Scientist turns people’s mental images into text using ‘mind-captioning’ technology

Scientist turns people’s mental images into text using ‘mind-captioning’ technology

https://www.aol.com/articles/scientist-turns-people-mental-images-174827640.html

Publish Date: 2025-12-31 20:19:00

Source Domain: www.aol.com

Here is a summary of the key points about the article regarding a new technique that converts brain activity into descriptive sentences using fMRI and AI:

• Mind-Captioning Technique: A Japanese scientist, Tomoyasu Horikawa, has developed a method called “mind-captioning” that uses brain scans and artificial intelligence to convert a person’s mental images into descriptive text.

• Brain Activity Analysis: Horikawa’s study involved scanning the brain activity of six participants as they viewed silent video clips. The aim was to gather data on how visual details, such as objects, places, actions, and their relationships, are processed in the visual cortex without relying on language-related regions.

• Use of AI and Decoders: Large language models were used to generate captions of the video clips. Separate AI models, known as “decoders,” matched the scanned brain activity to the captions, which were converted into numerical sequences. This allowed the AI to interpret brain activity related to video clips the AI had not encountered during training.

• Potential Applications: This novel approach has potential implications for helping individuals with communication difficulties, such as those with aphasia or ALS. It may even assist non-verbal autistic individuals.

• Ethical Considerations: The technology raises privacy concerns, as it could theoretically decode private thoughts before they are verbally expressed. Experts emphasize the need for strict regulations to protect mental privacy, especially if this technology is used outside of medical contexts.

• Challenges: While the method holds significant potential, it’s not yet accurate enough for practical use. The current approach, which relies on large data collection, cannot easily read private thoughts as it mainly works with typical, predictable mental images rather than less common scenarios.