The Download: gig workers training humanoids, and better AI benchmarks
Thousands of workers in 50+ countries record chores to create crucial training data for humanoid AI.
A new, global workforce is emerging to train the next generation of humanoid robots. Companies like Micro1 are hiring thousands of gig workers in over 50 countries, including Nigeria, India, and Argentina, to record themselves performing everyday chores. These videos, captured by workers strapping phones to their foreheads, have become a hot commodity for robotics firms racing to develop capable humanoids. While the work pays well by local standards, it introduces complex ethical dilemmas around data privacy and informed consent for the individuals involved.
Separately, a significant critique is mounting against how we evaluate AI. Professor Angela Aristidou argues that traditional benchmarks, which test AI on isolated problems, are fundamentally broken because they don't reflect real-world use. AI operates in messy, collaborative human environments over extended periods. This misalignment means we misunderstand its true capabilities and risks. The call is for new benchmarks, like Human–AI, Context-Specific Evaluation, that assess AI's performance within actual teams and workflows over longer time horizons.
- Micro1 hires thousands in 50+ countries to record chore videos for robot training data.
- Experts declare traditional AI benchmarks broken, needing new tests for real-world collaboration.
- The gig work raises major ethical questions about privacy and informed consent.
Why It Matters
This shapes how robots learn human tasks and forces a rethink of how we measure AI's real-world impact.