← Back to site PT
Behavioral Intelligence for Decisions
Key research on the topic
The spine, substrate and reservoir of research behind grano.Protocol, Granovetter’s method, and its three spheres: grano.Tropic (organizations), grano.Opsis (brands and markets) and grano.Polis (population and public sphere). Direct links where an open PDF exists; search links where access requires a subscription.
Reduce uncertainty about the future, not predict it.
00Document
Theoretical Framework · Granovetter (in Portuguese)
Foundations of grano.Protocol · version 2 · 2026-09-30
★Origin of the name
Mark Granovetter
The two papers by the Stanford sociologist that give the lab its name: individual thresholds decide whether a group stays still or cascades, and weak ties spread innovations better than strong ones.
The Strength of Weak Ties
Granovetter, M. · American Journal of Sociology 78(6) · 1973 · pp. 1360–1380
Threshold Models of Collective Behavior
Granovetter, M. · American Journal of Sociology 83(6) · 1978 · pp. 1420–1443
01Layer 1
Individual fidelity
A language model reproduces a specific person’s behavior with usable fidelity, as long as it is grounded in that person’s real data.
LLM Agents Grounded in Self-Reports Enable General-Purpose Simulation of Individuals
Park, J. S. et al. · Stanford · 2024 · arXiv:2411.10109
Generative Agents: Interactive Simulacra of Human Behavior
Park, J. S. et al. · 2023 · arXiv:2304.03442
Out of One, Many: Using Language Models to Simulate Human Samples
Argyle, L. P. et al. · Political Analysis 31(3) · 2023 · origin of the term silicon sampling
02Layer 2
Collective emergence
Populations of agents reproduce group dynamics (homophily, conformity, community formation) without being instructed to.
Artificial Intelligence Chatbots Mimic Human Collective Behaviour
He, J. K., Wallis, F. P. S., Gvirtz, A., Rathje, S. · British Journal of Psychology · 2026
Emergent Social Conventions and Collective Bias in LLM Populations
Ashery, A. et al. · 2025 · final reference to be confirmed
03Layer 3
Inside the organization
Simulation extends into a specific organization, not just generic external audiences. This is the grano.Tropic sphere, the method’s frontier.
Generative Organizational Behavior Simulation using LLM-based Autonomous Agents: A Holacracy Perspective
Zhu, C. et al. · 2024 · arXiv:2408.11826
04Limits and validation
What is not yet proven
It cuts across every layer. It is what gives the lab credibility before a scientific board.
LLM Social Simulations Are a Promising Research Method
Anthis, J. R. et al. · ICML 2025 · arXiv:2504.02234 · the five challenges; recommends fine-tuning on social science data
The PIMMUR Principles: Ensuring Validity in Collective Behavior of LLM Societies
2025 · arXiv:2509.18052 · audit of 39 studies; a crisis of methodological rigor
LLM-Based Social Simulations Require a Boundary
2025 · arXiv:2506.19806 · a fidelity-boundary approach
Synthetic Replacements for Human Survey Data? The Perils of Large Language Models
Bisbee, J. et al. · Political Analysis 32(4) · 2024 · narrow opinion distributions
Limited Ability of LLMs to Simulate Human Psychological Behaviours: A Psychometric Analysis
Petrov, N. B., Serapio-García, G., Rentfrow, J. · 2024
LLMs fail to reproduce human distributions in economic games
Gao et al. · 2024 / 2025 · the 11-20 game case
Temporal instability and the limits of social simulation with LLMs
Wu et al. · 2025 · simulations are more defensible at the collective than the individual level
Generalization limits of AI surrogates for social science
Crockett, M. and Messeri, L. · 2025 · non-representative training data
05Layer 4 · Umbrella
AI Behavioral Science
The lab’s identity: treating agents as measurement instruments and the model itself as an object whose behavior must be characterized.
AI Behavioral Science
CASBS · August 2025 · arXiv:2509.13323
AI Agent Behavioral Science
2025 · arXiv:2506.06366 · the Machine Behaviour lineage applied to agents
Machine Behaviour
Rahwan, I. et al. · Nature · 2019 · treats AI systems as empirical subjects of study
06Substrate
The books, the conceptual basement
Not appeals to authority. They are the floor of Layer 0: behavior is patterned and can be modeled as a distribution.
Thinking, Fast and Slow
Daniel Kahneman · dual systems, heuristics and biases
Misbehaving / Nudge
Richard Thaler · bounded rationality, choice architecture
Determined / Behave
Robert Sapolsky · behavioral determinism
The Culture Code
Daniel Coyle · culture as an observable behavioral pattern · bridge to Layer 3
Thinking in Systems
Donella Meadows · feedback loops, emergence, leverage points
Clear Thinking
Shane Parrish · deciding under defaults · the product side
07Reservoir
The two lists, the deep well
Not part of the spine. Mined on demand: Social Science feeds Layer 2, Psychometrics feeds Layer 1.
Awesome-LLM-in-Social-Science
ValueByte-AI · Layer 2 reservoir
Awesome-LLM-Psychometrics
ValueByte-AI · Layer 1 reservoir