Regression Games' CEO Aaron Vontell talked about the AI features his platform offers to game developers, explained the process of real-time bot development, and discussed why game studios should embrace AI.
Image credit: Regression Games
My name is Aaron Vontell, and I'm the Founder and CEO of Regression Games. We are a team of four engineers building the future of AI agents in games. The company was founded in May of 2022, and over the past year, our team has built tools and products around intelligent agents in games (i.e. systems and tools for controlling characters in a game using code and machine learning).
I originally founded the company out of my excitement for the intersection of AI and gaming! I've been a gamer for most of my life, from playing GameCube as a kid to playing League of Legends now. I've always loved technology and was heavily involved in FIRST Robotics from 6th-12th grade. In college, I also participated in and then ran Battlecode, MIT's annual AI gaming competition. Before Regression Games, I started my career at Instabase, working on applied AI and ML for use in enterprise applications.
With all of these activities and experiences and my interest in both esports and practical AI tooling, I knew I had to start a company at the intersection of my passions. In late 2021, I and my college friend started to chat frequently about the idea of AIs for games for both fun and practical use cases, and by early 2022, I knew I had to start the company!
Aaron Vontell, Founder and CEO of Regression Games
Regression Games is building a platform that makes it easy to build useful and practical AI agents for games. When we say AI agents, we mean more broadly this idea of building logic for bots or characters in a game that can complete tasks by themselves with little human input. For example, you may want a bot that can play intelligently against human players in a game, agents that can progress through a level and test that things are working, or you may even want a bot that can coach players when they first start playing a game.
Programming these bots is hard, even for experienced game developers, and it is also an activity that developers often want to rush through, so they can focus on the core gameplay. Our platform aims to provide tools to make building these bots easier, whether it be through easy-to-use SDKs, drag-and-drop interfaces, machine learning, and reinforcement learning tools, or using generative AI to create bots for us.
On top of these AI-building tools, we also want to build the products around them that make them useful – tools for quality-testing bots, analysis tools for multiplayer engagement with bots, reports on game balancing through agent gameplay, etc. We are building our platform to be applicable to many different game genres, such as first-person shooters, battle arena games, virtual card games, dungeon crawlers, and more! We have also built the platform to be game-engine agnostic, so while our current tools focus mainly on Unity, we will also be able to provide tools for engines such as Unreal, Roblox, Godot, and others.
Image credit: Regression Games
The Evolution of AI Agents in the Game Industry
The progress of AIs in games is a fascinating topic with a rich history. Developers have always thought outside the box to develop interesting and new ways to define NPCs (non-playable characters). Besides simple-rule-based logic (e.g. if the player is X distance away, then attack), early algorithms were used to represent more complicated AI behavior. Finite state machines (FSM) and Goal Oriented Action Planning (GOAP) are two examples of popular techniques to encode the logic of an AI, due to the fact that they give a way to organize the state and actions of a bot.
More recently, deep learning and reinforcement learning have been used heavily to program bots to behave in a game based on a training process, where agents iteratively learn the best way to play a game. For example, DeepMind created AlphaZero and AlphaGo around 2015-2017 to play chess and Go, and in 2017, OpenAI announced OpenAI Five, a bot that could play Dota 2 and successfully beat many top players/teams.
More recently, not only has training these models become more accessible, but new approaches using generative AI through models like GPT have been shown to produce interesting AI agents for games, by writing code for these agents. For example, Unity and Unreal Engine now have tools to train machine learning models for agents, and works like Voyager have shown that GPT can be used to program and control Minecraft characters to complete difficult tasks.
Regression Games Agent Platform
The Regression Games Agent Platform runs outside of the core game engine and connects to the game in the same way that a human player may connect to the game. For developers, this means that your AI agent-building process is not tied to your game-build process – our system can automatically reload bot logic and apply new bot behavior in real time.
This is incredibly useful during the development of your game, because a bot can connect to your game, and you can program it to do various things without even leaving your game. For example, let's say you have a multiplayer dungeon crawler game you are developing, and two players are needed to move to the next room because one player needs to stand on a switch that opens a door. Within our tool, you can add logic for your bot to stand on the switch, and it will automatically apply this new logic within your game, with no need to restart your level. This is just one of many features that we hope will make the bot development process easier for developers.
Image credit: Regression Games
Using AI to Build AI at Regression Games
As a company that's building practical agents for games, it is only natural that we would improve our workflow and products using AI itself! As developers, we use tools like Copilot and GPT to help us write code, documentation, and other assets for our product. However, even within our product itself, we use AI to help game developers program bots. For instance, using tools like GPT, various parts of our platform generate bot code for the developer, reducing the time needed to develop an AI for a game. In the near future, we will be launching a few of these tools for developers to try out!
Using AI In Game Development
Over the past year, our team has had the opportunity to talk to countless studios about their usage of AI, both with older techniques and newer approaches such as GPT. Having worked in AI for the past 5+ years, it's been quite amazing seeing how open game developers are to improving their workflows with AIs. Compared to other industries, I find that game developers are incredibly passionate about their craft, and many of them are embracing AI not as a tool to outsource their craft of developing amazing games, but rather as a tool to help them realize their games faster.
There are a lot of new tools coming out for game developers that utilize AI – in addition to tools like Regression Games for AI agents, there are tools for creating assets such as Scenario.gg, 3D model generation with Kaedim, avatars with Ready Player Me, and so many more. With the current state of AI, these tools are helping reduce the cost and time needed to develop certain aspects of a game, in a way that allows game developers to focus on the core pieces of their craft. My main piece of advice is to simply try out these tools! A lot of them are free to try, and are easy to jump into – you may find that one of these tools speeds up your processes and improves quality with little effort needed on your part.
Image credit: Regression Games
Advice to Studios Considering Integrating AI Agents Into Their Workflows
When considering adding AI to your workflow, I encourage studios to at least try new tools all the time! Even just 15 minutes out of your day to initially test a tool can lead to large gains. However, in terms of optimal timing, I suggest integrating AI tools either at the very beginning of development or after your initial launch.
To really see if an AI tool is useful, you will need to invest some time pushing it to its limit and seeing if it will actually improve your process. From our experience at Regression Games in considering AI tools for our own processes, we've found that dedicating one person to really dive deep into a product has worked well, and sometimes it only takes a few days to really determine if a tool is worth putting more time into.
When it comes to tools that use AI, and game development in general, the two most important aspects to determine are the quality of the tool (e.g. is it generating good assets, are the bots useful?) and the integration path (i.e. the required code, SDKs, and APIs to use the tool). Since integration can be the biggest blocker to adoption, it helps to become acquainted with their documentation and support options. If you can get the help of the community or the tool developers themselves, that helps a lot as well.
Regression Games' Future Plans
Regression Games is really excited about our plans this year. We recently launched an early preview of our bot-building platform, and are working closely with developers to get feedback and provide as much value as possible. In the near future, we plan on launching new product features in both our bot-building tools and products around use cases such as QA testing, so we encourage developers to check it out!