Startup pays gig workers for AI data

A startup is paying gig workers in India to wear cameras and sensors, capturing real-world data to train AI and robotics models.

Human Archive, a startup from Berkeley and Stanford researchers, is using a novel method to gather AI training data. The company pays gig workers in India to wear camera-equipped caps and sensor devices while performing daily tasks. This equipment captures first-person video and physical interaction data, creating a large-scale dataset of real-world human activities. The objective is to supply AI and robotics labs with the high-quality information needed to build more capable and physically grounded models.

This approach tackles a major bottleneck in robotics and embodied AI: the lack of authentic, real-world training data. Simulated data can be limited, failing to capture the nuances of physical environments. By sourcing data directly from human experiences, Human Archive aims to bridge this gap. For developers and researchers, such datasets are crucial for training models to understand context, navigate spaces, and manipulate objects. This business model also offers a new template for startups in the AI data supply chain.

The project reflects a broader industry shift towards physical AI. As companies develop everything from warehouse robots to assistive technologies, the demand for data on human physical interaction is surging. This method of using a distributed workforce for data collection offers a scalable solution for gathering diverse, first-person data that is difficult to obtain in controlled lab settings. It represents a new intersection between the gig economy and the advanced AI sector.