Shift Robotics: AI Cleans Your Apartment for Free

A New Approach to Training AI with Shift Robotics

Shift Robotics, an innovative startup emerging from the German lab microagi, offers an intriguing proposition: clean your apartment for free, provided that the AI can observe. Shift's cleaning agents are equipped with front-facing cameras that capture their movements while performing household tasks. These video recordings are then used as training material for artificial intelligence labs and robotics companies.

Based in Munich, Shift was founded by Formula 1 engineers and an AI researcher. In New York, a city where free services are rare, Shift Robotics offers a zero-cost cleaning service. However, the catch is that the agents record their actions, such as doing the dishes or mopping the floor, to generate training data for the AI.

A Viral Success and Growing Demand

Before last week, Shift was relatively unknown. However, its launch video quickly garnered attention, amassing over 8 million views. The first 250 free cleaning sessions were booked in the blink of an eye. "We received thousands of booking requests," said Harry Kilberg, CEO of Shift in the United States, to Business Insider.

Shift is part of a growing trend where AI companies seek to acquire real-world data to develop machines capable of functioning in domestic, industrial, and commercial environments. Companies like OpenAI, Nvidia, Meta, and Tesla are increasingly interested in robotics.

Kilberg, who appears in the launch video, is not surprised by the excitement. "We knew our idea was revolutionary and that it would go viral," he shared.

A Global Ambition for microagi

Shift is the consumer-facing branch of microagi, a research lab founded in Munich last year. This lab focuses on developing physical AGI, or artificial general intelligence, for machines capable of interacting with the real world. The founders, Bercan Kilic and Yoan Iliev, are former aerodynamic engineers from Formula 1, and Anton Poletaev is a former researcher at the Alan Turing Institute.

According to Kilberg, Shift operates in 15 countries and employs 14,000 operators who collect real-world data. He describes this initiative as a way to accelerate the transition to an economy where everyday goods and services become more accessible.

The Challenges of Training Robots

Kilberg acknowledges that robots capable of reliably performing household tasks are not yet a reality. In the meantime, individuals can be compensated for providing the necessary data to train these systems.

Real-world data is crucial for startups and large tech companies looking to evolve AI from chatbots to machines capable of physical interaction. Large language models are trained on vast amounts of text and images online. However, a similar corpus for robots does not yet exist, prompting the industry to create it from scratch.

A Viable Economic Strategy

The offer of free cleaning raises questions about its economic viability. "Unit economics are better than one might think," assures Kilberg. The technology from microagi processes data in a way that enhances its quality, allowing it to be sold at a premium price to AI labs and robotics companies. The videos are anonymized by blurring faces and screens, and no audio is recorded. Microagi also uses this data for its own research.

Kilberg explains that the idea for this service originated from early users who were already recording their household tasks and wanted to do more. "They started offering cleaning services to their neighbors because we covered the costs," he said. Some even recorded their activities in grocery stores or soup kitchens.

Planned Expansion in the United States

New York is just the beginning for Shift. Kilberg announced that the company plans to expand across the United States and add other free or subsidized services, such as cooking and plumbing.

Shift is part of a rapidly growing market for data collected in the physical world. Companies like Scale AI, Turing, and micro1, which contributed to the rise of chatbots, are now turning to real-world data collection. The goal is to bridge what UC Berkeley roboticist Ken Goldberg calls the "100,000-year data gap."

A Diverse Data Collection

The more Shift expands its services, the more valuable its videos become. AI labs and robotics companies need diverse training data for robots to adapt to the complexities of the real world.

Shift is also focusing on geographical diversity. Kilberg notes that the company operates in countries where few others are collecting this type of data, such as Bulgaria, Georgia, and South Africa. He mentions that Shift has been particularly successful in France.