Latent Lab - Navigating Innovation: Interactive Data Exploration & Inspiration
Latent Lab is an AI-driven, graphical knowledge exploration system. It uses text and media analysis to reveal conceptual overlaps between documents at various scales. The interface is designed for exploration rather than traditional search, aiming to inspire a graphical conversation from which new ideas will emerge.
Latent Lab includes data from the MIT Media Lab, the US Patent Office, social media posts on COVID-19, and various user-uploaded datasets. The system creates a high-dimensional vector embedding space for each uploaded dataset, which is then reduced to 2D and rendered on-screen for visualization and interactive exploration. Large Language Models are used to extract topics and sub-topics for each document, describing the map’s layout, generating summaries as users explore the map, and synthesizing new ideas from components of user-selected documents and predefined prompts. These generated ideas are shown in the latent space of existing ideas, serving as a starting point for further ideation. A timeline slider enables users to visualize the evolution of themes over time within a dataset, and a search bar allows users to identify semantic regions of the map visually.