Google’s AI Genie Transforms Imagination into Playable Games – Unveils Groundbreaking Technology
In a groundbreaking demonstration on Monday, Google’s AI venture DeepMind introduced Genie, an innovative generative AI model capable of transforming simple prompts into fully playable 2D platformer games. Developed in collaboration with the University of British Columbia, Genie, short for Generative Interactive Environments, utilizes over 200,000 hours of gameplay videos to create immersive gaming experiences.
The experimental model, as showcased in a live demo, operates by learning game mechanics from an extensive dataset, enabling it to generate side-scrolling games reminiscent of classics like Super Mario Brothers and Contra. Users can prompt Genie with a single image, and the AI leverages a latent action model, video tokenizer, and dynamic model to create interactive environments.
According to Tim Rocktäschel, a developer at Google DeepMind, Genie goes beyond the realm of 2D side-scrollers. The AI model can convert various media types into playable games, including action-controllable virtual worlds generated from different inputs. Rocktäschel demonstrated Genie’s ability to bring human-designed creations, such as sketches and artwork, into playable 2D worlds.
“Genie can bring to life human-designed creations such as sketches, for example, beautiful artwork from Seneca and Caspian, two of the youngest ever world creators,” said Rocktäsche.
The potential of Genie extends beyond gaming, as the AI model showcased the ability to teach other AI models or “agents” about 3D worlds. By training Genie on robotics data (RT-1) without actions, the developers demonstrated the capability to learn an action-controllable simulator, a promising step toward artificial general intelligence (AGI).
Google DeepMind emphasized that the dataset used to train Genie was generated by filtering publicly available internet videos, specifically focusing on titles like “speedrun” or “playthrough.” This meticulous approach ensures that the AI model learns from relevant 2D platformer gameplay videos.
“With Genie, our future AI agents can be trained in a never-ending curriculum of new, generated worlds,” Google DeepMind stated. The proof of concept in their research paper showcases how Genie’s latent actions can transfer to real human-designed environments, hinting at the vast possibilities awaiting exploration.
As technology companies, including Google, Microsoft, and Amazon, heavily invest in generative AI, Genie stands out as a testament to the continuous evolution of artificial intelligence. The launch of OpenAI’s GPT-4 has further accelerated advancements, with Google introducing a subscription-based version of its Gemini AI model.
The future seems promising for Genie, offering a glimpse into a world where AI seamlessly integrates creativity and gameplay. Stay tuned for more updates on the transformative potential of Google’s Genie AI.
In a groundbreaking demonstration on Monday, Google’s AI venture DeepMind introduced Genie, an innovative generative AI model capable of transforming simple prompts into fully playable 2D platformer games. Developed in collaboration with the University of British Columbia, Genie, short for Generative Interactive Environments, utilizes over 200,000 hours of gameplay videos to create immersive gaming experiences.…