building over time and space.

[12:23] Measuring health and living conditions

So as I was saying earlier, much of the work that we have been doing is focusing on measuring health and living conditions. And a way to think about this from a causal perspective is that you have a process that you’re interested in measuring, which is poverty and inequality or living conditions, where you want to test for different causal questions. And then what you build is this graphical model in the observational world. But what happens oftentimes is that you face this data scarcity.

This data scarcity of not being able— because there aren’t enough surveys. Surveys are very costly to measure things at scale. So if you look at Burundi, one of the examples, we only have two surveys in Burundi: 2010 and 2017. And so with Earth observations, we might ask: can we use computer vision and images to capture what is going on? So if you look at Cape Town, South Africa here, and then you meditate on the growth that happens over time— Yes, thank you.

Please, yes. Now you see that the neighborhood development that has occurred is indicative of some economic growth at the neighborhood level. And this is then the hypothesis: can we set up this computer vision model that enables us to do this at scale, where we can target the population then with that machine? We are not only confined to the sample; we can scale this.

[13:40] From survey training data to downstream analysis

And so how do we do this? This is getting into the nitty-gritty of the concept, but you can think about chunking up this world into two steps, where the first step is the upstream data creation, where our methodologists get to work to capture this using a supervised machine learning approach, where you have some training data here from the DHS that you can geographically resolve. And then the idea is that you’re going to train the machine on these blue points. And then once you have trained the machine, you could then essentially generalize to these other data points that you don’t have surveys for.

But not only do you want to do this spatially, you want to also do this temporally. The whole thing is to see developments. After training on these blue points, you can now go to new places, both temporally and geographically, and you can infer what is happening there at that location that you didn’t have a survey for, and you can see past trends over time. And then the idea is that you build up a dataset that now, compared with what would happen without our project, without our algorithms, you’ll be able to impute and see developments.

Obviously with some loss in terms of measurement error, but it gives you a way of also thinking about other methodologies to enhance your prediction. And so this is basically then using this data source for downstream analysis. You can think about the way that you can connect cause and effect, and it enables you to think also causally about this process in a planetary social computing setup. And so this is now a dataset that you can use for any type of analysis, not only one, enabling you to go further here in the analysis.

Use this dataset for downstream analysis. That’s the whole point as well, in terms of what we’re doing in enabling us to address the problem of data scarcity: we create these data sources that can be used for that type of analysis. And so you might ask, how well can you do in terms of this approach?

[15:40] How well does satellite-based estimation work?

Empirically, we have a bunch of work. One that I would like to focus on, just a couple of slides, one that we published in IJCAI two and a half years ago, which is exactly showing that you can do this at scale and create this data. We have several new data source products coming out. There’s one on India from the past. We’re working with Satya and others on a planetary scale approach. And so, empirically, as I was saying, what can you do?

How well can you do empirically? So in the interest of time, and this is a whole talk in its own right, is to focus on however you split the data. If you imagine how we have— you have a whole Africa, you have about 70,000 data points to train on, you have the image data as input, you have a survey, a measured wealth index as a target, and however you split the data, splitting by area, by country, by time, you get an average performance of above 0.7 in terms of R-squared.

And that is without doing any annotations on the images. And if you think about this from a social science perspective, that’s an amazing result. We get an R-squared of 0.3, you’re like, “Oh, this looks suspicious!” This is 0.7. We’re doing this without any additional annotation, as I was saying. So this is really just scratching the surface of what we can do in terms of planetary social computing. What would you think about doing with your own work? And one thing that I’m now thinking about is this longer-term measurement of developments and economic growth.

This is from Gapminder, the 1800s. You see the planet’s population growing over time. So the x-axis is economic conditions, and you see the density of that moving over time. And this is an immense development that happened. And the idea that I’ve proposed here is: to what extent, how much of this was made possible by actually just thinking about the way that we transform the surface of our planet? Planetary social computing.

[17:47] Simulating collective intelligence

And so, other ideas that I’m playing around with in our lab are to think about these as big simulators, using image data, using other agent-based models, potentially networks, to think about ways that this is essentially almost like a computer game. I don’t know how many of you played Civilization. I was a big fan when I was younger. Spent a little bit too many hours playing that. But there are these simulators that think about how humanity, as multiple cluster civilizations, is working together, fighting each other, cooperating to build something that we call a grander humanity of civilizations.

So these are tools that we can use to think about predictions into the future. And this is a work in progress. In two days, I’m going to meet with the colleagues at Google DeepMind. We’re going to see how much we can play this out within their own machinery.

