Physical AI and Snabbit’s Role in Data Collection

Physical AI is becoming a significant factor in data collection, as highlighted by the recent emergence of Snabbit, a startup within the home services sector. Unlike Pronto, however, Snabbit has opted to walk away from their discussions.

Human Archive’s Proposal to Snabbit

Earlier in the year, Snabbit received an outreach from Human Archive, a startup dedicated to creating datasets for embodied intelligence and robotics systems. The aim of Human Archive is to gather multimodal human activity data, helping AI systems comprehend and emulate real-world tasks. The platform refers to various technologies such as egocentric video capture, motion tracking, hand tracking, pose estimation, tactile sensing, depth mapping, and action annotations. Potential environments for data collection identified include homes, hotels, restaurants, retail spaces, and industrial sites.

The Stakes Are Rising

This development signifies a shift beyond traditional methods of AI companies retrieving internet data. The new focus is on physical human behaviour—activities such as cleaning, cooking, laundry, and navigating spaces. Startups operating in home environments are increasingly positioned at the centre of this burgeoning opportunity.

Snabbit’s Engagement with Human Archive

Human Archive approached Snabbit with what was described as an exploratory engagement. Both companies confirmed to Startup Superb that an NDA was signed before any discussions commenced. Snabbit noted that they reviewed “preliminary draft concepts” and carried out a “limited assessment” in a controlled training environment. Nevertheless, Snabbit clarified that they did not move forward with any proposal.

No Formal Deployment

Snabbit clarified that there was no pilot test, no rollout in customer homes, no operational deployment, and no sharing of customer data. Human Archive corroborated this, stating that the dialogues remained exploratory and no formal rollout in customer homes occurred.

Implications of Real-world Behavioural Data for Physical AI

The overarching trend is clear: Physical AI systems require real-world behavioural training data, which generative AI models do not. Robots cannot master household tasks purely from text; they necessitate direct interaction with environments involving kitchens, appliances, and everyday human activities.

Strategic Value of Home Services Platforms

Platforms like Snabbit hold strategic importance as they already function in real-world settings, employing distributed workforces to generate structured human task flows daily. Raj Patel, co-founder of Human Archive, remarked on social media that the economic factors may soon compel startups to enter this space. He claimed that Human Archive reached out to multiple companies, including Urban Company, which declined the offer. However, it later collaborated with a smaller entity, providing subsidised services in exchange for consent-based recording. Patel reported that nearly 98% of users opted for these cheaper services, leading to a swift increase in bookings.

Ethical Considerations Surrounding Consent

The situation raises vital ethical concerns regarding consent, especially in a sensitive market like India. When consumers trade privacy for discounted services, the meaning of consent becomes murky. Can it genuinely be considered freely given if linked to economic advantages? Are users informed about how their recordings may be used, whether for operational monitoring or as training data for AI systems?

Regulatory Oversight

The Digital Personal Data Protection Act, 2023 emphasises that consent must be specific, informed, and purpose-bound. If recordings made during discounted services are utilised for AI training, the distinction between service delivery and data extraction blurs. If such practices gain traction, India’s low-cost labour market could inadvertently become one of the largest training grounds for global robotics companies.

Addressing Questions on Data Practices

An urgent question arises from this discussion: which company may have conducted large-scale recording inside homes to create Physical AI training data? Was there proper communication with consumers about how their data would be processed or monetised? These are questions that need addressing by regulators like India’s Data Protection Board and the Ministry of Electronics and Information Technology.

Snabbit’s Position on Data Usage

Startup Superb reached out to Snabbit to determine if workers were informed that their movements could be used as AI training data. Snabbit responded that no such framework exists on their platform and that the earlier assessment was voluntary and based on consent. They also stated that they have not shared customer data outside India or with Human Archive.

Future Directions for Snabbit

Snabbit maintains a focus on operational excellence and customer trust, asserting that it is not pursuing any strategy involving data recording within customer homes. Abhiraj Bhal, another key figure, reiterated that the company has no intention of engaging in such activities.

The Growing Interest in Physical AI Partnerships

The attention that various companies are now receiving from AI firms seeking Physical AI-linked data partnerships illustrates the increasing demand for real-world behavioural datasets. As it stands, homes in India are becoming a significant interest area for technological advances in this realm.