The Good AI Is Doing · November gratitude series
A man with ALS attempted to speak, and an experimental system turned signals from his brain into audible speech. UC Davis researchers reported the work in June 2025, demonstrating a way for artificial intelligence to help a person whose speech had become extremely difficult to understand. The research involved one participant, and the technology remains investigational. That is already a result worth paying attention to. 1
After a month spent looking at AI through Halloween, horror, and the things that could go wrong, I want to spend November looking closely at the good it is doing. There are real reasons to be grateful for this technology, and I think we owe those reasons the same attention we give the threats.
The UC Davis work gives us somewhere specific to begin. We can look at what the researchers built, what the participant could do with it, and how much work remains. Gratitude becomes more useful when we can explain what we are grateful for.
When speaking becomes difficult
ALS can weaken the muscles involved in speech. Communication tools already include letter boards, eye tracking, and computer-generated speech. Each offers ways for people to express themselves as their physical abilities change. The National Institutes of Health describes these approaches alongside research into brain-computer interfaces, which use brain activity to help operate a device. 2
When I think about speaking, I tend to think about the words. Reading about this research makes me think about the decisions wrapped around them. There is the moment you want to enter a conversation, the word you want someone to hear more strongly, and the change in your voice that makes a sentence sound like a question. Those are ordinary parts of how we make ourselves understood.
A communication system has to deal with those details as well as the words themselves. A person may have a sentence they want to say while the physical act of speaking has become difficult. The practical question is how to carry that attempted speech into something another person can hear.
What the researchers built
The participant had four implanted microelectrode arrays that recorded activity in brain regions involved in speech. A trained AI decoder used signals associated with attempted speech to predict acoustic features, and a speech synthesizer converted those predictions into sound. The participant could hear the output continuously. This was decoding attempted speech under research conditions, with implanted equipment and a trained system. 3
UC Davis also reported demonstrations involving changes in intonation, emphasis, and simple melodies. Those capabilities matter because expression carries meaning alongside vocabulary. The university described the project as a collaboration involving neuroscientists, engineers, and clinical researchers, including Maitreyee Wairagkar, Sergey Stavisky, and neurosurgeon David Brandman. Its announcement came from UC Davis Health in Sacramento. 1
I appreciate being able to follow that chain of work. The participant attempts to speak; the system records activity, makes predictions, and produces sound he can hear. AI has a defined job in that process. Its contribution can be examined in terms of whether the resulting speech helps someone communicate.
That is a useful standard for the rest of this gratitude series, too. When we describe an AI breakthrough, we should be able to explain the task it performed and the human benefit that followed.
What the results actually mean
The study used different listening tests, and their results need to stay attached to those tests. Across 956 held-out sentences, listeners choosing among six sentence options achieved a mean accuracy of 94.34%. In a separate test requiring listeners to transcribe speech without those choices, the synthesized speech had a median word-error rate of 43.75%, compared with 96.43% for the participant’s residual speech. The paper appeared online in Nature on June 12, 2025. 3
Choosing the sentence you heard from six options gives you information that an open transcription task does not. A headline saying the system was “94% accurate” would leave out a condition that changes how a reader understands the achievement.
The open-transcription result deserves attention on its own terms. It documents improvement while showing that listeners still encountered substantial errors. Both observations belong in an honest account of the work.
I think we can give people enough detail to appreciate that. A breakthrough does not have to be described as finished before it becomes worth celebrating. Reporting the conditions gives a reader something firmer to hold onto than an impressive percentage separated from its test.
The distance between a demonstration and everyday access
This research involved one participant, invasive implants, and specialized equipment. Speech intelligibility remained imperfect, and broader use requires further research. 3
The BrainGate2 trial is a device safety and feasibility study of an intracortical neural interface. Its purpose is to investigate ways to control computers and assistive devices through recorded brain activity. It is not a trial of a treatment to reverse ALS. 4
For anyone wondering whether an implanted speech interface is available through ordinary clinical care, the NIH’s October 2026 overview says these devices remain available only through clinical trials. 2
That gap between a research demonstration and something people can obtain matters. A family reading this story deserves to understand where the technology stands. So does a reader trying to decide what to believe about AI. The published results give us a reason to follow this work; they do not tell us when a particular person will be able to use it.
Gratitude can live with that uncertainty. We can appreciate an improvement in communication while remaining precise about who has used it and under what conditions.
Giving the people their share of the credit
The participant belongs near the center of this story. Reporting an AI result should also leave room for the person who participated in the research and the people who built and tested the system. The UC Davis announcement makes that human collaboration visible. 1
Calling this an AI breakthrough is reasonable, provided we explain AI’s role. Otherwise, the phrase makes the technology sound as though it completed the work by itself. In this case, understanding the contribution means paying attention to the participant, the implanted recording system, the software, and the clinical and engineering work around them.
I want that kind of attention to shape November’s stories. For each one, we should be able to ask what improved, how it was measured, and who experienced the benefit. Those questions help us appreciate useful work without asking anyone to accept a general promise about the future.
With this study, attempted speech became audible through an experimental system, and listeners could understand more of it under the conditions the researchers tested. A person’s ability to be heard is worth the effort of understanding those conditions. It is also worth being grateful for.
Sources
- UC Davis Health. “First-of-its-kind technology helps man with ALS ‘speak’ in real time.” June 11, 2025. Read the institutional report.
- NIH News in Health. “Restoring Communication: Technologies to Support Speech.” October 2026. Read the NIH overview.
- Wairagkar et al. “An instantaneous voice-synthesis neuroprosthesis.” Nature 644, 145–152 (2025). Published online June 12, 2025. DOI: 10.1038/s41586-025-09127-3. Read the study. Public author manuscript.
- UC Davis clinical trials. “BrainGate2: Feasibility Study of an Intracortical Neural Interface System for Persons With Tetraplegia.” NCT00912041. Read the trial description.