This is The Age of AI Series, where we talk to the foremost entrepreneurs and innovators around the planet using ML to transform industries. (Join our special mailing list!)
Video games are a bigger industry than movies, sports, music and several other entertainment forms COMBINED. It’s also much riskier to build and sell good games.
Today, we discuss how AI shows a path for the future of this business. As competition heats up, game development will benefit from machine learning.
I interviewed Christoffer Holmgård, CEO and Founder of Modl AI, a software company that builds cutting-edge ML tools for game developers so they can be more profitable and scalable!
- 01:36 — How the games industry is structured (studio sizes, types of games, etc)
- 07:09 — The varying economics of game dev for different game types
- 11:05 — How game dev can be more profitable and less risky!
- 14:50 — Why video game prices are trending towards zero
- 18:00 — How machine learning is revolutionizing the games industry
- 25:55 — Types of ML models used by Modl AI
- 29:43 — Evolution of Modl AI’s team and product
- 32:00 — Lessons from building a reinforcement learning product: don’t go too hardcore
Aman’s 2-Minute Summary and Key Takeaways
To understand how AI is changing the way games are made, let’s first understand the economics of game dev:
- Games are of many kinds: from the simplest casual games like Candy Crush and Angry Birds, to complex RPG games like Fortnite or Assassins Creed. Game dev studios can be as small as indie teams of 2-3 people to several hundred people.
- Most games fail. The cost of buying a game to play is also quickly trending towards zero, so most games are free to download, focusing on in-app purchases and other ways to monetize. This means the gold is in retention and engagement.
- You have to make several games as quickly and cheaply as possible, to find one that becomes a hit. When you DO get a hit, you suddenly have to pour more money and resources — to retain your hard-earned gamers, you must scale up, and quickly add new levels and elements — while maintaining high quality.
- Game dev is a creative process, but most of it is notoriously hard, manual, low-paid labour. A big part of this is QA — testing the game for bugs and issues (which anger gamers like nothing else).
By using ML to automate as many parts of game dev as possible, you greatly reduce the risk factor in game dev. This will help the studio business be much more sustainable and profitable.
Modl AI is one company that builds such plugins and APIs for game developers.
Given that it’s a complex problem, their technology is new and cutting-edge, and obviously can’t work for every kind of game. They currently serve a niche market of games that, one, are technically a good fit for ML agents to work on, and two, make a lot of commercial sense.
Here’s how it works:
- The first challenge is building bots that play newly added levels on their own, trying to find glitches. For very simple games, you can hand-code the bots — but such bots are not only myopic, but also unscalable. For complex games, you want bots agents which can intelligently explore environments to find hard-to-find glitches.
- The second challenge is building bots that not just play the game, but play impressively against human players, and keep learning to get better! This means training the bots using deep reinforcement learning and other techniques.
- The third challenge is to create new levels automatically, which keep your hard-core gamers interested. They do this by feeding manually-created levels to train a “level generator” AI model. I suspect that this is the toughest technical challenge of the three, and works for only very simple games as of now.
I was very impressed with Christoffer and their incredibly logical approach to attacking their market. He’s run a game studio himself in the past, and also has a PhD in ML, and he deeply understands the issues of his customers and how to choose the right ones to solve.
I think we’ll hear a lot more about Modl AI and activity in this space in the near future!
(Ethics Policy: These opinions are 100% my own as an independent observer and educator. I don’t own stock in guests’ companies or their competitors, nor do I get paid by them in any form for any reason at the time of publishing, unless specifically stated. Episodes are also not intended to be an automatic endorsement of any company or its products and services.)