When Social Networks Become Predictive Systems

Series: Seeing the Social Program (4/4)

Your Social World as a Graph

You experience relationships as human and contextual.

Large-scale systems represent them as graphs:

• Nodes (people)
• Edges (connections)
• Weights (interaction strength over time)

This representation enables analysis at scale:

• Which connections are active
• Which individuals are central
• How information moves through a network

On its own, this is descriptive. Combined with behavioral data, it can be used for prediction.

From Description to Prediction

Research across network science has shown that social graphs can be used to:

• Identify influential nodes (e.g., high centrality)
• Detect communities and clusters
• Model likely paths of information flow

Work from institutions including Stanford University and MIT has demonstrated that, given sufficient interaction data, it is possible to estimate how behaviors and information propagate within networks.¹

These models do not guarantee outcomes.

They provide probabilistic forecasts:

• Who is likely to adopt an idea
• Where influence is concentrated
• How quickly signals may spread

When Prediction Meets Targeting

The implications of network-level modeling became widely visible during the Cambridge Analytica scandal.

Public reporting and regulatory findings indicated that data derived from millions of Facebook users was used to:

• Build psychographic profiles
• Segment audiences into groups
• Deliver targeted political messaging

The effectiveness of such campaigns relied not only on individual profiling, but on group segmentation and influence pathways.

Rather than attempting to reach everyone equally, strategies focused on:

• Identifying receptive segments
• Prioritizing key audiences
• Delivering tailored messaging at scale

This reflects a broader principle:

Influence at the network level often operates through selective targeting and amplification, not universal persuasion.

How Groups Are Shaped Indirectly

Modern recommendation systems do not need to explicitly direct group behavior.

They can:

• Surface content differently across clusters
• Prioritize interactions within specific communities
• Adjust visibility of topics over time

These mechanisms affect:

• What conversations are sustained
• Which viewpoints gain prominence
• How group norms evolve

This influence is typically indirect.

It operates through:

• Selection (what is shown)
• Ranking (what is prioritized)
• Repetition (what is reinforced)

The Environmental Effect

At the group level, influence becomes environmental.

Individuals tend to:

• Calibrate beliefs against peers
• Trust familiar sources
• Adopt norms within their network

If the surrounding environment shifts, even subtly, perception can shift with it. This does not require coercion. TikTok trends like unboxing Labubu and claiming a StanleyQuencher mirror the frenzy of earlier products like Furby or Ugg boots.

To catch our attention, and convert purchase intention, it requires:

• Consistent exposure
• Reinforced patterns
• Social network-aligned signals

The Core Tension

This leads to a final tension in this series:

Community vs. steerability

Social groups provide:

• Support
• Identity
• Shared understanding

At the same time, when modeled as networks, they can become:

• Predictable
• Segmentable
• Influenceable through system design

Why This Matters

Most systems no longer focus on a single message or a single user. They operates on patterns across users, relationships between users, and the dynamics within groups.

Being able to anticipate trends means being able to amplify specific narratives in order to adjust iformational environments.

That's not direct control. It's not even coersioon. Yet, structured influence over what surrounds you still feels overly persuasive.

Companion Intelligence Perspective

Companion Intelligence is designed with the belief that your data belongs to you.

• Relationship data remains local
• Social context is not exported for external analysis
• Systems are not optimized to shape group behavior

Instead, we focus on user-directed understanding:

• Identifying personal communication patterns
• Recalling commitments and interactions
• Observing changes over time

This shifts the role of the system from predictive modeling to personal clarity.

Closing

If systems can map your relationships, they can model your network. If they can model your network, they can estimate how it moves. And if they can shape what the network sees, they influence the environment the group depends on.

Citations

1. Easley, D., & Kleinberg, J., “Networks, Crowds, and Markets: Reasoning About a Highly Connected World”, Cambridge University Press (2010); research and coursework associated with Stanford University and MIT on network analysis and information diffusion.
2. UK Information Commissioner’s Office, “Investigation into the use of data analytics in political campaigns” (2018), related to the Cambridge Analytica scandal.
3. U.S. Federal Trade Commission, Facebook, Inc. Privacy Settlement (2019), concerning data handling practices by Facebook.