Research has recently been conducted to further bridge the gap between machine understanding of human behavior and computer games. Six scientists, namely Joon Sung Park, Joseph O’Brien, Carrie Cai, Meredith Ringel Morris, Percy Liang, and Michael Bernstein, have incorporated large language models like ChatGPT in order to simulate human behavior and create personalities and memories. This resulted in agents, which they call “generative agents,” that can interact and pursue goals while using text communication with more natural language.
The demo of this project displays agents similar to that of The Sims – an EA game that released in 2000. This offers insight into the state data of agents, with a focus on one agent named Isabelle and her plan to host a Valentine’s Day Party.
ChatGPT has been a significant part of this project, by enabling a vast range of human behaviors to be encoded. Each agent has its own memory stream which consists of observations and reflections on the current situation. Additionally, agents are programmed with a scheduling module to set goals for their days.
The success of this experiment resulted in emergent behavior of the software. For example, as the simulation progressed, knowledge about one agent running for mayor spread to 32% of other agents by the end of two days without human intervention. Of the invited guests to the party, four had expressed interest, but did not respond due to calendars conflicts.
The research team did note that there were issues with the results. For instance, some hallucinations occurred and some agents did not recognize that they could not enter the store after it had closed. They concluded that the architecture was expensive and further steps need to be taken to address bias, inadequate model data, and safety.
The Google research group is a subsidiary of Alphabet Inc. and focuses on topics such as cloud computing, machine learning, and natural language processing. They specialize in the development of applications, computer hardware, operating systems and technologies that can benefit from Google services. Michael Bernstein is the company’s lead researcher in natural language processing and language models. His primary focus is on enabling computers to understand and generate language that accurately reflects a human conversation.
