Exploring the Future of Human Behavior with Humanoid Agents
Introduction
In a previous post, The Dawn of AI Teammates: Communicative Agents for Software Development, we dug into research that explores using AI-based communicative agents in software development. A recently published whitepaper, “Humanoid Agents: Platform for Simulating Human-like Generative Agents,” dives into the world of simulating human behavior. Authored by researchers from the University of Washington and The University of Hong Kong, this paper explores the role of Humanoid Agents in mimicking human activities and emotions.
The Core Concept
Humanoid Agents are computational models designed to simulate human-like behavior. These AI agents are not just lines of code; they have “basic needs,” “emotions,” and “social relationships.” They exist in virtual worlds and can interact with each other and their environments, offering an interesting level of realism in computational simulations.
Why It Matters
The application of Humanoid Agents goes beyond mere technological advancement; it signifies a shift in how we study human behavior. For example, social scientists can use these agents to run simulations that could help us understand complex human interactions. Furthermore, the technology has the potential to revolutionize gaming, virtual reality, and even AI ethics by creating more lifelike virtual beings.
Personal Experimentation
I tried out the demo and reviewed the code available on GitHub. I quickly spun up a virtual world with autonomous AI agents representing various characters from the television show, “Friends,” autonomously interacting with one another. The researchers use Unity to visualize the world and the character interactions in a manner reminiscent of role-playing video games such as Final Fantasy. The project provides a framework for defining different agents and a dashboard for visualizing the social relationships between the agents, including emotions and conversation details.
For the Academically Curious
The whitepaper is a modest ten pages long and available online here. I also loaded it into a “ChatGPT for PDF” service to save reading time. You can ask questions and get a summary of the whitepaper at https://askyourpdf.com/conversations/d/3335d919-9095-4e86-a2d6-0da36dd3669f.
For the Experimenters
Check out the HumanoidAgents repo on GitHub for a demo, walkthrough, and the codebase.