Dreaming of Genie
Did you ever watch I Dream of Genie, the TV show? Probably not, and most likely I’m dating myself horribly. Well it wasn’t bad despite the times… Having a magical girl in a bottle to fulfill your every wish probably wouldn’t fly in today’s gender balanced biases. If you want, give it a look see and tell me what you think, but this isn’t about that… the names just match.
No this post is about artificial intelligence and where that technology intersects with games. That’s what got me into AI in the first place, and it continues to be an area where I pay more attention than usual. Recently, a new model created by Deepmind, Google’s AI research branch made new strides into game development, especially level design. This model is a second iteration of their first and appropriately named “Genie 2”.
Genie 2 can generate an “endless” number and variety of playable 3D environments. These worlds are interactive, consistent in theme, real time and are generated from a single image and a text description.
The players in these world can take actions appropriate to the environment, including swimming and interacting with objects and Non Player Characters or NPCs. The environments or levels also include animations, lighting (including reflections) and physics (like gravity). The model can generate games from a variety of perspectives including 1st and 3rd persona as well as isometric views.
Unfortunately, the model and those like it (more on that shortly), aren’t very good at remembering their own levels, which means your progress gets erased as the model continues to render the levels. So there’s a ways to go, but in the meantime the hope is that it can be used for creative expression, research, training, including the training of AI agents.
We can’t peer behind the curtain to see how it makes such magic, but the output resembles AAA titles, minus some resolution. It’s possible that a lot of the training content was pulled from Google’s YouTube service, but we might never know and Google isn’t very vocal on the subject until their lawyers advise them to talk.
Google continues to poach talent, especially in the area of video and games, including Tim Brooks and Tim Rocktaschel who come from OpenAI and Meta respectively. The video games industry is brewing some discontent while advancing this technology as the big box game companies are already laying off workers and using Gen AI to try to fill the gaps. Obviously the technology isn’t there yet, and hopefully the gamer community can support the humans still trying to make games.
Despite our conflicts with capitalism, hopefully we will find a way to maintain our standard of living and still research and develop AI that can help create magical digital worlds to play in. The two aren’t mutually exclusive, and it’s best we wise up to the necessity of a universal basic income, or something to help humans compete with the quickly evolving technology. For more on that, you will have to wait, because we haven’t written it yet.
A lot of this information was parsed from the Tech Crunch article I’ll list below along with a few of the screeenshots taken from Genie’s output.
