top of page

Generative AI Bridges the Gap Between Thought and Text

12/24/23

Editorial team at Bits with Brains

In a groundbreaking development from the University of Technology Sydney (UTS), researchers have taken quite a leap forward in both neuroscience and generative AI.

They have introduced a cutting-edge system capable of 'reading' thoughts and transcribing them into written text through a sophisticated interface that captures brainwaves via an EEG cap. Unlike other attempts, this innovative technology stands out for its non-invasive approach, bypassing the need for intrusive surgical implants or the excessive costs associated with MRI scans.


At the heart of this system is DeWave, an AI model carefully trained to parse through extensive datasets of EEG information. The model does what large language models do well – translate information. The AI converts the complex electrical activity of the brain into comprehensible words and sentences. By associating specific patterns within the EEG signals to linguistic elements, DeWave enables the articulation of silent thoughts with promising precision.


The current translation proficiency of this AI system stands at a 40% accuracy rate. While this figure may initially appear modest, it is nonetheless a significant step towards a future where our thoughts could seamlessly convert into text, promising a revolution in how we communicate, especially for those unable to speak due to various disabilities.


The research team notes that the AI model sometimes opts for synonymous word pairs rather than precise translations—illustrating with the example of interpreting "the author" as "the man." It’s likely this phenomenon stems from the brain's processing of semantically similar words, which can produce analogous brainwave patterns.


Despite these linguistic and technical hurdles, the results are substantial. With a sample size of 29 participants, the researchers underscore the adaptability and robustness of their decoding technology, which represents a marked improvement from previous efforts that relied on a far more limited dataset. The goal is to achieve a 90%+ efficiency rate.


This novel use of AI is allowing us to construct a bridge between thought and language, and it is only a matter of time before we will be able to cross it, ushering in a new form of human-computer interaction.


Sources:

https://interestingengineering.com/innovation/mind-reading-ai-thoughts-text

Sources

bottom of page