Google Wants to Simulate the World With AI

Apparently not content with its takeover of the world, Google is in the process Staff up Its DeepMind research lab develops generative models capable of simulating the physical world. The project—which will be led by Tim Brooks, one of the people who helped create Sora, OpenAI’s video generator—will be an important part of the company’s efforts to achieve artificial general intelligence. job list Belonging to the new party.

Brooks, who later joined DeepMind Escape from OpenAI Back in October, and his team has “ambitious plans to create massively productive models that mimic the world.” According to the role description, efforts to create world models “will power many domains, such as visual reasoning and simulation, planning for embodied agents, and real-time interactive entertainment.” If you are willing to take on one of these roles, you can understand what that ambiguity means and get back to us.

A world modelPut as simply as possible, one usually wants to simulate how the world actually works. Generative models like Sora are able to replicate things it has seen before in its training data, without having any real understanding of why that thing is happening. So it can successfully create a video of a person throwing a baseball, but it has no idea of ​​the physics of what’s going on. The goal of the world model is to equip the machine with enough information to parse through how an action occurs and its possible outcomes.

Ian LeCun, Meta’s chief AI scientist, described the world model this way Speech at the Hudson Forum Earlier this year: “A world model is your mental model of how the world behaves…you can imagine a sequence of actions you can take, and your world model will allow you to predict what effect that sequence of actions will have on the world.”

World models are difficult to build for a number of reasons, including the large amount of compute required to run a model and the lack of sufficient training data to build an accurate model, resulting in most world models only working for limited and specific contexts.

DeepMind’s team seems intent on expanding the world model further. The plan is to build “real-time interactive generation” tools on top of the models and potentially see how they can integrate their world model into Google’s large language model Gemini.

One possible area that DeepMind will try to tackle is video games. The new team’s job description states that they will collaborate with the Veo and Genie teams at Google. the genie Video generators like Google’s Sora and Genie are existing world models that can simulate 3D environments in real time. The video game industry already Eager to adopt AI toolsThousands of workers displaced. A CVL Economics The study found that more than 86% of all gaming organizations have already adopted generative AI tools and that by 2026 around 15% of all gaming jobs could be disrupted.

Maybe improving this world would be a better use of time than modeling it.