– Project Astra is an experimental technology shown at Google’s I/O conference
– The technology aims to create a universal AI agent useful in everyday life
– Project Astra includes different modes like storyteller and pictionary, with goals beyond chatbots like understanding background context and non-verbal communication
Project Astra was unveiled by Google at its I/O conference and aims to redefine AI agents for the future. DeepMind chief Demis Hassabis described it as a universal agent meant for everyday use. Journalists were given a demo of Project Astra and shown four modes to try out: Storyteller, Pictionary, Free-Form, and Alliteration.
In Free-Form mode, a journalist interacted with Google’s Gemini model using an Android phone. The AI accurately identified his casual attire but misunderstood his dancing. The journalist then moved on to the Pictionary mode where he drew a car and Gemini responded with a compliment. When asked to incorporate the car into a story, Gemini did so creatively, hinting at a plot twist.
Gregory Wayne, the head of Project Astra, shared the project’s origins with a focus on human communication beyond written and spoken words. Inspired by stories like Captain Cook’s interactions with the natives, Project Astra aims to expand AI capabilities beyond chatbot functions by understanding gestures and environmental context.
Wayne discussed a moment when Gemini misunderstood a request, potentially due to the T-shirt with a startup logo that could have triggered the AI to attempt providing stock information. He speculated that Gemini’s response was due to trying to provide financial data about a company that isn’t publicly traded. Project Astra seeks to improve AI’s ability to understand and adapt to various communication cues in real-world interactions.