Researchers from the GrapheneX-UTS (University of Technology Sydney) Human-Centric Artificial Intelligence Centre have demonstrated a non-intrusive way of translating human thought into text and speech. The researchers say the portable device uses an in-house developed AI called ‘DeWave.’
The feat is achievable using a portable and non-invasive electroencephalography (EEG) headset connected to a multitasking-capable EEG encoder to detect coherent, readable sentences from brainwaves. It identifies and captures information from the raw EEG waves and passes them through a specially designed and trained AI to process waveform EEG signals into text. Importantly, this innovation was tested on 29 participants, showing that this technology can be used with different people — this is important because all individuals have unique EEG brain signals.
Lead professor Lin said, “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.”
The application of this is practically limited to one’s imagination. The most obvious one will be its use as a thought-to-text (and speech) aid for those with paralysis, stroke, or any injury affecting their speech. The researchers also say this would further bionic limb development in combination with this technology, as it will allow seamless, direct, and non-invasive communication between the user and the machine. In other applications, it will help people convert thought into text (and speech), hence being more productive than typing or writing.
The Challenges Ahead…
DeWave’s AI model needed to be trained to ensure it did not make simple rookie AI mistakes like using ‘too’ instead of ‘two,’ all depending on the context of the sentence structure. While it may look very well-made and ready to be mass-produced and deployed, the tech still needs to go through the peer review process like any other innovation. The researchers also say there are challenges as the current system is only 40% efficient. There are other issues, too, such as the AI tends to prefer synonymous pairs instead of using precise translations when it uses nouns. For example, if the EEG brainwave specifies ‘the author,’ the AI would mistranslate that as a person. The team is confident it can boost its efficiency up to 90%.
As of now, the paper has been presented at the NeurIPS conference that was held in New Orleans on 12th December 2023.
Better Than Elon’s Invasive, Permanent and Controversial Neuralink?
One couldn’t help but compare this non-invasive application with Elon’s Neuralink, which is a permanent and extremely invasive system that requires brain implants. The Neuralink experiment has been controversial as Wired found out that at least a dozen monkeys, which were Neuralink’s primate subjects, had to be euthanized, contradicting Elon’s claims. Neuralink also aims to resolve similar issues, as it allows people with paralysis to control devices with their brain activity.
The researchers made it clear that their new technology can be used in the development of artificial appendages, but its clear that it could be used for a wide range of other uses as well.
The research was headed by Professor CT Lin, the Director of GrapheneX, and includes Yiqun Duan and a PhD candidate, Jinzhou Zhou, from the university’s faculty of engineering and IT.