Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Monkey MindPong

neuralink.com

Post reply on HN
Print view View original

221–230 of 291 posts

Re: Monkey MindPong

#221

Post by elil17 »

jonplackett wrote: post #168
elil17 wrote: post #150

For those who don’t know much about this field, this seems incredible. The reality is that this a tech demo that Neuralink is using to practice the basics. For other research groups in this field, this is tech from 20 years ago. The current state of the art is humans, mostly quadriplegic people, with chips which let them control robot arms with enough accuracy to pick up small delicate objects, play games, and type b…

I had no idea these things were possible. Are they currently insanely expensive / only experimental? How available are they?

Very much insanely expensive and only experimental. The man I met had a team of like five people needed just to set the thing up and calibrate it for his "brain waves" that day (they had to constantly fine tune the neural net he used to control it because the human brain is so plastic it would change every day and become unusable after a couple days) - plus more researchers running the experiments he was doing. The team had a portable supercomputer running the neural net (this thing was maybe a two or three server cabinets on wheels) that they brought around with them. They had a safety managers to make sure the arm didn't hurt anyone and make sure he was medically okay. This was not a device he used around the house, he was an employee of the research lab and using it for experiments was his full time job.

The risks of these devices are also significant: you have to get risky brain surgery to install them and then you might still reject the implants and have to get them removed.

A very limited number of people have these - the lab this guy worked at had I think about a dozen people, all with BMIs installed using different surgery techniques (some had the implants on the areas where the brain met the nerves going to the muscles so the output to the neural net was a more "direct" representation of movement whereas others had implants on motion planning areas on the brain giving a more "high level" output to the neural net). There are several such labs in the world. My best guess is there are less than one hundred BMIs installed on people today, as well as quite a few more on animals.

Re: Monkey MindPong

#222

Post by abledon »

When they 'decode' the info, are they using something akin to the Autoencoder tutorial found on the tensorflow site (ofc x1000 more complex)

Re: Monkey MindPong

#223

Post by elil17 »

jdsalaro wrote: post #174
elil17 wrote: post #150

For those who don’t know much about this field, this seems incredible. The reality is that this a tech demo that Neuralink is using to practice the basics. For other research groups in this field, this is tech from 20 years ago. The current state of the art is humans, mostly quadriplegic people, with chips which let them control robot arms with enough accuracy to pick up small delicate objects, play games, and type b…

I'm really interested in learning more about the process of putting existing research into practice as a process from an economics and innovation point of view. Does anyone have book, talks and other material you'd suggest looking at?

Some great places to look:

The Annual BCI Research Award highlights interesting advances each year

"Search for Paradise: A Patient's Account of the Artificial Vision Experiment" is an older book (2012) about a research subject's personal experience with an attempted vision-restoring implant

The research put out by the University of Pittsburgh Human Neural Prosthetics Program and Braingate at Brown University

Re: Monkey MindPong

#224

Post by elil17 »

icoder wrote: post #190
elil17 wrote: post #150

For those who don’t know much about this field, this seems incredible. The reality is that this a tech demo that Neuralink is using to practice the basics. For other research groups in this field, this is tech from 20 years ago. The current state of the art is humans, mostly quadriplegic people, with chips which let them control robot arms with enough accuracy to pick up small delicate objects, play games, and type b…

The chips you describe, are they responding to brain activity, or are they connected to something outside it, and if so, is that remnant muscle tissue or remnant (peripheral) neural tissue? And if so, do you agree that if that's missing, this is an unprecedented (possible) solution?

Both techniques have been used! Some are connected to the top of the spinal cord where as others are on different areas of the brain. Both techniques work and it's not clear which is superior right now.

Re: Monkey MindPong

#225

Post by antattack »

trombonechamp wrote: post #209
graderjs wrote: post #201

Earlier quoted context omitted.

I think that's a little unfair. I guess what they are doing is achieving the basics, but with a different interface. This is an important proof of concept. Remember, Musk is all about scalable manufacturing and products. If this was other's tech from 20 years ago, where's the massive commercialization? They failed. Musk is focused on this aspect, and he will (given his past achievements) succeed.

