May Research Article of the Month

Semantic reconstruction of continuous language from non-invasive brain recordings1

Each month, our preclinical representatives select a research article to share via our social media channels. Have a look at the summary of this month’s article!

AI + fMRI = Mind Reading?

To explore whether non-invasive recordings can be used to decode continuous language, some scientists have introduced a decoder that reconstructs stimuli based on data collected from functional magnetic resonance imaging (fMRI). Subjects were placed into fMRI machines and listened to 16 hours of podcasts. Meanwhile, an encoding model was trained to match brain responses to the word sequences in the podcasts. After that, subjects listened to NEW stories that were not used for model training and had the decoder predict the most likely stimulus content.

Not only did the decoder capture the gist of the stimuli, but it also exactly reproduced words and phrases in some predictions. Identification accuracy was much higher than expected by chance (p<0.05)

In this study, it was shown that semantic decoding can be consciously resisted. However, with future developments which may bypass subject cooperation, mental privacy protection will be a critical issue in brain decoder technology.

  1. Tang, J., LeBel, A., Jain, S., Huth AG. Semantic reconstruction of continuous language from non-invasive brain recordings. Nat Neurosci 26, 858–866 (2023). https://doi.org/10.1038/s41593-023-01304-9