What’s amazing, of course, is that Smith can steer a pointer with his brain well enough to text with his wife at home and answer our emails. Since he’d only been semi-famous for a few days, he told us, he didn’t want to opine too much on philosophical questions about the authenticity of his AI-assisted posts. “I don’t want to wade in over my head,” he said. “I leave it for experts to argue about that!”
The eye tracker Smith previously used to type required low light and worked only indoors. “I was basically Batman stuck in a dark room,” he explained in a video he posted to X. The implant lets him type in brighter spaces—even outdoors—and quite a bit faster.
The thin wires implanted in his brain listen to neurons. Because their signals are faint, they need to be amplified, filtered, and sampled to extract the most important features—which are sent from his brain to a MacBook via radio and then processed further to let him move the computer pointer.
With control over this pointer, Smith types using an app. But various AI technologies are helping him express himself more naturally and quickly. One is a service from a startup called ElevenLabs, which created a copy of his voice from some recordings he’d made when he was healthy. The “voice clone” can read his written words aloud in a way that sounds like him. (The service is already used by other ALS patients who don’t have implants.)
Researchers have been studying how ALS patients feel about the idea of such language assistants. In 2022, Klein interviewed 51 people with ALS and found a range of different opinions.
Some people are exacting, like a librarian who felt everything she communicated had to be her words. Others are easygoing—an entertainer felt it would be more important to keep up with a fast-moving conversation.
In the video Smith posted online, he said Neuralink engineers had started using language models including ChatGPT and Grok to serve up a selection of relevant replies to questions, as well as options for things he could say in conversations going on around him. One example that he outlined: “My friend asked me for ideas for his girlfriend who loves horses. I chose the option that told him in my voice to get her a bouquet of carrots. What a creative and funny idea.”
These aren’t really his thoughts, but they will do—since brain-clicking once in a menu of choices is much faster than typing out a complete answer, which can take minutes.
Smith told us he wants to take things a step further. He says he has an idea for a more “personal” large language model that “trains on my past writing and answers with my opinions and style.” He told MIT Technology Review that he’s looking for someone willing to create it for him: “If you know of anyone who wants to help me, let me know.”
