How researchers, including at Meta's AI lab, use AI to study EEG readings, decoding how neurons in the brain communicate and exploring the nature of cognition
Thought is ever-changing electrical patterns unconnected to individual neurons. Meta is working on a system to read your mind. Tweets: @hellohypercube , @hkanji , @wsj , @singularityumia , and @kevinokeefe Tweets: @hellohypercube : Neural nets can reprogram other neural nets if there is a communication link between them. That includes neural nets in servers reprogramming humans. Traditional computers using Ethernet etc can't do it. But any neural net using KVM link can. https://hellosemi.com/... https://twitter.com/... Hussein Kanji / @hkanji : A little less than half of the time, Meta's AI algorithm was able to correctly guess what words a person had heard, based on the activity generated in their brains https://www.wsj.com/... @wsj : Tools from cutting-edge AI systems are unlocking things about human beings which were difficult or impossible to figure out, simulate, or convincingly demonstrate before. https://www.wsj.com/... @singularityumia : Thought is ever-changing electrical patterns unconnected to individual neurons. Meta is working on a system to read your mind. 🧠 https://www.wsj.com/... via @WSJ @WSJTech @mims Kevin O'Keefe / @kevinokeefe : We all saw this. > AI research could one day lead to humans connecting with computers merely by thinking-as opposed to typing or voice commands. But there is a long way to go before such visions become reality. https://www.wsj.com/... - @WSJ's Christopher @Mims https://twitter.com/...
Context & Ripple Effects
The WSJ piece lands between two milestones in non-invasive mind reading. On one side sits the hardware-heavy lineage of Neuralink and DARPA's 64,000-electrode implant program, where decoding meant surgery; on the other, Meta's AI lab showing an EEG-based model can guess which word a person heard a little under half the time from brain activity alone.
A month later, researchers pushed the same idea further with a GPT model paired with fMRI to decode continuous language non-invasively, suggesting Meta's EEG work was an early data point in a fast-moving arc rather than a one-off. The throughline: language models turning passive brain recordings into decoded content.
First-order effects
- Meta gains a working non-invasive decode pipeline — EEG plus its AI guessing heard words at just under 50% accuracy — positioning it against implant-based programs like Neuralink that require surgery for richer signals.
Second-order effects
- The follow-on fMRI/GPT decoding result validates language models as general-purpose brain-signal interpreters, pressuring both neurotech startups and big labs to treat foundation models, not electrode counts, as the differentiator.
Third-order effects
- If decoding accuracy keeps climbing on wearable-grade signals, consumer devices that passively read neural activity become a privacy-regulation question distinct from today's medical-device framing — the gap the current regulatory landscape has not yet addressed.
The trend: Brain decoding is migrating from surgical implants toward non-invasive signals interpreted by large language models, with consumer platforms like Meta positioned to commercialize it first.