Predicting your social interactions | Sune Lehmann | TEDxAarhus
The speaker presents groundbreaking research showing that observing social networks via detailed, short-duration smartphone data reveals group formation patterns, which the speaker then theorizes correlate with geographical movement patterns, suggesting social life is most predictable when physical movement is unpredictable. This method uses an analogy of an "anthill" to describe the massive data collection effort from the Technical University of Denmark, culminating in a theory that exploration predicts close social bonds. The speaker concludes by demonstrating this theory by asking the audience to identify if they are with someone they know well versus a first date. ## Speakers & Context - Young researcher, previously worked at US universities like Harvard and Northeastern. - Currently presenting findings based on a study of the Technical University of Denmark freshmen. - The project involved the development of a massive observational system to map human social interactions. ## Theses & Positions - Analyzing raw phone call data alone is insufficient for understanding human connection because *"only a fraction of our communication happens over the phone."* - Observing a social system at a very granular level—specifically, in short five-minute snapshots—allows for the discovery of novel patterns invisible in broader data sets. - The structure of social groups can be modeled by analyzing transitions between social contexts, rather than requiring the analysis of the entire network. - There is a deep similarity between the way people move through social space (context to context) and how they move through geographical space (place to place). - A crucial finding is that social life is most predictable precisely when physical movement is unpredictable, suggesting that exploration tends to occur with familiar companions. ## Concepts & Definitions - **Social Network Analysis:** The study of relationships (phone calls, location, communication) between individuals. - **Social Context:** The specific group or activity within which people are interacting (e.g., study group, soccer team). - **Anthill:** Used as a metaphor for the massive, teeming activity of a social system. - **Social Predictability:** The tendency for individuals to interact with the same people in established patterns over time. - **Geographical Predictability:** The tendency for individuals to return to the same physical locations in established patterns. ## Mechanisms & Processes - **Data Collection:** Purchased **1,000 identical top-of-the-line smartphones** and installed **custom software** on each to track phone calls, text messages, Facebook feed activity, and co-location data. - **Group Discovery:** By analyzing a five-minute snapshot of all connected people, the system could isolate and define small, temporary clusters of people hanging out in the real world, solving the "fundamental problem of finding groups in social networks." - **Transition Modeling:** The ability to compare adjacent five-minute windows allowed researchers to study how people transition from one social context to another. - **Prediction Model Transfer:** Because the mathematical equations for moving from context to context were simple, the speaker was able to transfer location prediction methods directly to social data, creating a "machine that could predict people's future encounters." - **Comparative Analysis:** The research compared social predictability against spatial predictability to find commonalities and divergences. ## Timeline & Sequence - Initial work involved analyzing phone call networks from US universities like **Harvard** and **Northeastern**. - The project focused on the **freshmen at the Technical University of Denmark**. - The initial breakthrough occurred when analyzing a five-minute snapshot on an unspecified **Tuesday**. - The study progressed to comparing adjacent time windows (e.g., noon to 5 minutes past noon, and 5 past to 10 past). - The final comparison phase involved comparing derived social predictability against known spatial movement data. ## Named Entities - **Technical University of Denmark** — The institution whose freshmen were the primary data source. - **Harvard** — University where the speaker previously worked with network data. - **Northeastern** — University where the speaker previously worked with network data. ## Numbers & Data - Number of phones purchased: **1,000**. - Scale of initial data: *"billions of phone calls."* - Duration of key observation window: **five-minute snapshot**. - Example comparative time slots: **noon and 5 minutes past noon**, and **5 past to 10 past**. ## Examples & Cases - **The initial limitation:** Phone calls only capture a fraction of human connection; the data was insufficient. - **The five-minute snapshot:** A "humble picture" of everyone together between noon and 5 minutes past noon on a specific Tuesday, which revealed patterns invisible in the broader data. - **The group formation:** The network data *fell into bits and pieces*, each piece representing a small, temporary group of people. - **The exploration paradox:** When movements are unpredictable (e.g., exploring dangerous mountainous terrain), the social life tends to be *most predictable* (i.e., accompanied by familiar friends). ## Tools, Tech & Products - **Custom software:** Installed on the smartphones to measure network activity. - **Top-of-the-line smartphones:** Used in quantity (**1,000**) for data collection. - **Whiteboard:** Used by the research team for visualizing and "drawing on" the findings. - **Location Prediction Methods:** The transferrable mathematical framework derived from geographical tracking. ## References Cited - None explicitly cited; the work is presented as a novel internal discovery. ## Counterarguments & Caveats - The original data set (phone calls) was insufficient because *"only a fraction of our communication happens over the phone."* - The speaker had no prior methodology to scale this level of observation; *"no one had ever done anything like this on a scale like this before."* ## Conclusions & Recommendations - The primary recommendation is to adopt a framework that models transitions between social contexts, rather than attempting to map the entire network at once. - The theory suggests that documenting both location and social data can reveal that periods of physical novelty predict stable social companionship. ## Implications & Consequences - The research methodology provides a "new mental toolbox" for other researchers in various sciences. - The study implies that understanding human connection requires analyzing transient, short-term, highly detailed interactions rather than broad longitudinal views. - The theory suggests that people undertaking novel or unpredictable physical activities (exploration) are more likely to do so with people they are already deeply connected to. ## Verbatim Moments - *"imagine seeing a university campus from a helicopter but with detailed data about every single individual so how they move around who they're hanging out with and also all the online communication mapped out that's what you need to really understand human social networks"* - *"I had this idea that we could take our understanding of social networks to the next level if we could just zoom out and collect data about an entire social system"* - *"I purchased 1,000 identical top-of-the-line smartphones"* - *"the green network that's real data and it's for the same people and it's what researchers had looked up up until then and that's why the patterns that my team and I had found were invisible to them"* - *"this gives us a whole new way of thinking about how people move through the social space"* - *"if people are truly friends they don't just meet once they meet again and again over days and weeks and months"* - *"if I go hiking in some dangerous mountainous terrain I don't bring some random stranger with me I go with my friends"* - *"I think almost all of you in the audience are out exploring today by being here you're doing something unusual"* - *"We should take these methods of location prediction and transfer them directly to our social data"*