A thread for those who missed @neuralink live stream on their progress in building brain interfaces so far. While their progress is laudable, how they are approaching the problem is even more interesting
1/
1/
They have a device that you can implant in a day, and can read and write to neurons in the top few layers of the cortex
2/
2/
They demonstrated reading in pigs. The device "read" what the neurons connected to the snout were doing.
3/
3/
The "device" is 1024 threads, and a little watch battery sized thing that fits snugly in your skull. For which you have to take out a little skull for, but it's worth it assures @elonmusk
4/
4/
Now they are going to do a human clinical trial with tetraplegics to try to restore limb function
5/
5/
They have been granted breakthrough device status by the FDA, which doesn't mean much (yet) except that the FDA are helping them with the trial setup and eventual approval process. fda.gov/medical-device…
6/
6/
This was a recruiting event, so there were technical conversations about how the threads were made and the surgical robot used to put it in. Installing is not fully automated, but will be. Lots of conversation about the on-chip signal processing
7/
7/
"Spikes" in neural activity, the team clarified, were basically action potentials traveling through the axons to excite other neurons. But they can't take all the data, it's too much, I say! So they have to sample and process that, which they do on the chip
8/
8/
Oh Bluetooth Low Energy (BLE) is used for external communication. The 1024 channels have simultaneous read and write capability. I don't think I appreciate what this means fully yet
9/
9/
Write happens by sending electric current through the threads that stimulate a bunch of neurons together. From the video they showed, it looked like about 100k neurons per thread
10/
10/
A key limitation is the brain region they are working with right now. The threads go about 4 mm into the cortex. This limits access to inner regions of the brain, e.g. the hypothalamus to treat things like anxiety and addiction
11/
11/
The team explicitly said that there is nothing that fundamentally limits how deep we can go into the brain and eventually work a higher neuron resolution, so we will be able to do fine-grained things with neurons
12/
12/
The major concern will be to make sure there's no damage to brain tissue. There is also a limit to how thin the threads can be if they have to carry all that electric current to stimulate those neurons!
13/
13/
Elon Musk said he eventually wants to bring this down to cost around the price of LASIK. Which here in the UK is about £4,000. Obviously the first few devices will be expensive
14/
14/
They had a pig that had the device and then had it removed, which Musk explicitly said demonstrates reversibility. Obviously this is important from a safety perspective, but this makes upgrades easy too. This was explicitly called out in the Q&A
15/
15/
They're shooting for multiple orders of magnitude in improvements between versions. That's 100x better. They're at v0.9, an aspirational v1. v2 will be 100x better and v3 will be 100x better than v2. Probably by 2022
16/
16/
So, brain interfaces are coming faster than we think! They're really far from broad consumer availability, but Neuralink are shooting for a platform and once that is safe and the price comes down, we'll start wading from medical device territory to consumer devices
/END
/END
