Machine Built Video Games

Generative Content in Video Games (GPT-3)

Over the past month, the tech space has seen a lot of interest in OpenAI's new GPT-3, a general machine learning language model trained on billions of parameters and trillions of words. It's an incredibly powerful model and step-up from GPT-2 in terms of sheer training data and some very cool new applications have already been built on top of it.  

In the video gaming space, one project leveraging OpenAI's GPT models is AI Dungeon from Latitude, a text based adventure game with generative content. The story line and text is generated completely from the machine learning models and the player can respond however they like. The game is not limited by the main developer's imagination and creativity, an interesting theme that is paralleled in the user generated content (UGC) space.

Generative content in games has been around for quite some time, but it's been limited mostly to generative music, where music and audio is procedurally created from in-game interactions. You can see an example here of Otocky from 1987. One of the reasons generative content and more advanced uses of machine learning methods have not been used in video games to date is that they are inherently unpredictable.  

Game design and development today is a well crafted and orchestrated production, very similar to most entertainment: cinema, music, broadway, etc. One exception being improv, both in theatre as well as music (jazz), where most of what is performed is unplanned and unscripted.  

If you try AI Dungeon, you'll notice the experience is always different, but it's not always all that great. Unpredictability in games (and any product really) has serious negative consequences for user experience over time, which is obviously tough for game developers to accept (much less adopt).  

Takeaway: adoption of this type of iterative technology (GPT-3) in video games will be slow for larger developers until more startups like AI Dungeon prove out the reliability. However, we think the promise is there for a great user experience as this vertical of game technology matures.

Hypercasual Games are Accelerating in Mobile

The rise of hypercasual games is an accelerating vertical within the mobile gaming space. In mobile gaming in 2019, hypercasual generated about 5% of the total revenue from mobile games and 12.5% of the installs, but took 38% of the total pie for ad revenue and 76.4% of the install growth (Pocket Gamer).  

While games like Fortnite, PUBG Mobile, and Call of Duty Mobile have taken over headlines, consumers seem to be increasingly looking for bite-sized rather than immersive experiences. Average attention spans have steadily decreased over the past 15 years from 12 seconds to 8.25 seconds (The Human Attention Span).  

Hypercasual games have taken full advantage of this shift in consumer behavior. Companies like Voodoo, a prolific hypercasual game publisher, have started expanding to platforms like Snapchat that have built businesses around bite-sized entertainment. Hypercasual as a market is estimated to be worth ~$3B (+25% YoY growth) and has seen an average of 17.6M daily installs since January of this year.  

Hypercasual games have quick development and release cycles, which is why 40-50% of the total available in-game ad inventory is in the hypercasual genre. This creates a vast library of games for consumers to quickly jump in and out of while also optimizing the ad monetization funnel.