Editorial illustration for Brain Waves Could Guide AI on When to Learn, Neuroscientist Says
Brain Waves Could Guide AI Learning, Study Shows
Brain Waves Could Guide AI on When to Learn, Neuroscientist Says
Andrew Ceja spent part of last week in a San Leandro warehouse pulling wooden blocks from a Jenga tower, slowly, one at a time, while wearing a headset that tracked his eyes and his brain waves. Ceja works as a "pilot" for Encord, a company that builds data tools for training AI models, and his job that day was to generate the kind of physical training data that humanoid and warehouse robots need but mostly don't have. The headset came from Zander Labs, a German neuroscience startup that thinks brain activity, specifically the signals tied to error, intent and surprise, can be turned into a data layer that makes robot training more precise.
Encord's bet is that the bottleneck in physical AI isn't the models themselves but the scarcity of real-world data to feed them. Instead of just organizing what robotics companies already collect, the company is trying to manufacture entirely new kinds of data, starting with a trial run alongside Zander to see if brain wave tagging actually changes how well a robot performs a task. Lucas Gehrke, the Zander neuroscientist overseeing the project, has been watching how much brain activity gets generated in exercises like Ceja's block-pulling.
Encord is one of a small but growing number of startups betting that the next real constraint on humanoid and warehouse robotics won’t be model architecture but instead the sheer scarcity of real-world physical training data.
Why this matters
Encord's Jenga experiment is a small-scale test of a real problem: robot foundation models are expensive to run at full power, and nobody has a great way to decide when a task actually needs that power. Gehrke's pitch, using EEG signals to flag moments of high cognitive load, gives model builders a biological proxy for task difficulty instead of guessing from pixels alone. For developers building physical AI systems, that's a potential shortcut to the kind of adaptive compute allocation that's currently handled with crude heuristics or brute-force redundancy.
We'd stay skeptical about how far this generalizes. One pilot pulling wooden blocks in a warehouse is a controlled, low-noise environment; brain activity in a factory or a home is a messier signal. Gehrke himself frames this as the "bleeding edge," not a shipped solution, and Encord is still a data-tooling vendor with an interest in selling novel collection methods. The idea is worth tracking, but the gap between a research headset in San Leandro and a scalable training pipeline is still wide open.
Related Reading
Market Trends AI drives three initiatives to accelerate global action for planet protection
Market Trends Akamai data shows AI training bots and content‑fetching bots rise since July
Market Trends NVIDIA and Google Cloud let developers scale AI from prototype to production
From Archives Trump to Announce Utility Companies’ AI Energy Pledge Thursday
From Archives NVIDIA Decoding Cuts Color Code Error Rates by Over 300X
Common Questions Answered
How does Zander Labs use brain waves to improve AI training for robots?
Zander Labs uses EEG headsets to track brain waves and identify moments of high cognitive load during physical tasks like the Jenga experiment. By detecting when humans experience cognitive difficulty, the company can flag which tasks require more computational power, giving model builders a biological proxy for task difficulty instead of relying on visual data alone.
What is the main constraint limiting humanoid and warehouse robotics according to Encord?
According to Encord, the primary constraint is not model architecture but rather the scarcity of real-world physical training data needed to train these robots effectively. The company is working to generate and collect the kind of physical training data that humanoid and warehouse robots need but currently lack.
Why is adaptive compute important for physical AI systems?
Robot foundation models are expensive to run at full power constantly, and developers need a way to determine when a task actually requires maximum computational resources. Using EEG signals to identify task difficulty provides a potential shortcut to implementing adaptive compute, allowing systems to allocate processing power more efficiently based on actual cognitive demands rather than guessing from visual information.
What was Andrew Ceja's role in the Encord and Zander Labs experiment?
Andrew Ceja worked as a "pilot" for Encord, performing physical tasks like pulling wooden blocks from a Jenga tower while wearing a brain wave tracking headset from Zander Labs. His role was to generate the kind of real-world physical training data that humanoid and warehouse robots require for effective training.
Further Reading
- Aligning Brain Waves and Machine Learning - Neuroscience News
- AI Reveals How Brain Activity Unfolds Over Time - Stanford HAI
- Brain waves could help AI know when to learn - Max-Planck-Gesellschaft
- Critically synchronized brain waves form an effective, robust and flexible basis for human memory and learning - Scientific Reports
- Using human brain activity to guide machine learning - Scientific Reports
