Reading thought is one of the great myths of humanity, and in recent years some interested steps have been made. Recently, at the University of Texas, scientists were able to create text from scans of brain activity. The short-term objective is to enhance the communication skills of patients with specific diseases, but in the long term the potential is also wider.

It’s not entirely new, but today, voice decoders use data from surgically implanted devices in the brain, which limits their usefulness.

This new research instead takes the data from an fMRI scan (Functional Magnetic Resonance) made on the areas of the language, and makes them "translate" by a specially trained AI algorithm. You can get the general sense of the message and some isolated words, which is an excellent first result.

For example, one user thought "I don’t have a driver’s license yet" and the decoder predicted "he hasn’t started learning to drive yet".

To carry out the experiment, the original GPT language model was used, which is the precursor of the current GPT-4 model.

PUBLICITY

A fundamental difference between this work and the previous ones is represented by the decoded data. Most decoding systems link brain data to motor characteristics or recorded activity from the brain regions involved in the last stage of speech emission, mouth and tongue movement. Instead, this decoder works at the level of ideas and meanings.

Should we be worried?

The idea of a technology that can "read the mind" might raise some concerns, but it’s not that tomorrow we’ll have mind-reading machines" on the street.

The necessary equipment is complex and expensive, it is not that you can make a resonance to one passing on the street. In addition, each GPT model must be trained specifically on the individual person.

So for now we can be amazed and hope that this discovery will lead to a good step forward in the medical field, without worrying too much about robots reading our minds.

cover image: quickshooting