DeWave, AI Can Understand Thoughts Without Implants, Just by Reading Brainwaves

Researchers from the University of Technology Sydney created DeWave, an innovative AI system decoding silent thoughts into text using EEG waves and a snug-fitting cap. With 40% accuracy, it shows promise for communication in stroke patients and machine control, aiming to reach 90% accuracy through further refinement.

DeWave, AI Can Understand Thoughts Without Implants, Just by Reading Brainwaves

Decoding Dreams: A World-First AI System, DeWave Reads Your Thoughts Directly from Your Brain

For centuries, the ability to read minds has captivated our imaginations. From telepathic conversations in science fiction to the mystical arts of fortune tellers, the idea of directly accessing another person’s thoughts has held an undeniable allure. Now, in a groundbreaking development, researchers from the University of Technology Sydney (UTS) have taken a monumental step towards making this dream a reality. They have unveiled DeWave, a world-first, non-invasive AI system that can translate silent thoughts into text with just a snug-fitting cap.

This revolutionary technology holds immense potential to revolutionize communication, particularly for individuals with speech impairments due to stroke, paralysis, or other conditions. Imagine a world where someone who has lost their voice can simply think a sentence, and DeWave seamlessly converts it into text on a screen, enabling them to express themselves and engage in meaningful conversations. But the applications extend far beyond communication; DeWave could also empower people to directly control bionic limbs or robots with their thoughts, opening up a new era of human-machine interaction.

Unlocking the Secrets of the Brain:

DeWave works by harnessing the power of electroencephalogram (EEG), a technology that measures electrical activity in the brain. Participants in the study wore a cap equipped with EEG sensors while silently reading passages. DeWave’s AI model, trained on a massive dataset of brain activity and text, analyzed the captured EEG waves and decoded them into corresponding words and sentences.

While DeWave’s current accuracy stands at just over 40%, which may seem modest, it represents a significant 3% improvement over existing benchmarks for thought-to-text translation from EEG recordings. The researchers’ ultimate goal is to push this accuracy to around 90%, matching the performance of conventional language translation or speech recognition software.

A UTS researcher testing DeWave. (University of Technology Sydney/CC BY-NC-SA)
A UTS researcher testing DeWave. (University of Technology Sydney/CC BY-NC-SA)

Breaking Barriers and Overcoming Challenges:

Unlike previous methods that require invasive brain implants or bulky, expensive MRI machines, DeWave’s non-invasive approach makes it practical for everyday use. This is a crucial advantage, as invasive procedures carry inherent risks and are often inaccessible to many individuals. DeWave’s portability and ease of use also eliminate the need for controlled laboratory settings, potentially paving the way for wider accessibility.

However, translating thoughts into language presents unique challenges. Brain wave patterns vary greatly between individuals, and even for the same person, the way they process different types of words can differ. DeWave tackles these challenges by employing a sophisticated AI architecture that includes:

  • Encoder: This component transforms raw EEG waves into a code that captures the underlying meaning and structure of the thoughts.
  • Codebook: This database maps the encoded EEG signals to specific words and phrases.
  • Large Language Model (LLM): This powerful AI tool uses the decoded words to generate grammatically correct and meaningful sentences.

The researchers trained DeWave using a combination of BERT and GPT, two cutting-edge LLMs, on existing datasets of brain activity and text. This training data allowed the system to learn the intricate relationships between brain waves and language, enabling it to accurately translate silent thoughts into written words.

UTS HAI Research – BrainGPT (Credit: YouTube/Thomas Do)

Glimpses of the Future:

While DeWave’s accuracy for nouns currently lags behind verbs, the system still manages to produce meaningful results by generating semantically similar word pairs. This suggests that even with imperfect translations, the technology holds immense potential for communication and interaction.

The study’s large sample size further strengthens its reliability. By testing DeWave on a diverse group of participants, the researchers ensured that the system’s performance wouldn’t be skewed by individual variations in brain wave patterns.

Despite the significant progress, DeWave’s journey is far from over. The researchers acknowledge the limitations of EEG caps compared to implanted electrodes, which capture brain signals with greater precision. However, they remain optimistic about the future of non-invasive thought translation, highlighting the rapid advancements in LLMs and the potential for similar encoding methods to bridge the gap between brain activity and natural language.

The presentation of DeWave at the prestigious NeurIPS 2023 conference and the publication of a preprint on arXiv signal a new era in brain-computer interfacing. This groundbreaking technology represents a giant leap towards unlocking the secrets of the human mind and opening up a world of possibilities for communication, expression, and human-machine interaction. As DeWave continues to evolve, its implications for individuals with disabilities, healthcare, and our understanding of the brain itself are truly immeasurable. The future of silent conversations may no longer be confined to the realm of science fiction, but a tangible reality within our grasp.

The research was presented at the NeurIPS 2023 conference, and a preprint is available on ArXiv.

Google News Icon

Add Slash Insider to your Google News Feed

Source(s): Science Alert

The information above is curated from reliable sources, modified for clarity. Slash Insider is not responsible for its completeness or accuracy. Please refer to the original source for the full article. Views expressed are solely those of the original authors and not necessarily of Slash Insider. We strive to deliver reliable articles but encourage readers to verify details independently.