Google DeepMind Launches AI ‘Genie’ Instantly Creating Playable Games

Google’s DeepMind on Monday, February 26, held a televised illustration of Genie – a product profiled as a generative AI model. The model can create playable games by using simple prompts. 

The AI venture unveiled Genie after training on game mechanics for over 200,000 hours. The live demonstration indicated that Genie can translate images and ideas into the 2D platformer. 

The experimental model is a product of collaboration with the University of British Columbia. The model Genie is an acronym of Generative Interactive Environments illustrated to have the capacity to generate 2D platformer games. 

The model can leverage user prompts such as Super Mario Brothers and Contra via a single image. 

Genie Creating Interactive and Playable Environments

Google’sGoogle’s AI venture acknowledged generative AI’s emergence in the last few years. The space has seen several models that generate novel and creative content using images, videos and languages. DeepMind illustrated its Monday statement that it was unveiling the new paradigm for the generative AI space through Genie. 

Genie can generate interactive yet playable environments using the single image prompt. The model is a product of input from Google researchers, who consider it the latent action model capable of inferring actions between the video frames. 

Genie infers the video tokenizer and converts the raw video frames into discrete tokens besides the dynamism model that determines the subsequent frame.

Tim Rocktäschel, a developer behind the product, hailed the Genie orientation to the scale rather than focusing on the inductive biases. He indicated that the model utilizes a dataset exceeding 200,000 hours of videos comprising 2D platformers blended with the 11B world model. Genie utilizes an unsupervised approach to learning the latent actions in controlling the characters to realize consistency. 

Rocktäschel expressed excitement about the delivery by the Open Endedness Team to unveil the Genie, which he described as generating diverse action-controllable 2D games. He echoed the Google DeepMind paper that Genie can convert various media types into games.

Rocktäschel illustrated that Genie brings various human-designed creations to life, including sketches. He described that the model had proven the creation of Seneca and Caspian artwork, which are regarded as the youngest-ever creators. 

Genie Proficiency in Generating 2D Worlds

Although Genie has proven proficiency in generating 2D worlds using text or image prompts, Rocktäschel indicated that the model has additional capacity besides building the side-scrollers. The developer stated that Genie can teach AI models and agents regarding 3D worlds. 

Rocktäschel confirmed training Genie on robotics data (RT-1). It affirms the capability to learn action-controllable simulators and is a critical milestone in realizing general world models for artificial general intelligence (AGI).

AGI is identified as a singularity representing the AI with the capacity to understand and deploy learned knowledge via diverse tasks just as humans. DeepMind confirmed that the dataset utilized in training Genie was attained by filtering the publicly available internet videos. The model was trained using titles featuring the playthrough and dpeedrun, excluding the movie and unboxing titles. 

Genie to Train Future AI Agents

Google DeepMind hails the advances in AI technology facilitated by bolstered hardware and datasets. Such is behind the improved capability to generate coherent and conversational language without compromising the aesthetic appeal of images. 

The researchers carefully selected keywords to eliminate the 2D platformer gameplay videos outnumbered by others sharing similar keywords. Rocktäschel added that Genie involves a general model and is not restricted to 2D. 

Google DeepMind indicated that Genie will help in training future AI agents in the eternal curriculum of generated worlds. The research paper illustrates that the latent actions learned by Genie are transferable to real human-designed contexts. 

The unveiling of Genie adds to the subsequent additions following the GPT-4 launched by OpenAI in 2023. Technology companies led by Microsoft, Amazon, and Google have accelerated investment towards generative AI. Such is evident in Google’s move in early February to unveil the subscription-based Gemini version as a follow-up to rebranding from the Google Bard.

Editorial credit: Robert Way / Shutterstock.com  

Tokenhell produces content exposure for over 5,000 crypto companies and you can be one of them too! Contact at info@tokenhell.com if you have any questions. Cryptocurrencies are highly volatile, conduct your own research before making any investment decisions. Some of the posts on this website are guest posts or paid posts that are not written by Tokenhell authors (namely Crypto Cable , Sponsored Articles and Press Release content) and the views expressed in these types of posts do not reflect the views of this website. Tokenhell is not responsible for the content, accuracy, quality, advertising, products or any other content or banners (ad space) posted on the site. Read full terms and conditions / disclaimer.