5 ms·

This seems to be very similar to the early 2000s demos at Brown University/Cyberkinetics[1]. In 2008 they demonstrated it in humans[2] after lots of work getti

by dlg 5y ago

This seems to be very similar to the early 2000s demos at Brown University/Cyberkinetics[1]. In 2008 they demonstrated it in humans[2] after lots of work getting FDA approval. The core difference here (from my non-expert reading) seem to be electrodes are along a thread rather than on a chip. See https://www.youtube.com/watch?v=TJJPbpHoPWo https://www.youtube.com/watch?v=TJJPbpHoPWo for a video of the pong demo in humans. The motor cortex seems to have a clear, learnable signal though it is rather noisy sampling <1000 locations. Mary Lou Jepsen at Open Water has ideas about how to do this non-invasively [3]. [1] https://www.braingate.org/publications/ https://www.braingate.org/publications/ [2] https://www.brown.edu/Administration/News_Bureau/2006-07/06-002.html https://www.brown.edu/Administration/News_Bureau/2006-07/06-... [3] https://www.ted.com/talks/mary_lou_jepsen_how_we_can_use_light_to_see_deep_inside_our_bodies_and_brains https://www.ted.com/talks/mary_lou_jepsen_how_we_can_use_lig...

cyrux004 5y ago
Can somebody who is knowledgeable in this field of work explain how/if neuralink is better to the previous approaches and if its easier to "scale" this tech compared to previous works ?
- cblconfederate 5y ago
  The article describes how it has advanced: it's wireless and reads relatively more electrodes. Producing the device is something that can probably scale up easilt . 1024 electrodes is still nowhere near enough to decode precise movements imho. These pong experiments have been done in humans since 20 years ago, but real fluidity seems very difficult to achieve , because signals are noisy, the brain adapts, distractors are always present etc. The article doesnt mention how long it takes until the device needs recalibration. OTOH brain surgery is not something easy to scale and i doubt people will accept surgery for implanting 2 devices just to play pong. Such a device should be precise enough to allow using a robotic arm effectively. I think the applications of this are going to be in research for years, not as implants that paralyzed people can readily use.
- etrautmann 5y ago
  This is primarily an advance in the systems integration required to make such a sensor fully implantable abs capable of wireless transmission. It also ~10x’s the number of channels of the Utah array, which is the primary point of comparison as a product that can currently be used in human patients (in an investigational context as part of clinical trials). The Utah arrays do not have any active electronics however, and pass a bundle of wires through the skin, which has obvious disadvantages and scaling limitations. NL’s approach is scalable due to their robotic insertion system which can implant a (multi-channel) thread every few seconds. It should be possible to hit a few thousand channels within the window of a few hour surgery. They do face the same challenges with size, weight, and power that everyone else does, which forces trade offs on the bandwidth, ability to isolate spikes from individual neurons, and number of active channels. The primary limitation of this approach is that the needles cannot easily insert deeper than the outer layer of cortex (to my knowledge). This limits the application space to anything with recording or stimulation targets on the surface. Motor prostheses and gaming are perfect for this due to the anatomy, but many other medical applications require deeper targets, which their sensor cannot readily hit at the moment.
  - natosaichek 5y ago
    This is a great summary. I will just add that my understanding is that Neuralink is heavily focused on surgical speed, and a few seconds per thread is the slowest it will ever be. It will likely be much faster than that.
cblconfederate 5y ago
IIRC the major challenges was brain plasticity itself. Brain signals changed over time and the device needed new calibration.
londons_explore 5y ago
In this video, it looks like they are probably only sampling ~200 locations (assuming that's a 1080p screen and they have gone for one pixel per signal)
jonplackett 5y ago
I'm curious - does taking a signal straight from the brain reduce reaction times? Is there a big loss of time sending the signal down to the arm, or is the reaction time mostly cognitive anyway?
- mryall 5y ago
  Yes. Nerve signals travel at about 50-60 m/s through the limbs, so it takes about 20-30ms for a signal to travel from brain to the hand muscles. Source: https://en.m.wikipedia.org/wiki/Nerve_conduction_velocity https://en.m.wikipedia.org/wiki/Nerve_conduction_velocity
  - neuronic 5y ago
    Still better than Sonos Multiroom Speaker sync (70-75ms).
  - jonplackett 5y ago
    So Neuralink have a large secondary market of FPS players!
    - BlueTemplar 5y ago
      And about to be incarcerated cyberpunk guys.
    - tomcooks 5y ago
      Primary being the army who wants soldiers to quick-fire real guns

