In the #study, participants trained the decoder by listening to several hours of podcasts on an fMRI scanner, which is a large piece of machinery that measures brain activity. The system does not require surgical implants.
Once the #AI is trained, it can generate a text stream when the participant listens to a conversation or imagines themselves telling a new story. The resulting text is not an exact transcription, but the #researchers designed it with the intention of capturing general thoughts or ideas.
According to a press release, the AI produces text that closely or accurately matches the intended meaning of the participant's original words about half the time.
For example, when a participant heard the words “I still don't have my driver's license” during an #experiment, the thoughts were translated to “She hasn't even started to learn to drive yet”.
"For a non-invasive method, this is a real leap forward compared to what's been done before, which are typically single words or short sentences," said Alexander Huth, one of the study's leaders. "We're getting the model to decode continuous language over long periods of time with complicated ideas."
For now, the #decoder cannot be used outside of a lab environment because it relies on the fMRI scanner. But researchers believe that eventually AI could be used through more portable systems.
