A Generative Model of Conspicuous Consumption and Status Signaling
LLM-based agents in a simulated society spontaneously created luxury markets and status symbols.
A team from DeepMind and the University of Toronto has published a groundbreaking paper, 'A Generative Model of Conspicuous Consumption and Status Signaling,' proposing a new computational theory of social status. The research moves beyond classical frameworks like Costly Signaling Theory by using Large Language Model (LLM)-based agents within the Concordia multi-agent simulation framework. The core idea is that status symbols emerge endogenously through a feedback loop: agents observe social patterns and complete them predictively, a process grounded in the 'theory of appropriateness.' This allows the model to explain how the meaning of goods can drift dynamically over time, a phenomenon difficult for static preference models to capture.
In their simulations, the researchers manipulated social visibility within the agents' daily routines. They demonstrated that this social observation transformed basic functional demand into conspicuous consumption and status-seeking behavior. Crucially, the simulations produced real-world economic phenomena like price run-ups and Veblen effects—where demand for a good increases as its price rises—for both real luxury items and procedurally generated synthetic goods. This latter finding helps rule out the idea that the behavior was solely a bias from the LLMs' pretraining on human data.
The study also showed that designated 'influencer' agents could drive the formation of distinct subcultures through targeted sanctioning, and that similar social influence effects generalized to non-monetary signaling behaviors. By bridging micro-level agent cognition with macro-level economic patterns, this work offers a powerful new methodology. It provides a generative, testable platform for forecasting how cultural conventions, market dynamics, and social norms can emerge organically from complex multi-agent interactions.
- Used LLM-based agents in the Concordia framework to simulate how status symbols form endogenously through social observation.
- Demonstrated the emergence of Veblen effects and price run-ups for both real and synthetic goods, ruling out simple pretraining bias.
- Showed 'influencer' agents can create distinct subcultures, offering a new method to model cultural and economic convention formation.
Why It Matters
Provides a testable simulation framework for economists and sociologists to model complex social phenomena like luxury markets and trend formation.