This is really basic. Here is a harder task from 2015 in monkeys with only 96 implanted electrodes (Utah array): https://youtu.be/lVFWIv5lupo?t=1673 Here is a video from 2014 of someone controlling a quadcopter with their mind with zero implanted electrodes: https://www.youtube.com/watch?v=lsraC04Mm8w

That's akin to saying that phones are basic since we already had desktops in the 1990s.

Having 10x more electrodes that stay active for a long time, attached to self powered unit under skin with low latency wireless comm link is a major step forward.

Re: Monkey MindPong

#226

Post by anonymouse008 »

anonytrary wrote: post #82
anonymouse008 wrote: post #32

Here is the ultimate question.... What smoothing algorithms did they use to guess the 'certainty' of the 'intended' movement in a Y-Plane game vs. the X,Y grid they learned from? Having researched in this space, you don't get better results from 'training' on a X,Y space and reducing that training to a Y-grid predictor. There is a ton of smoothing going on in the video... or the metal conductor plays a huge role in t…

> or the metal conductor plays a huge role in the electrical signals they get from the implanted electrodes I highly doubt that, you must be talking about non-invasive electrodes such as EEG. When the electrodes are inside the brain and thus in the cranium, they are effectively protected from outside EM activity since they're in an effective Faraday cage, so your signal has much higher fidelity.

> I highly doubt that, you must be talking about non-invasive electrodes such as EEG. When the electrodes are inside the brain and thus in the cranium, they are effectively protected from outside EM activity since they're in an effective Faraday cage, so your signal has much higher fidelity.

Yeah, I am speaking through non-invasive experience, however, unless your grounding is LIGO level sensitive, sticking a 3 ft metal pole into your mouth takes some significant processing to get a good ground, especially if it comes into contact with the earth and then you change your connectivity with opening and closing your mouth.

Again, low probability, but that just looks like it's asking for problems.

Re: Monkey MindPong

#227

Post by BlueTemplar »

jonplackett wrote: post #217
mryall wrote: post #206

Earlier quoted context omitted.

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

So Neuralink have a large secondary market of FPS players!

And about to be incarcerated cyberpunk guys.

Re: Monkey MindPong

#228

Post by tomcooks »

jonplackett wrote: post #217
mryall wrote: post #206

Earlier quoted context omitted.

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

So Neuralink have a large secondary market of FPS players!

Primary being the army who wants soldiers to quick-fire real guns

Re: Monkey MindPong

#229

Post by elil17 »

codeulike wrote: post #214
elil17 wrote: post #150

For those who don’t know much about this field, this seems incredible. The reality is that this a tech demo that Neuralink is using to practice the basics. For other research groups in this field, this is tech from 20 years ago. The current state of the art is humans, mostly quadriplegic people, with chips which let them control robot arms with enough accuracy to pick up small delicate objects, play games, and type b…

One of the neuralink innovations is having the electrodes installed very precisely via a robot surgeon, which supposedly allows them to make things easier/better/more precise - https://www.dezeen.com/2020/09/02/neuralink-elon-musk-brain-... Previous discussion of surgical robot https://news.ycombinator.com/item?id=24311152

That's really cool! I think one of the things that will hold them back is that there are still a lot of issues on even knowing where to place the electrodes - but as those issues are solved this will have such amazing potential!

Re: Monkey MindPong

#230

Post by subw00f »

cloudking wrote: post #111
alexslobodnik wrote: post #10

It looked like the dexterity of the mind was quicker than the dexterity of the hand.

No brain signal to hand latency

That's just replaced with brain signal to chip then "decoder" then game software, which I would think is greater than nerve conduction velocity[1]. But then again, there's no "latency" of hand movement + joystick signal -> game software, so maybe it evens out?

[1] https://en.wikipedia.org/wiki/Nerve_conduction_velocity

Post reply on HN