[18:38] A vision for planetary prediction

And so to round things off, in terms of a vision for planetary social computing, this is a recurring theme of collective intelligence. If you think about what makes us as humans, as a species, different from other types of animals, much of this boils down to the cumulative cultural evolution of our collective brains. And what I’m saying is that that is absolutely true, but it boils down to thinking about the material foundation of that collective intelligence, and how we change the planet to serve that collective intelligence, and other ways of organizing the planet and societies to make even further collective intelligence possible.

And so if you indulge me, I don’t know how many of you are fans of Foundation, Isaac Asimov’s work, but this is leading us to… Yeah, I think there are some aliens trying to disrupt this knowledge spread, because it might be a total unlock of everything that we might be doing here. The Radiant. The Radiant is this prediction machine that, in this book, they use to see how psychohistory is developing over time. And in our world, the Prime Radiant would be translating this to a big multimodal planetary prediction machine, to think about using images, text, audio, other types of modalities that we can fuse together, to think about ways that we can predict and understand collective intelligence.

Related work

@article{daoud2026chinese,
  title={Chinese vs. World Bank Development Projects: Insights from Earth Observation and Computer Vision on Wealth Gains in Africa, 2002-2013},
  author={Daoud, Adel and Cindy Conlin and Connor T. Jerzak},
  journal={World Development},
  year={2026},
  volume={202},
  pages={107328},
  publisher={}
}

–

@article{fucheng2025optimizing,
  title={Optimizing Multi-Scale Representations to Detect Effect Heterogeneity Using Earth Observation and Computer Vision: Applications to Two Anti-Poverty RCTs},
  author={Fucheng, Warren Zhu and Connor T. Jerzak and Adel Daoud},
  journal={Proceedings of the Fourth Conference on Causal Learning and Reasoning (CLeaR), Proceedings of Machine Learning Research (PMLR)},
  year={2025},
  volume={},
  pages={},
  publisher={}
}

–

@article{pettersson2026debiasing,
  title={Debiasing Machine Learning Predictions for Causal Inference Without Additional Ground Truth Data: 'One Map, Many Trials' in Satellite-Driven Poverty Analysis},
  author={Pettersson, Markus B. and Connor T. Jerzak and Adel Daoud},
  journal={Proceedings of the AAAI Conference on Artificial Intelligence (AAAI-2026)},
  year={2026},
  volume={40},
  pages={39106-39115},
  publisher={}
}

–

@article{jerzak2024effect,
  title={Effect Heterogeneity with Earth Observation in Randomized Controlled Trials: Exploring the Role of Data, Model, and Evaluation Metric Choice},
  author={Jerzak, Connor T. and Ritwik Vashistha and Adel Daoud},
  journal={ArXiv Preprint},
  year={2024},
  volume={},
  pages={},
  publisher={}
}

–

@article{murugaboopathy2026platonic,
  title={Platonic Representations for Poverty Mapping: Unified Vision-Language Codes or Agent-Induced Novelty?},
  author={Murugaboopathy, Satiyabooshan and Connor T. Jerzak and Adel Daoud},
  journal={Proceedings of the 7th International Conference on Social Computing (ICSC 2026), of Communications in Computer and Information Science},
  year={2026},
  volume={},
  pages={},
  publisher={Springer Nature}
}

–

@article{jerzak2023image,
  title={Image-based Treatment Effect Heterogeneity},
  author={Jerzak, Connor T. and Fredrik Johansson and Adel Daoud},
  journal={Proceedings of the Second Conference on Causal Learning and Reasoning (CLeaR), Proceedings of Machine Learning Research (PMLR)},
  year={2023},
  volume={213},
  pages={531-552},
  publisher={}
}

–

@article{sakamoto2026scoping,
  title={A Scoping Review of Earth Observation and Machine Learning for Causal Inference: Implications for the Geography of Poverty},
  author={Sakamoto, Kazuki and Connor T. Jerzak and Adel Daoud},
  journal={Hall, Ola and Ibrahim Wahab (eds.), Geography of Poverty},
  year={2026},
  volume={},
  pages={},
  publisher={Edward Elgar Publishing (Cheltenham, UK)}
}

–

@article{jerzak2023integrating,
  title={Integrating Earth Observation Data into Causal Inference: Challenges and Opportunities},
  author={Jerzak, Connor T. and Fredrik Johansson and Adel Daoud},
  journal={ArXiv Preprint},
  year={2023},
  volume={},
  pages={},
  publisher={}
}