• A
  • A
  • A
  • ABC
  • ABC
  • ABC
  • А
  • А
  • А
  • А
  • А
Regular version of the site

Petr Parshakov Presented a Paper on LLM Behavior at a Bank of Russia Seminar

On June 30, 2026, Petr Parshakov, Head of IDLab, presented at the 18th Economic Research Seminar "Artificial Intelligence and Economics," held at the North-Western Main Branch of the Bank of Russia in Saint Petersburg. The event was co-organized by the Bank of Russia, NES, and HSE University, bringing together experts from leading research centers and the banking sector to discuss the impact of AI on economic growth, the labor market, inflation, and financial stability.

Petr Parshakov Presented a Paper on LLM Behavior at a Bank of Russia Seminar

As part of the session dedicated to modeling human behavior using large language models, Petr Parshakov presented the paper "Replicating Experimental Behavior in the Trust Game Using LLMs." The study was co-authored with Iuliia Naidenova, Senior Research Fellow at IDLab, and Nikita Matkin, Research Assistant at IDLab. Anton Suvorov, Rector of NES, served as the discussant for the paper.

In their work, the authors examined whether modern language models can replicate the decisions of real participants in economic experiments, including the amounts transferred to a partner in a trust game, expectations regarding returns, and reciprocal responses. The responses of eight modern language models were compared with experimental data previously collected from participants in Moscow, Tatarstan, and the Republic of Sakha (Yakutia).

The results showed that in certain tasks, language models are capable of approximating average human responses. However, this finding is not robust across different models, roles assigned to synthetic respondents, or regional contexts. The main takeaway of the study is that language models cannot yet be considered a direct substitute for human subjects in economic experiments without additional validation; nevertheless, they can be useful for piloting experimental designs and formulating hypotheses.
The study was conducted within the framework of RSF project No. 25-18-00539 "Comparative Analysis of AI-Based Agents and Real Individuals in Economic Decision-Making".