Exploiting Objects as Artifacts: How Social Learning Saved the Pueblo Indians

TL;DR

Researchers from the University of Windsor have developed a new model that treats the environment itself as an "artifact" that agents must learn to use. By simulating the survival of Pueblo Indian settlers (A.D. 600–1300), the study proves that evolving social networks—specifically learning who to listen to within a flexible radius—is the key to surviving extreme environmental shifts.

Background: Beyond Simple Agents

In the world of Multi-Agent Based Social Simulations (MABS), agents often interact with their environment in hard-coded ways. However, human history is defined by our ability to treat objects and landscapes as artifacts—functional tools that require specific knowledge to exploit. This paper positions itself at the intersection of cognitive science and archaeology, using the "Village Eco-Dynamics Project" as a high-stakes testing ground.

The Problem: Survival in a Shifting World

Prior models of artifact exploitation were often too rigid. They assumed agents either learned alone (Individual Learning) or followed a fixed set of tribal rules (Cultural Beliefs). In reality, the landscape of the American Southwest from A.D. 600 to 1300 was a "heterogeneous" mess of varying soil productivity and climate shifts. Fixed strategies simply couldn't keep up with the variability of the land.

Methodology: The Landscape as a Tool

The authors treated the landscape as a complex artifact defined by five functional attributes:

• Elevation and Slope
• Slope Direction
• Depth to Bedrock
• Biomass (Big Sagebrush density)

Learning Strategies

The study compared several "evolutionary" approaches:

1. Individual Strategy: Using a Genetic Algorithm (GA) to find the best farming cells.
2. Social Strategy: Agents share knowledge. Critically, agents could evolve their network radius ($r$), deciding how far afield they should look for "successful" mentors.
3. Cultural Strategy: A Cultural Algorithm (CA) where a central "Belief Space" stores the best examples of survival for everyone to follow.

The simulation maps the evolutionary strategies onto the historical Village Ecodynamics landscape.

Experiments and Results: The Power of Social Networks

The simulation tracked agent survival across four archaeological periods (Basketmaker III to Pueblo III).

Key Findings:

• Social Victory: The "Social Strategy with Learned Radius" was the undisputed champion. It wasn't just about learning from others; it was about adapting the size of the social circle based on environmental pressure.
• Culture vs. Network: Surprisingly, learning from social networks outperformed learning from cultural belief spaces. The authors suggest this is because the landscape was too diverse for a "one-size-fits-all" cultural rulebook to work.
• Archaeological Alignment: The model’s results peaked in accuracy during the Pueblo II and III phases (A.D. 900–1280), exactly when archaeological records show the greatest environmental variability and social complexity.

Evaluation of survival rates across different learning paradigms.

Critical Insight: Why Social Learning Wins

The success of the evolving social radius provides a profound insight into Inductive Bias. In a stable world, individual learning or rigid culture might suffice. But in a "dynamic and heterogeneous" environment, the most valuable trait is the ability to dynamically scale your information-gathering network. The social network acts as a distributed sensor array, allowing agents to "outsource" the risk of experimentation to their neighbors.

Conclusion & Future Work

This research demonstrates that complex social behaviors, such as the expansion of trade and communication networks observed in the Pueblo record, can be modeled as an evolutionary adaptation to artifact exploitation.

Future Outlook: The next step for this framework is moving beyond farming. Could this model explain the evolution of complex tool-making or the transition to urban living? By treating the "technosphere" as a learnable artifact, we can gain deeper insights into both our past and the future of artificial intelligence.