Google and Stanford University researchers recently conducted a study demonstrating the potential of using ChatGPT to create “generative agents” that simulate human-like behaviour.
In this study, 25 ChatGPT instances were used to create fictional characters living in a fantasy town, with each instance assuming the role of a specific person and conversing with other ChatGPTs to recreate the town.
The generative agents were prompted with specific information about their character, such as their job, family, and relationships with others. Based on this information and the situation presented, the agents output realistic actions, such as brushing their teeth, getting dressed, and engaging in conversation.
While the generative agents do not physically move around the virtual environment, they can infer human behaviour and interact with each other to mimic real-life interactions. This has a wide range of potential applications in games and simulations of human interactions in virtual environments.
However, this approach is still in its early stages and requires further refinement. The use of complex text layers can introduce delays, and the study has not yet been peer-reviewed or formally published.
Nonetheless, this experiment with ChatGPT highlights the potential of current technology and provides a valuable reference for future research. As technology advances, it will be interesting to see how ChatGPT and similar generative agents develop and are applied in various industries.