Since Elon is so publicly facing with his other ventures, it seems like Neuralink kinda gets pushed under the rug, but as time passes, it’ll be fun to see how the company grows and what societal awareness they raise. I’m most excited about Neuralink’s future than any other company. The prospect of a remedy for Alzheimer’s and/ or decreased limb functionality is super inspiring.
Put yourself in the circumstances of someone who has decreased ability to walk, limited ability to grab objects in daily life, or has severe paralysis. It’s no guarantee Neuralink will develop a fully functional remedy, but I personally believe the prospects are good. How amazing would it be to help one, two, ten, or even one hundred people be able to regain an ability they likely once took for granted?!
Based on the California Secretary of State’s business search, Neuralink was formally created around four years ago: in July 2016. LinkedIn shows 82 employees, so assuming this is a slight undercut, they have a team of close to 100 people- not very big, however at every public event, and nearly any time Elon mentions the company, he emphasizes that the company is hiring and looking for all-star talent, who are wanting to make a big impact on solving real problems. It wouldn’t surprise me if five years from now, Neuralink is as well known as SpaceX and Tesla are today.
Recently I’ve been digging through the information flow from mid-2016 to now. This doesn’t relate to the technology, but in an effort to go in somewhat chronological order, I’ll say it’s the first thing I thought was kind of interesting: Neuralink was originally trademarked by a different company. It looks like there are a couple of startup partners: a professor and medical partner at a hospital who are probably kicking themselves. An MIT Technology Review article says these guys didn’t know who they were selling the Neuralink trademark to, and after they agreed, they found out it was Elon Musk.
That happened at the very beginning of 2017. Then the real information blast hit. There was a pretty substantial time delay between when they first created the company to the first major release of information. That release came around a year later in April 2017, from Tim Urban of Wait But Why. It is a pretty lengthy article, and by pretty lengthy, I mean the PDF file is 196 pages. Despite being pretty detailed, the novel is really well done. Neuralink and The Brain’s Magical Future.
It has a bunch of pictures and helps bridge the gap between those who truly understand the details of the brain, versus people like me who are learning and benefit from a little more help via pictures and easy words. If you have a spare minute or 500, I highly encourage checking it out. Tim Urban and his articles on Wait But Why make complicated concepts less intimidating. He uses humor and pictures to make topics easier to understand. Tim’s article is thorough enough that it’s worth referencing in future posts, so I’ll plan on discussing sections of it pretty frequently.
Neuralink has one primary ambition: to merge man with machine. Elon Musk tweeted about this saying Neuralink’s mission statement is: “If you can’t beat em, join em.” In order to address the concern of artificially intelligent beings taking over, the team at Neuralink is developing technologies to connect human brains with computers.
Elon’s been concerned about the rapid progress of artificial intelligence for quite some time- in fact, he’s often stated it’s among the top issues we must address as a human race. In an interview conducted by Axios, Elon specifies the way artificial intelligence could destroy civilization is the same way us humans destroyed the habitats of primates. So, it’s not necessarily that artificial intelligent beings would choose to kill us, but instead, they could treat us in the same way we currently treat pigs. We tend to confine them to a designated habitat, but also tend to leave them alone without much unnecessary slaughtering… oink. *BACON*
Okay, replace pigs with turtles!
If the future unfolds a little differently, artificially intelligent beings could treat us like how we currently treat ants- if they’re not bothersome, most of us won’t go out of our way to kill them, but if they’re a nuisance, we’ll squish them and think nothing of it. This is the future we’re hoping to avoid. We don’t want some infinitely smarter being to treat us like we’re worthless. Staying at or near the top of the food chain is worth prioritizing!
By connecting our brains with computers, humans will have intelligence and functionality unlike ever before. We would essentially have the digital cloud in our brain. For example, one day I go to the kitchen, make breakfast, and sit down with grandma as she’s reading the newspaper. After a few bites, she asks what a “tur” is.
I also don’t know what a tur is, so from this point, almost any person with internet access knows what to do. The steps are actually incredibly long, so I’ll list them out:
1. Pick up phone
2. Open browser
3. Think about appropriate way to search
4. Type query with two thumbs into search bar
5. Find pertinent result
6. Retrieve information: “A tur is a wild goat native to the Caucasus mountains.”
7. Process result in brain after visual input
8. Synthesize information into a way grandma can understand
9. Vocalize the answer: “a tur is a goat that lives in mountains in southern Russia, near the country of Georgia.”
Contrast this list to a brain equipped with a mature brain-machine interface and an equally powerful supercomputer:
1. Think about question
2. Share result
Obviously the difference is pretty significant. This becomes more substantial with increased scale. If we’re able to have 24/7 access to any files we want and the cloud, I don’t see how it’s possible for anything to be smarter than us- hence the Neuralink mission statement: “if you can’t beat ’em, join ‘em.” However, this situation is pretty far down the road. Before we see a fully functional, brain-machine interface with few issues, there is an immense amount of work the team needs to complete.
Let’s move on to how this might work-
Our brain is our body’s primary tool to process information. That information is coded in the form of electric signals called action potentials. The magnitude, frequency, and location of these electric signals provides most, if not all the information we need to know what the brain is doing.
Although this is quite an oversimplification, I think this is a good, relatable example:
Let’s say Mike grew up in the Chicago suburbs and is planning on going downtown. When his buddy says to meet at the corner of Michigan Avenue & Ohio Street, Mike’s brain is gonna fire a series of action potentials of varying location, frequency, and magnitude. These electric signals will enable him to develop a generic layout of downtown Chicago and recall where the two roads intersect. If we could repeat this process countless times and determine which specific signals indicate what the brain is doing, we can replicate these signals for someone who lost their memory.
Neuralink is working on this by building out a handful of different technologies that must work well together. One major tool they are creating is an electrode thread. Electrodes are used to measure electric signals. The current thread iteration has 32 electrodes per thread. These threads are much smaller than the width of a human hair, so it’s not possible for a human surgeon to implant these threads into a patient’s brain. As such, the Neuralink team has developed a surgical robot that has micron scale precision.
Just so we can continue to appreciate how cool this is, it’s worthwhile to compare how much less invasive this in relation to comparable surgeries. In these surgeries, the process is pretty invasive, with the patient on the table and a pretty large hole in the skull. A surgeon would place electrodes by hand, with significantly less accuracy than a robot, with much greater likelihood of damaging blood vessels. The process would also take an order of magnitude or two longer.
This robot is custom-made by Neuralink, for Neuralink. This feat alone is pretty remarkable. Once the robot has placed threads with electrodes in the appropriate locations, the signals must be sent to a chip to be analyzed. One of the challenges with the recording of the signal information is that it’s difficult to distinguish between signals and noise. Noise is the undesirable information. We can imagine this in the form of a professional singer in a recording studio. If the microphone were able to only pick up the desired voice, and no background noise, that would be a pretty remarkable microphone.
In the case of the electric signals, the noise is difficult to get rid of; however, one method of differentiating the two is to amplify the signal, find the spikes, and use machine learning to figure out what noise typically looks like. Going back to the singer-microphone example, it’d be like recording a song then looking at the audio track to see what is wanted sound versus unwanted noise. It is much easier if you first increase the volume, then remove the noise. If you repeat this thousands of times, you can become really good at distinguishing the difference between the two. This is no different from a computer. When programmed correctly, the computer can use tons of data about tens of thousands of neurons firing and it can learn to separate between useful signal and unwanted noise.
After the noise is filtered out, a chip processes the signal spike information and converts it from analog to digital (0’s and 1’s). This chip is also custom made by Neuralink. It’s only 4 mm x 5 mm!