The Generative AI Opportunity in Gaming
Yes it’s massive, but different than what many might think
In conversation with our LPs, co-investors, and strategic partners I’ve often been asked what I believe about generative AI’s applications in gaming, and ultimately its potential for exponential returns on investment. The drama at OpenAI this past week of course catalyzed even more of those discussions. Spoiler alert, I believe that our industry is perhaps the ripest and most lucrative space for the technology to demonstrate its full potential, but not quite in the way that the investor community in large part seems to believe.
The question of whether “AI will be able to make games itself” invariably arises eventually. The prospect is certainly intriguing; if AI can so coherently frame a prose narrative already and compose beautiful images, power AI characters, and stitch together video, how far are we from it being able to build together a compelling digital experience? Wouldn’t that have massive implications for even greater profitability and/or expansion of the already massive market?
Our answer is both yes and no.
Yes, in that the technology can and will have a massive impact on specific disciplines within game development due to serving as a multiplier alongside existing tools, which will translate into far more efficient content generation via human augmentation. That in turn translates into less upfront capital required for delivering compelling experiences that can gain traction, much greater profitability for these experiences at scale, and expansion of the market through more varied and personalized experiences for players.
No, in that within the liquidity timeframe we underwrite against (dollars have to generate distributions within ten years) AI will not be able to solo-generate experiences capable of competing for the attention of the player with those developed by humans.
Our AI investments thus focus on 1) working prototype builds that are already capable of delivering value to a particular segment of development and/or player experience, that need to be productionized, delivered, and translated into enterprise value; and 2) teams integrating the technology on the cutting edge within their development stack or what they are offering to players.
Some examples of generative AI tech in the portfolio and pipeline for (1):
AI level design augmentation tool that allows a human with zero experience to instantly “paint” playable spaces complete with terrain, flora/fauna, NPCs, etc. directly in-engine, and instantly drop in with friends to the gameworld to play the game loop together
AI that tracks and analyzes every action taken by players within the gameworld to integrate their behavior real-time into the lore and narrative of the game, and even tunes the gameworld’s variables in accordance with play styles
AI designer that tunes difficulty in puzzle game ($B genre) for maximum engagement, retention, monetization
And for (2):
AI texture painting in 3D environments so that artists no longer need to hand-draw the surfaces of assets
Much faster iteration on aesthetic at the concept stage with AI-generated concept art
AI agents capable of responding and taking actions autonomously rather than programmatically
We view the AI opportunity through a similar lens to other technologies that have both expanded the market as well as radically accelerated the ROI on development, such as the advent of the cloud (dedicated servers and server-authoritative gameplay paved the way for multiplayer live service $B+ annual revenue franchises) and engines (allowed developers to build on standard front-end foundations rather than starting games from scratch or with bespoke legacy tooling). Like cloud provisioning and engine-first development, the use of AI can allow developers to create more, better content with higher fidelity faster and with fewer resources.
The latter of course is the most exciting from the perspective of backing new entrants vying for market share. Startups are inherently more willing to use the latest tech for the sake of competition and differentiation, versus incumbents with legacy assets driving massive inertia. One of our most experienced portfolio CTOs left his position running the architecture of a $B+ franchise because it was never going to be able to migrate from its 20-year-old engine; the most experienced designer in the portfolio, on the other hand, has achieved a level of visual fidelity with his alpha-state game (using AI engine plugins) never before seen outside of non-playable handcrafted cinematics at his previous $B incumbent franchise.
For allocators considering where the technology has the potential to deliver portfolio-defining returns, consider this: virtual spaces in which humans behave (whether building them or playing them) are one of the richest sources of data to power reinforcement learning available to man. Every action taken by a human in building an interactive experience in a virtual space, and every corresponding action taken by players in that space, leaves digital footprints for interpolation by neural networks.
Google's AI investment arm, Gradient, led my last startup's final financing round before acquisition by Motorola Solutions not because of any AI capability we had developed to date, but because of the data generated by human pilots completing flights and tasks using our near-zero latency camera+control interface from their home/HQ devices to fly drones hundreds or even thousands of miles away. Every mouse movement and keystroke, coupled with the corresponding movement of camera and drone, generated a data point for machine learning.
In an entirely digital space, player action continuously translates to data and learning; it is no coincidence that the most heralded milestones for AI have been it learning how to play increasingly complex games. Of higher ROI, however, will be the adaptation of that capability not to the playing of a game but designing the contours of it for paying human players.