Google Researchers Can Develop an AI That Mirrors Your Thinking After Just Two Hours of Conversation

In a groundbreaking study, Google researchers, in collaboration with Stanford University, have developed an AI system that can closely simulate human behavior after just a two-hour interview. The research, titled *Generative Agent Simulations of 1,000 People*, involved 1,052 participants who first read a passage from *The Great Gatsby* and then engaged in a detailed interview about their personal lives. Using the data from these interviews, the researchers created AI agents capable of emulating the participants' actions with 85% accuracy.

The primary aim of this project is to provide policymakers and businesses with a more accurate and continuous method for understanding public behavior. These AI agents, which mirror real individuals, could potentially replace traditional techniques like focus groups and polling, offering a more dynamic and efficient way to gauge public opinion.

To develop these agents, participants were interviewed by a 2D sprite AI, and their responses were processed using a large language model (LLM). The researchers then assessed the agents' performance by comparing their reactions to those of the real participants in different scenarios, such as economic decision-making games like the Prisoner's Dilemma and the Dictator's Game.

While the AI agents replicated participants' responses with 85% accuracy in general questions drawn from the General Social Survey, their performance dropped to about 60% when tested in the economic games. This suggests that while the AI is capable of mimicking basic behavioral patterns, it still faces challenges in replicating complex decision-making processes that involve economic reasoning and social dynamics.

The study also raises important ethical concerns about the potential misuse of such technology. If AI can simulate human behavior so accurately, it could be used to manipulate public opinion or target individuals based on their online habits and preferences.