The Semantic Decoder for Human Thoughts

semantic decoder, AI, fMRi, brain activity

The Semantic Decoder for human thoughts is a groundbreaking collaboration between AI and fMRI that has the potential to translate brain activity into a continuous stream of text.

  • The AI-based system decodes a person’s brain activity while listening to or imagining a story, translating it into continuous text using non-invasive fMRI brain recordings.
  • The transcription is not perfect, capturing the overall message rather than word-for-word accuracy.
  • The limitations of the decoder prevent it from transcribing the thoughts of unwilling participants.

Humans are very curious creatures, especially when it comes to how we function. Can you blame us? Our bodies are so complex, especially our brains. Our whole being is controlled by what is essentially a pile of gooey mush. And while we know which parts of it are responsible for certain things, we actively seek to narrow it down. The popular kid on the block, AI, is the key to unlocking further knowledge in a groundbreaking collaboration with functional Magnetic Resonance Imaging (fMRI): The Semantic Decoder.

The Semantic Decoder

Research from September of last year resurfaced following its publishing in Nature Neuroscience. Titled Semantic Reconstruction of Continuous Language from Non-Invasive Brain Recordings, the paper introduces a new AI-based system, called a semantic decoder, capable of translating a person’s brain activity, as they are listening to a story or silently imagining telling a story, into a continuous stream of text. Up until this point in brain research, such decoders were very invasive, involving the surgical placement of electrodes into the brain. The semantic decoder, however, relies on fMRI brain recordings. So, you can see why people are really excited about the prospects of this device.

How Does It Work?

Researchers trained the decoder on brain activity information gathered from a person who spent hours listening to podcasts inside an fMRI scanner. Once trained, the machine generated a text corresponding to what the subject was either listening to or telling themselves by corresponding the participant’s brain activity to the original data set.

Let’s Talk Results!

This whole situation is a step forward in neurology as it is a non-invasive method that allowed the creation of a text instead of the usual single words or short sentences. But it does have some limitations. The transcription is not perfect, meaning word-for-word. The system captures the overall message and then produces a text that closely matches the original meaning.


Personally, I love ethically conscious researchers. Sure, it would feel like the technology comes with a warning label but rather from the get-go than when it’s too late. It has brought to light concerns over unauthorized access to people’s thoughts. But they do explain that its limitations include its inability to transcribe unwilling participants’ thoughts.

Final Thoughts

I’m very excited that this collaboration opens new avenues for medicine. Patients are known to sometimes either downplay or exaggerate what they are feeling. And that’s understandable, in my opinion. I can never accurately pinpoint my own pain when I go see my physician. And that does indeed affect the quality of care I receive. How are you supposed to know what qualifies as a 5 on a 1-10 pain scale? This could simplify doctor-patient interaction. Not to mention give a voice to the voiceless. Can you imagine how the quality of life of non-verbal people would improve? I wonder what the semantic decoder makes of the brain of someone with aphasia.

Inside Telecom provides you with an extensive list of content covering all aspects of the Tech industry. Keep an eye on our Medtech section to stay informed and updated with our daily articles.