The paper challenges the dominant cognitive‑science assumption that the individual alone is the proper unit of analysis for human intelligence. Using a custom “socioscope” – a suite of smartphones, electronic badges, and body‑worn audio/video sensors – the author and his team collected millions of hours of high‑resolution behavioral data from MIT students, business‑school participants, and other groups. Four hypotheses are advanced: (1) most human behavior is reactive rather than deliberative; (2) a person’s behavior is strongly driven by the behavior of others in their social network; (3) goals, plans, attitudes and intentions can be inferred reliably from brief non‑linguistic social displays; and (4) the emergent network of unconscious social interactions contributes to group fitness. Empirical analyses show that (a) daily activity patterns can be captured by a small set of “situated behaviors” and predicted with >50 % accuracy; (b) conditional‑probability “influence” relationships between individuals explain 95 % of variance in proximity data and accurately recover affiliation and friendship; (c) short‑duration vocal‑body “social displays” (excitement, freeze, determination, accommodation) predict outcomes in business‑card trading, speed‑dating, salary negotiations, and other tasks, accounting for ~40 % of outcome variance; and (d) aggregating information across a small group while weighting objective network structure yields collective estimates that are twice as accurate as any individual’s and four times the average individual’s. The author argues that these findings demonstrate a substantial “network intelligence” component – often 40 %–80 % of behavioral variance – that operates via unconscious, non‑linguistic signals. The paper concludes that human cognition must be understood as a hybrid of individual and social‑network processes, with implications for cognitive science, organizational design, education, and the engineering of collaborative systems.
Topics
Four Foundational Hypotheses on Reactive, Network‑Mediated Human Behavior
Articulation of the four core propositions: (1) predominance of reactive behavior, (2) dominance of social cues, (3) inferability of mental states from non‑linguistic displays, (4) fitness benefits of network dynamics.
The Socioscope: Multi‑Modal Sensor Platform for Continuous Behavioral Capture
Description of the three sensor subsystems (smartphone location/Bluetooth, electronic badge with audio/accelerometer, body‑worn microphone/camera) and the scale of data collected (≈330,000 h, 94 subjects, etc.).
Variability and Predictability of Daily “Situated Behaviors”
Analysis showing that three principal components explain ~80 % of daily variance; clustering into 30 (or 10) situated behaviors yields >50 % to 75 % prediction accuracy, supporting hypothesis 1.
Network Influence: Conditional‑Probability Relations and Proximity Prediction
Quantitative results that influence models explain 95 % of proximity variance, recover work‑group and friendship affiliations with >90 % accuracy, and link vocal turn‑taking influence to betweenness centrality (r = 0.90).
Social Displays: Short‑Term Non‑Linguistic Signals Predicting Outcomes
Definition of four display clusters (excitement, freeze, determination, accommodation) and empirical findings that each predicts specific task outcomes (business‑card exchange, speed‑dating, salary negotiation) with ~40 % explained variance.
Wisdom of the Network: Information Aggregation via Objective Social Structure
Treasure‑hunt experiment showing that weighting objective network metrics and risk profiles yields collective estimates twice as accurate as any individual and four times the average, illustrating hypothesis 4.
Implications for Cognitive Science and Organizational Design
Discussion of how the findings challenge the individual‑centric view of cognition, suggest redesign of companies, education, and collaborative tools to leverage network intelligence, and note limits (Dunbar’s 150‑person cap).