Imagine a computer reading your thoughts about a picture you just saw — not from an MRI machine, but from a simple cap covered in electrodes. That's exactly what a new study from researchers at multiple Chinese universities and the University of Cambridge has achieved. Their model, called MindAlign, can decode which of 200 images a person is viewing just from their EEG brain signals, with 54.1% accuracy — far above the previous best of 32.4%.
The Research
The team, led by Zexuan Chen and colleagues, built a three-part AI that connects EEG recordings, images, and text descriptions in a shared virtual space. They first trained an EEG encoder on unlabeled data to learn general brain patterns, then aligned it with visual and textual representations using contrastive learning — essentially teaching the model to match brain signals to the correct image and its description. The model uses innovative components like subject-specific adaptation (tailored to each person's brain) and graph-attention over EEG channels. On the Things-EEG2 benchmark, MindAlign achieved 83.4% top-5 accuracy, meaning it often included the correct image among its top five guesses. They also tested on MEG data, showing the approach generalizes across brain recording methods. Statistical tests (paired Wilcoxon, p < 0.01) confirmed the improvements were not due to chance.
Why It Matters
This research brings us closer to thought-to-text or thought-to-image interfaces that don't require expensive, bulky MRI machines. EEG caps are portable and relatively cheap, making them accessible for everyday use — think helping people who cannot speak communicate, or new ways to interact with computers. The finding also aligns with known neuroscience: the model's decoding patterns match early visual processing stages, suggesting the AI is actually learning meaningful neural signatures, not just statistical noise.
What You Can Do
While you can't try MindAlign at home yet, you can explore your own visual processing speed with the free IQ test at iqgenio.com, which includes tasks that challenge pattern recognition and visual memory — skills that this research shows are deeply tied to brain signals.
Source: arXiv q-bio.NC
Curious about your own brain? Take our free adaptive IQ test or try 306 brain training levels.
