Bitpart.AI

Multi-Agent. Games are multi-agent. The world is multi-agent. And Bitpart.AI is natively multi-agent as well. The western saloon demo that we built to show to folks at last GDC exhibits several aspects of multi-agent simulation: ● Characters navigate the environment. ● Characters can carry props from place to place. ● Characters can proactively initiate conversation or interaction with other characters. ● The AI Director organizes subgroups of characters into casts, focused on shared activities. ● Characters have configurable and dynamic affinities toward other characters.  ● Characters can hear nearby conversations, and chime in if they have something relevant to add. ● Conversations can involve 3+ characters, including human players. There are numerous decisions to make when designing a multi-agent system.  Is each agent fully autonomous, and group behaviors are emergent?  Or is there an explicit hive mind that can organize and orchestrate?  How can simulated characters collaborate with unpredictable human players? If you’d like to delve into these topics, take a look at our Substack, and sign up for our Tech Preview.  Links in the comments below.

We’ve been working on some cool character tech since late 2022.  We started development about a month before the debut of chatGPT. The release of ChatGPT ushered in a wave of developers and startups focusing on chatbot NPCs.  While this new capability was intriguing, we decided to stay the course with our focus on AI characters designed to participate in core 𝐠𝐚𝐦𝐞𝐩𝐥𝐚𝐲 in mainstream PC and console games. Our characters are aware of their surroundings, interact with props and furniture, execute interruptible multi-step plans.  They do converse as well, of course.  Players can strike up a conversation with our characters at any time.  But our dialogue content is authored by designers (with optional AI assistance), rather than generated on the fly, ensuring meaningful, grounded interactions.  For some applications, like XR, open-ended chat based dialogue could be beneficial, for example enabling hands-free voice input. It is possible to layer in this mode of interaction, integrated with our behavioral engine. We started showing a demo of our characters to studios at GDC 2024, behind closed doors.  We’d like to start sharing quick peeks at what we’re building more publicly in short video clips.  To get started, here is a brief look at a bartender in a western saloon that we created in March for GDC, running with the Bitpart Unity SDK. I’ve written a bit more about our bartender on Substack.  If you’d like to try our demos or explore Bitpart.AI for your own project(s), we are signing people up for our Tech Preview.  Links to both in the comments below.

We spoke with MIT Technology Review about how AI is enabling improvisational casts of characters for games. Niall Firth does a good job of highlighting the tension between content generation and human authorship -- NPCs who could support games with infinite content, and NPCs who participate and converse flexibly within the bounds of human-directed stories. We're in good company in the article alongside others exploring and inventing in this new frontier, including Inworld AI, Character.AI, modl.ai, and Hidden Door. Article linked in the first comment.

My talk is now live in the GDC Vault. I had a short 7-minute segment titled “Creating Planners from Examples” within the AI Summit: Simplest AI Trick in the Book session. The talk gives an overview of our approach to learning HTN planning domains, and how we combine HTNs and LLMs.  I end with a peek at a fun prototype we built called Prompt-to-Action. When we use the term Action, we mean two things: First, we mean a performance, like “Lights, Camera, Action!” Second, we mean physical action in the environment — characters walking around, sitting on furniture, picking up props and using them in different ways, and using body language to interact with each other. The Prompt-to-Action prototype uses Unity, OpenAI GPT-4, ElevenLabs voices, and the Bitpart.AI SDK. More details are in our Substack post.  In related news, we now have a Substack. Links to the GDC Vault and our Substack in the comments 👇

To celebrate 10 years of AI and Games, there was one topic that warranted being revisited one last time: the AI of Monolith's 2005 shooter F.E.A.R. But this time we're doing something special. Over the past 3 months I've sat down and had several (online) interviews with Jeff Orkin, Ph.D.: the AI programmer on F.E.A.R. In this hour-long retrospective, we get into the backstory of how the AI of F.E.A.R. came to be. The academic influences, the crazy risks being taken, the lack of oversight, a pivot from 60s espionage to sci-fi horror, the creativity of the development team behind it, and the legacy the game has in the video games industry. A huge thank you to Jeff for participating in these interviews. I've had a great time listening to his stories. I learned a lot, and equally am excited for my audience to hear what he has to say. On a more personal note, F.E.A.R. was massively influential in my own career path and without it, my trajectory through grad school would have been drastically different. And of course, given it was the inaugural episode of AI and Games, without Jeff's contributions my YouTube channel may never have been created. I hope you enjoy it! Watch on YouTube: https://lnkd.in/eCG4466t Read on Substack: https://lnkd.in/e3Rtu5PS