Researchers from the GrapheneX-UTS (University of Technology Sydney) Human-Centric Artificial Intelligence Centre have achieved a groundbreaking feat by developing a non-invasive method to translate human thought into text and speech.
This remarkable advancement is made possible through the utilization of a portable electroencephalography (EEG) headset connected to an advanced EEG encoder, which, in turn, leverages an in-house developed AI known as ‘DeWave’ to decode coherent, readable sentences from brainwaves.
Innovative brain-to-text translation
The core of this revolutionary technology lies in its ability to identify and capture information from raw EEG waves and subsequently process these waveforms through the DeWave AI. This sophisticated process transforms EEG signals into text, making it possible to convert thoughts directly into written or spoken language.
What sets this innovation apart is its adaptability, as it has been successfully tested on 29 participants, proving its efficacy across a diverse range of unique EEG brain signals.
Lead professor Lin, the driving force behind this innovation, emphasizes its significance, stating,
“It is the first to incorporate discrete encoding techniques in the brain-to-text translation process, introducing an innovative approach to neural decoding. The integration with large language models is also opening up new frontiers in neuroscience and AI.”
Limitless applications
The potential applications of this breakthrough technology are limited only by one’s imagination. Foremost among these is its role as a thought-to-text and speech aid for individuals with conditions such as paralysis, stroke, or speech impairments.
Additionally, this technology promises to advance the field of bionic limb development by enabling seamless, direct, and non-invasive communication between users and their prosthetic devices.
Beyond healthcare, this innovation offers the prospect of boosting productivity by allowing individuals to convert their thoughts into text and speech, eliminating the need for traditional typing or writing.
Challenges on the horizon
While the DeWave AI model represents a remarkable achievement, it has not been without its challenges. The AI required extensive training to ensure it could avoid common errors, such as homophones or synonyms, depending on the context of the sentence structure.
Despite its promise, the technology still requires rigorous peer review, similar to any other groundbreaking innovation.
Furthermore, the current system exhibits an accuracy rate of only 40% on BLEU-1, indicating that improvements are necessary to achieve optimal performance. An additional issue is the AI’s tendency to opt for synonymous pairs rather than providing precise translations for nouns.
For instance, when presented with the EEG brainwave input “the author,” the AI may inaccurately translate it as “a person.” However, the research team is confident in their ability to enhance accuracy to an impressive 90% on BLEU-1 with continued refinement.
This groundbreaking research was recently presented at the NeurIPS conference held in New Orleans on December 12th, 2023, garnering significant attention from the scientific community and the wider public.
A Non-invasive alternative to Neuralink
Comparisons with Elon Musk’s Neuralink are inevitable, as both technologies aim to enable direct brain communication. However, GrapheneX-UTS’s approach distinguishes itself by being non-invasive and portable. In contrast, Neuralink requires invasive brain implants, making it a permanent and highly controversial solution.
Neuralink’s experiments have faced ethical concerns, with reports suggesting that a number of monkeys involved in Neuralink’s primate trials had to be euthanized, contradicting Elon Musk’s claims.
Both technologies seek to empower people with paralysis to control devices through their brain activity, but the non-invasive nature of GrapheneX-UTS’s solution offers a more ethical and accessible alternative.
The researchers behind this groundbreaking technology, led by Professor CT Lin, Director of GrapheneX, along with Yiqun Duan and PhD candidate Jinzhou Zhou from the university’s faculty of engineering and IT, are keenly aware of the vast potential it holds.
While its immediate applications in healthcare and communication devices are evident, the technology’s versatility suggests a multitude of other possibilities waiting to be explored.