by dlg 5y ago

cyrux004 5y ago

Can somebody who is knowledgeable in this field of work explain how/if neuralink is better to the previous approaches and if its easier to "scale" this tech compared to previous works ?

cblconfederate 5y ago
The article describes how it has advanced: it's wireless and reads relatively more electrodes. Producing the device is something that can probably scale up easilt . 1024 electrodes is still nowhere near enough to decode precise movements imho. These pong experiments have been done in humans since 20 years ago, but real fluidity seems very difficult to achieve , because signals are noisy, the brain adapts, distractors are always present etc. The article doesnt mention how long it takes until the device needs recalibration. OTOH brain surgery is not something easy to scale and i doubt people will accept surgery for implanting 2 devices just to play pong. Such a device should be precise enough to allow using a robotic arm effectively. I think the applications of this are going to be in research for years, not as implants that paralyzed people can readily use.
etrautmann 5y ago
This is primarily an advance in the systems integration required to make such a sensor fully implantable abs capable of wireless transmission. It also ~10x’s the number of channels of the Utah array, which is the primary point of comparison as a product that can currently be used in human patients (in an investigational context as part of clinical trials). The Utah arrays do not have any active electronics however, and pass a bundle of wires through the skin, which has obvious disadvantages and scaling limitations. NL’s approach is scalable due to their robotic insertion system which can implant a (multi-channel) thread every few seconds. It should be possible to hit a few thousand channels within the window of a few hour surgery. They do face the same challenges with size, weight, and power that everyone else does, which forces trade offs on the bandwidth, ability to isolate spikes from individual neurons, and number of active channels. The primary limitation of this approach is that the needles cannot easily insert deeper than the outer layer of cortex (to my knowledge). This limits the application space to anything with recording or stimulation targets on the surface. Motor prostheses and gaming are perfect for this due to the anatomy, but many other medical applications require deeper targets, which their sensor cannot readily hit at the moment.
- natosaichek 5y ago
  This is a great summary. I will just add that my understanding is that Neuralink is heavily focused on surgical speed, and a few seconds per thread is the slowest it will ever be. It will likely be much faster than that.

cblconfederate 5y ago

IIRC the major challenges was brain plasticity itself. Brain signals changed over time and the device needed new calibration.

londons_explore 5y ago

In this video, it looks like they are probably only sampling ~200 locations (assuming that's a 1080p screen and they have gone for one pixel per signal)

jonplackett 5y ago

I'm curious - does taking a signal straight from the brain reduce reaction times? Is there a big loss of time sending the signal down to the arm, or is the reaction time mostly cognitive anyway?

mryall 5y ago
Yes. Nerve signals travel at about 50-60 m/s through the limbs, so it takes about 20-30ms for a signal to travel from brain to the hand muscles. Source: https://en.m.wikipedia.org/wiki/Nerve_conduction_velocity https://en.m.wikipedia.org/wiki/Nerve_conduction_velocity
- neuronic 5y ago
  Still better than Sonos Multiroom Speaker sync (70-75ms).
- jonplackett 5y ago
  So Neuralink have a large secondary market of FPS players!
  - BlueTemplar 5y ago
    And about to be incarcerated cyberpunk guys.
  - tomcooks 5y ago
    Primary being the army who wants soldiers to quick-fire real guns