I'm Innovative & Old!
Well, you may have seen it online already, but there’s my big dumb face getting honored by Forbes. (Stick around; there’s more below.)
Now back to my regularly scheduled (and thoroughly self-indulgent) exploration of recent research, in this case: Translating neural decoding into real-world communication.
Let’s take a look at recent breakthroughs in invasive and noninvasive BCIs restoring speech, typing, and digital autonomy for individuals with paralysis…and then back to me :)
Research Roundup
Imagi-typing
What if you couldn’t move…but you could still imagine moving? Well then, you can type, and pretty damn well.
A new study introduces “intracortical brain–computer interface (iBCI) typing neuroprosthesis” [1] that decodes attempted bimanual finger movements on a standard QWERTY keyboard layout. You imagine typing with both hands and it types.
The iBCI achieved typing speeds of 110 characters per minute (22 wpm) at a 1.6% word error rate in two participants with tetraplegia, “with as few as 30 calibration sentences”.
Part of what I liked in this paper was the incorporation of “a 5-gram language model”, an AI that helps predict the next word being imagi-typed [2] based on 5 word patterns. Very cyborg!
[1] If the authors could just squeeze in 3 more 7+ letter words that name would really roll off the tongue!
[2] Hey, I could have had a career in branding!
Brain2Qwerty
𝐁𝐫𝐚𝐢𝐧𝟐𝐐𝐰𝐞𝐫𝐭𝐲. What if you couldn’t move but could still imagine moving…and you weren’t keen on brain surgery just to type? Meat Brain2Qwerty. [1]
The authors of a new paper describe a deep-learning architecture that narrows the performance gap between invasive and noninvasive BCIs. Their system decodes typed sentences directly from MEG or EEG.
With MEG, the average character error rate was 29%. EEG wasn’t nearly as good, with a character error rate of 65%. That’s a bit of a problem given that MEGs are much less portable than EEG rigs.
The iBCI paper used a 5-gram AI decoder and others have used a full LLM. These guys should do the same.
[1] But where’s my Brain2Dvorak?
Imagine an Email
What if you couldn’t move, can still imagine moving…and have ambitions that go beyond QWERTY typing? For this week, I saved the most exciting neuroprosthetic breakthrough for last: a life already transformed.
In this study, “a man with paralysis and severe dysarthria due to” ALS, used an intracortical BCI for over 2 years. He logged “more than 3,800 h at home with no researchers present”, communicating “183,060 sentences—totaling 1,960,163 words—at an average rate of 56 words per minute.”
In clinical tests, his iBCI speech was decoded with >99% word accuracy from a 125,000 word vocabulary, all while he was also able to control a computer cursor BCI. He could send text messages and emails and browse the internet just using his iBCI.
Takeaway
Zoom out and these 3 studies map the same frontier: how much brain access do you need to turn imagined movement into language? Go all the way in—electrodes actually in the cortex—and you get startlingly good typing (22 wpm, under 2% errors) from just 30 calibration sentences. Stay outside the skull and you trade accuracy for safety: MEG gets close-ish, EEG really doesn't, and neither is anywhere near implant-level yet. That's the real axis this field is negotiating right now: invasiveness versus performance. This week's papers show both ends of it clearly.
But the ALS study is the one that truly excites me. Two years of unsupervised, everyday use—texting, emailing, browsing, nearly two million words typed at speech-like speed with near-perfect accuracy. That's the gap between "we decoded a signal" and "we gave someone their voice back". The typing speed records are impressive, but benchmarks aren’t lives. [1]
[1] A truth AI companies frequently lose sight of.
Media Mentions
You same it at the top of the newsletter–now indulge me again...
Forbes just named me to their Forbes Over50 list, in the Innovation category. This was a genuine surprise (there are some serious heavy hitters on that list!) and I’m sincerely delighted. My lesson:
Keep doing things others are scared to do and you never grow old. [1]
You only get 11 or so lives; make them count! [3]
[1] Wrinkles notwithstanding [2]
[2] What a weird word.
[3] https://www.smbc-comics.com/?id=2722
SciFi, Fantasy, & Me
𝗦𝗰𝗶𝗙𝗿𝗶𝗱𝗮𝘆: Green City Wars delivers more of Adrian Tchaikovsky’s lessons on speculative comparative cognition as a greentech Noir wrapper.
Stage & Screen
- September 8, Virtual: How might AI change the world of investing?
- September 15, SF: Innovation Day with INSEAD!
- September 16, DC: AI and education–beyond dreams and dread.
- September 19, Phoenix: I'm giving the keynote for the Association of Science & Technology Centers annual conference.
- September 21, Stanford: We're still working on the details, but hopefully I'll be talking about my research on machine learning and neurodiversity for Stanford's Neurodiversity Project.
- September 24, UC Berkeley: It's my annual Berkeley Change-makers Lecture!
- September 29, Cincinnati: Still baking...
- September 30, Irvine: Hybrid Intelligence for innovation!
- October 6, SF: UCSD Alumni Association
- October 6, SF: Giving a talk at the Draper Richards Kaplan Foundation
- October 6, Park City: It's Robot-Proof in the Rockies with setups.
- October 21-23, Warsaw: So much good stuff is in the works for my first visit to Poland
- October 27, Bohn: Maybe, maybe a visit to Germany!
- October, Toronto: The Future of Work...in the Future
- November 19, NYC: Secrets in the dark!