Social Learning & DeGroot

Social learning explains how your beliefs form by averaging what the people around you think. Think of it like a conversation where you update your opinion based on weighted input from friends, colleagues, and experts you trust. The DeGroot model captures this mathematically: people repeatedly take weighted averages of their neighbors' beliefs until things settle down. Whether the group reaches agreement depends entirely on how everyone is connected. Try this: Map out who influences your thinking and who influences them.

Social learning models how beliefs spread through networks as people repeatedly update their opinions by averaging what those around them think. The key insight is that your influence depends on your position in the network structure—well-connected people act as hubs that shape everyone else's views. Whether a group converges to consensus depends on whether the social network is connected enough for information to flow between all members.

This pattern shows up everywhere. In organizations, corporate culture forms as employees repeatedly socialize and align with influential leaders and peers. In financial markets, investors watch each other's behavior and can drive prices away from fundamentals through cascades—herding into stocks just because others are buying. In online platforms, recommendation algorithms can contribute to filter bubbles by preferentially showing users content that aligns with their existing beliefs, potentially reducing exposure to diverse perspectives.

The framework reveals that influence is unequally distributed: centrally-positioned nodes disproportionately shape final beliefs. This becomes a problem when networks contain "stubborn" agents who never update their beliefs—they can anchor the entire group's consensus at their position, even if most people disagree. The model also exposes how social influence can undermine collective intelligence: when people pay too much attention to each other rather than their private information, groups can converge on wrong answers and lose the wisdom that diverse, independent opinions would provide.

Practical implications: First, diversify your information sources by seeking perspectives from different parts of your network. Second, recognize when influential actors might steer consensus for their own benefit. Third, encourage exposure to dissenting views—assign someone to argue against prevailing opinions, rotate people through teams, or use anonymous feedback.

The DeGroot model, introduced by statistician Morris DeGroot in 1974, provides the canonical framework for understanding belief updating through social learning. In this model, each person holds a belief and updates it by taking a weighted average of their neighbors' current beliefs, with weights representing how much trust they place in each connection. The process repeats until beliefs either converge to consensus or settle into persistent disagreement. Whether convergence occurs depends on network structure: the group reaches consensus if the network is strongly connected and aperiodic (or under certain other technical conditions).

• Golub and Jackson (2010): Demonstrated that social learning can undermine the wisdom of crowds because influence correlates with network centrality rather than information quality—central nodes shape group beliefs regardless of whether they're well-informed, which can lead groups away from truth rather than toward it [1].
• Acemoglu, Como, and Fagnani (2010): Showed that when networks contain 'stubborn' agents who never update their beliefs, flexible agents converge to a weighted average of the stubborn positions—and these stubborn actors can exert disproportionate influence even when they're a minority [2].
• De Dreu and West (2001): Found that minority dissent improves team innovation by triggering more thorough information processing and creative problem-solving—challenging consensus stimulates better group decisions [3].

The DeGroot model assumes people update beliefs mechanically by weighted averaging, ignoring strategic behavior, confirmation bias, motivated reasoning, and other cognitive biases that can distort social learning in real life. It also assumes fixed networks and trust weights, while in reality relationships form endogenously based on beliefs, creating feedback loops that can produce polarization or echo chambers. The model also treats beliefs as one-dimensional scalars, whereas real beliefs involve complex, multidimensional structures that don't easily average. Finally, the framework doesn't specify where initial beliefs come from or how they relate to objective truth, so it can't distinguish between converging on correct versus incorrect answers without external validation.