6 ms·

I wonder if after training, these paralyzed users would develop a sort of language of brain primitives (almost a "thought alphabet") which are easily recognizab

I wonder if after training, these paralyzed users would develop a sort of language of brain primitives (almost a "thought alphabet") which are easily recognizable by the software and easy signals for the human brain to create. It may vary by user, but I wonder what the general set of easy "letters" would look like.

nicklovescode 10y ago
I've been thinking along similar lines recently, but don't have fully formed enough thoughts to write. Would love to brainstorm if you're in the bay area at some point. (email in profile)
- got2surf 10y ago
  Will definitely let you know next time I'm around the Bay Area. I'm also interested in ML and education (my startup analyzes feedback comments, and we've been working with student surveys in education recently), so looking forward to chatting sometime
johnloeber 10y ago
I suspect this might be a candidate for PCA, or a similarly driven approach.
stinos 10y ago
From what I heard this is actually how it happens. Roughly speaking, upon first connection of the implant the subjects can't do anything at all basically. They have to train themselves and to some extent the software to work together, so that the different patterns in firing of the neurons get matched to different letters. (or movement, or whatever is attached at the output side of the software).
KON_Air 10y ago
But that's a stop gap measure at best, sure until the "personal calibrations" can be done programmatically. It would be a neat thing.
drzaiusapelord 10y ago
I doubt we have the signal fidelity for that. These are just electrodes that detect voltage changes. They can't see 'thoughtwaves' or somesuch. They see very basic information and with a little training you can set off a voltage threshold. These voltages are really aggregates and the tools used here just aren't granular enough to be very exacting. You can do this right now at home with the Neurosky headset or the Neurosky Jedi game. The former comes with a programmable API. The stuff in clinical labs isn't typically more complex than this. This team went with an implant which is going to provide a higher level of accuracy, but its not going to read thoughtforms directly. I imagine there's a more efficient way to handle typing for these low bandwidth cases. Maybe chording of common syllables. The layout in the article looks inefficient. You can probably lose accuracy in text to have easier 'speaking.' If you wanted to say 'father' you'd have to hunt and peck six letters. With chording you could click on 'fa' and 'der' and it should be understandable via context. Toss in some predictive logic and you can chord words at once or even entire sentences. Probably easier said than done, of course.
- hatsunearu 10y ago
  It's like having an SDR next to a digital computer pickup unintentional emissions to guess what the computer is doing, except the computer isn't designed by anyone and the design isn't understood at all yet. It's a miracle every time someone manages to put the signals together to get something meaningful done.
  - analogist 10y ago
    As a neuroengineer working in the field, this is quite accurate. Understanding the compute architecture goes a loooong way - after all, acoustic RSA key extraction (https://www.tau.ac.il/~tromer/acoustic/ https://www.tau.ac.il/~tromer/acoustic/) is possible. Whereas we're not exactly even sure how the brain is supposed to theoretically compute, other than that it's tremendously parallelized to a degree we don't quite fathom. The electronics explosion has primarily come out of computational motifs that rely on the lightspeed resolution of semiconductor gates and heavily rely on sequential processing, but the brain doesn't work this way AT ALL. An important concept here is the 100-steps-rule (https://www.teco.edu/~albrecht/neuro/html/node7.html https://www.teco.edu/~albrecht/neuro/html/node7.html) - neurons are SLOW! You can out-jog most non-myelinated neural signals, and the vast majority of sensory and motor computations finish in the order of 100 "clock cycles". Write me a computer vision algorithm that has enough parallelism to complete in 100 cycles, and we can talk about understanding the biological brain compute structure and true brain-computer interfaces.
pizza 10y ago
Interesting. Some stuff I've stumbled upon in the past which is kinda related to that idea: https://en.wikipedia.org/wiki/Language_of_thought_hypothesis https://en.wikipedia.org/wiki/Language_of_thought_hypothesis https://en.wikipedia.org/wiki/Private_language_argument https://en.wikipedia.org/wiki/Private_language_argument

nicklovescode 10y ago

I've been thinking along similar lines recently, but don't have fully formed enough thoughts to write. Would love to brainstorm if you're in the bay area at some point. (email in profile)

got2surf 10y ago
Will definitely let you know next time I'm around the Bay Area. I'm also interested in ML and education (my startup analyzes feedback comments, and we've been working with student surveys in education recently), so looking forward to chatting sometime

johnloeber 10y ago

I suspect this might be a candidate for PCA, or a similarly driven approach.

stinos 10y ago

From what I heard this is actually how it happens. Roughly speaking, upon first connection of the implant the subjects can't do anything at all basically. They have to train themselves and to some extent the software to work together, so that the different patterns in firing of the neurons get matched to different letters. (or movement, or whatever is attached at the output side of the software).

KON_Air 10y ago

But that's a stop gap measure at best, sure until the "personal calibrations" can be done programmatically. It would be a neat thing.

drzaiusapelord 10y ago

I doubt we have the signal fidelity for that. These are just electrodes that detect voltage changes. They can't see 'thoughtwaves' or somesuch. They see very basic information and with a little training you can set off a voltage threshold. These voltages are really aggregates and the tools used here just aren't granular enough to be very exacting. You can do this right now at home with the Neurosky headset or the Neurosky Jedi game. The former comes with a programmable API. The stuff in clinical labs isn't typically more complex than this. This team went with an implant which is going to provide a higher level of accuracy, but its not going to read thoughtforms directly. I imagine there's a more efficient way to handle typing for these low bandwidth cases. Maybe chording of common syllables. The layout in the article looks inefficient. You can probably lose accuracy in text to have easier 'speaking.' If you wanted to say 'father' you'd have to hunt and peck six letters. With chording you could click on 'fa' and 'der' and it should be understandable via context. Toss in some predictive logic and you can chord words at once or even entire sentences. Probably easier said than done, of course.

hatsunearu 10y ago
It's like having an SDR next to a digital computer pickup unintentional emissions to guess what the computer is doing, except the computer isn't designed by anyone and the design isn't understood at all yet. It's a miracle every time someone manages to put the signals together to get something meaningful done.
- analogist 10y ago
  As a neuroengineer working in the field, this is quite accurate. Understanding the compute architecture goes a loooong way - after all, acoustic RSA key extraction (https://www.tau.ac.il/~tromer/acoustic/ https://www.tau.ac.il/~tromer/acoustic/) is possible. Whereas we're not exactly even sure how the brain is supposed to theoretically compute, other than that it's tremendously parallelized to a degree we don't quite fathom. The electronics explosion has primarily come out of computational motifs that rely on the lightspeed resolution of semiconductor gates and heavily rely on sequential processing, but the brain doesn't work this way AT ALL. An important concept here is the 100-steps-rule (https://www.teco.edu/~albrecht/neuro/html/node7.html https://www.teco.edu/~albrecht/neuro/html/node7.html) - neurons are SLOW! You can out-jog most non-myelinated neural signals, and the vast majority of sensory and motor computations finish in the order of 100 "clock cycles". Write me a computer vision algorithm that has enough parallelism to complete in 100 cycles, and we can talk about understanding the biological brain compute structure and true brain-computer interfaces.

pizza 10y ago

Interesting. Some stuff I've stumbled upon in the past which is kinda related to that idea: https://en.wikipedia.org/wiki/Language_of_thought_hypothesis https://en.wikipedia.org/wiki/Language_of_thought_hypothesis https://en.wikipedia.org/wiki/Private_language_argument https://en.wikipedia.org/wiki/Private_language_argument