The inner monologue in the brain was originally a little secret that belonged only to themselves, no one else would know - but now, this secret has been cracked by scientists.

In the recent "Nature Neuroscience", published a "mind-reading" feeling research results: researchers have developed an AI "decoder", just use the instrument to scan brain activity and then handed to the decoder analysis, you can decode a person's mind The words that appear in a person's head!

Researchers from the University of Texas have developed this AI decoding system. It can analyze functional MRI data from the human brain and decode what the subject is listening to or thinking in their head. Functional MRI measures changes in blood flow to various parts of the brain, and it reflects the level of neural activity.

Before decoding, data is first collected to train the model. The researchers brought in three subjects and had them listen to 16 hours of a podcast storytelling program while their brains were scanned. This resulted in brain data corresponding to specific utterances, and the decoding model would learn the patterns.

Once the training is complete, the decoding model can analyze the new MRI scans and interpret the utterances that are emerging in the subject's brain. The system also draws on the power of natural language models, which are responsible for predicting the most likely words based on the above and helping the decoding model narrow down the alternative answers.

The ultimate goal of developing such a "mind reading" system is to help people who have lost the ability to speak due to illness, so that they can use their brains to communicate with the outside world without barriers. Although the textual details of the AI decoding results are very different from the original text, far from the accuracy of the automatic recognition of subtitles.

However, it is still much better than random guessing, and quite often reflects a similar general idea as the original text. For example, one participant listened to the original sentence "I have not yet obtained a driver's license", which became "She has not yet begun to learn to drive" in the decoding results. About half the time, the decoding produced words and phrases that were very similar in meaning to the original.

Although not perfect, this decoder is still an improvement over previous studies. On the one hand, it can be done with a non-invasive examination, no longer requiring surgical implantation of electrodes in the brain; on the other hand, it can decode large passages of text more freely, instead of recognizing words and phrases only to a small extent.

The technology to decode the brain is still in its infancy, but it does show the possibility of deciphering the human mind through instruments, which also raises new ethical concerns. Will bad guys use the new technology to steal private information from people's brains? At least at this stage, we don't need to worry about it yet. Current decoding techniques require a high degree of cooperation from the subject, and it is not possible to "read" a person's mind without their consent.

This is because the characteristics of each person's brain MRI data are different and the rules are difficult to generalize. Only when the data of the person concerned is collected in advance can the model be trained to ensure a high accuracy of the decoding results. If we force to train with other people's data, we can only get a very low accuracy and almost garbled interpretation result.

Moreover, this data collection process requires the subject to listen to the audio very intently in order to complete. Once distracted, the brain activity changes and can no longer correspond to the audio utterance. In other words, if someone really kidnaps you and forces you to read your brain, you just need to turn a deaf ear to the test audio while collecting data, and then desperately think about what to eat at night is enough to crush his plot.

Of course, as technology advances, decoding neural activity may become even easier in the future. Therefore, the researchers also called on policy makers to get involved and consider introducing relevant policies as soon as possible.