UC Davis Neurointerface: 1.96 million words in 19 months — a new record for everyday use
A research team from the University of California, Davis, has presented results that fundamentally change the understanding of the practical applicability of neural interfaces. A patient with amyotrophic lateral sclerosis (ALS) named Casey Harrell transmitted 183,060 sentences over 19 months — that's 1,960,163 words. The average speed was 56 words per minute.
The key difference of this work from previous ones is the focus on real, home use rather than laboratory experiments. Harrell used the system for over 3,800 hours, without researchers present. He communicated with family, friends, and colleagues, sent messages, participated in video calls, and continued working despite complete paralysis.
Accuracy and Adaptation: From Lab to Life
According to the patient's own assessment, 92% of sentences were decoded as at least "mostly correct." Formal tests showed even more impressive figures: accuracy exceeded 99% with a vocabulary of 125,000 words, with a peak rate reaching 99.2%. After assistants were allowed to independently connect the equipment, the average interaction time increased from 3.7 to 9.5 hours per day. The system was used on 444 out of 653 days after implantation.
Technological Architecture: How It Works
In 2023, Harrell had four microelectrode arrays implanted in the left precentral gyrus — the brain area responsible for speech coordination. The system reads signals from 256 cortical electrodes when the patient attempts to speak. An algorithm converts neural activity into phoneme probabilities every 80 milliseconds, and then a language model selects the most likely sequence of words. After the phrase is completed, the text is voiced by a synthesized voice tuned to Harrell's original voice before the illness.
Notably, the system also includes a cursor decoder for computer control. Previously, it was assumed that different brain areas were needed for speech and movements, but the team proved that both modes can be implemented through signals from the speech motor cortex.
From Experiment to Tool
Previous work by the same group demonstrated high accuracy in the lab: in 2024, the system achieved 99.6% accuracy with a vocabulary of 50 words after 30 minutes of training. However, the new publication shifts the focus to long-term, everyday use. According to study co-author Sergey Stavisky, 3,800 hours of recorded brain activity represent the largest individual dataset with resolution at the level of individual neurons.
It is important to emphasize that this is still a single clinical case. The system remains experimental: it uses wired connections, requires daily setup by trained assistants, and due to its size, is only suitable for home use. It is not yet clear how scalable the results are to other patients with different types of implantation and neurological conditions.
My comment as an analyst: This breakthrough demonstrates that brain-computer interfaces are moving from the stage of laboratory curiosities into the realm of real tools for everyday life. 3,800 hours of use is not just a number; it is proof that the technology can be sustainable and useful in real-world conditions. However, investors and developers should remember: the path to mass adoption will be long, and the key challenges remain miniaturization, wireless connectivity, and adaptation to individual patient characteristics.