This is a summary of the article posted by Christian on using brain scans such as fMRIs to recreate a person’s thoughts

In the article posted on Discourse, you can read about the use of fMRIs as a non-invasive technique for deciphering cognitive activity in a person’s brain. By combining the activity recorded from the scans with predictive large language models, the researchers are able to reproduce what someone is thinking.

The study referenced in the article performed “listening” experiments with participants, asking them to report what they hear after listening to a story given by the researcher. The general findings were as follows:

• From the neural activity of the participant, the AI model could reproduce a detailed, but fairly inaccurate excerpt of what the participant heard; however, there is similarity.
• There is a long way to go for fMRI language decoding, but the progress is still significant. Using GPT-1 with the scans posed promising findings by being able to predict certain words that make sense in the context of the story.
• Although the technology is still in its early stages, privacy concerns are a pressing topic in research of this nature.

Feel free to view the full article here or see the discourse post below: