Fully Autonomous NPCs 🤯 - Putting "Open World" To Shame (ChatGPT-Powered)

Generative Agents: Interactive Simulacra of Human Behavior

This video discusses a project from Google and Stanford that brings together artificial intelligence and video games to create generative agents, essentially characters that can live their own lives in a game world with very little input from the user.

Creating Generative Agents

• The researchers gave these generative agents about a paragraph of text to describe who they are and from that they're able to live their lives.

• Generative agents wake up, cook breakfast, head to work, paint, write, form opinions, notice each other and initiate conversations. They remember and reflect on days past as they plan the next day.

• As they experience the world they will actually form memories of the world; they'll reflect on these memories and that will inform their future behavior.

Emergent Social Phenomena

• Through all these interactions in-game emergent social phenomena occurs.

• Constantly growing memories as new interactions conflicts and events arise and fade over time while handling cascading social dynamics that unfold between multiple agents success requires an approach that can retrieve relevant events and interactions over a long period reflect on those memories to generalize and draw higher level inferences.

Applications

• One application is social role play scenarios or for an example interview preparation where somebody can rehearse safely a very difficult conversation.

• Another way generative agents can be used is with social simulations such as modeling traffic patterns or crowd behavior.

Introduction to Smallville

This section introduces Smallville, a sandbox world with a park, co-living space, bar, cafe and other amenities. It also describes the generative agents that inhabit the world.

Generative Agents in Smallville

• The community of 25 agents in Smallville are represented by simple Sprite Avatars.

• Each agent is given one paragraph of natural language description as seed memories to build their entire personality and life.

• Example: John Lin is a pharmacy shopkeeper who loves helping people. He lives with his wife who is a college professor and his son who's a student. He knows his neighbors and likes discussing local politics with them.

Conversations Between Generative Agents

This section discusses how generative agents communicate with each other in natural language just like humans do.

Examples of Conversations Between Generative Agents

• Isabella talks to Tom about the upcoming election.

• Tom expresses his dislike for Sam Moore because he thinks he's out of touch with the community.

User Control in Smallville

This section explains how users can control the world in two ways: steering the simulation through conversation or issuing directives to an agent through inner voice.

User Control Options

• Users can steer the simulation by communicating with an agent through conversation or issuing directives through inner voice.

• Inner voice prompts can inspire an agent to perform certain actions such as painting a picture.

Development of Routines for Generative Agents

This section explains how generative agents develop their own routines based on their memory stream.

Development of Routines

• Generative agents perceive something in the world and log it to their memory stream.

• They retrieve memories from the memory stream and plan or reflect on them to decide what their next action should be.

• Isabella plans a Valentine's Day party at Hobbs Cafe and forms relationships with other generative agents to invite them.

Algorithm for Deciding What's Next

This section explains the algorithm used by generative agents to decide what they will do next.

Algorithm for Deciding What's Next

• The algorithm uses recency, importance, and relevance as weights.

• The graph shows how Isabella planned a Valentine's Day party and formed relationships with other generative agents.

Introduction to Generative World

In this section, the speaker introduces a generative world where characters move around based on artificial intelligence.

Features of Generative World

• The generative world has different characters moving around.

• All 25 of these generative agents are living their life within this world.

• The game will be completely dynamic based on what all of these other generative agents are doing.

Benefits of Artificial Intelligence in Gaming

In this section, the speaker discusses how artificial intelligence can enhance gaming experiences.

Benefits of AI in Gaming

• AI can give NPCs entire lives and make the game more dynamic.

• Games like The Sims, Rimworld, Skyrim room or GTA can benefit from AI-generated NPCs.

Conclusion

The speaker concludes by inviting viewers to check out the game and leave comments.