With Neuralink, Elon Musk Promises Human-To-Human Telepathy
technologyreview.com
technologyreview.com
The author probably didn't read Tim Urban's WBW article[1]. Elon Musk never claimed that we were going to have telepathy n a few short years but rather that he was hoping that he would have some sort of BCI that would also be used by people who don't have some sort of disability.
"I think we are about 8 to 10 years away from this being usable by people with no disability … It is important to note that this depends heavily on regulatory approval timing and how well our devices work on people with disabilities."
"And Mark Zuckerberg gave a timeline of about 25 years. Mark Zuckerberg said: “I would be pretty disappointed if in 25 years we hadn’t made some progress towards thinking things to computers.”"
So I think the whole article has a clickbait-y theme, I would expect something more objective from the MIT Technology Review.
Finally as the author said Musk has indeed missed the deadlines on his achievements, but nonetheless also he manages to go through with his plans, both with SpaceX and with Tesla.
So even though I don't expect the neural lace in "a few short years" I think it will come eventually.
But all that aside, developing sensor technology that could get accurate measures of brain activity in a non-invasive and non-destructive way would be quite a breakthrough (and really the first step to anything more substantial. Until we invent nano-robots, implants will simply be way too dangerous).
We have no real idea of how the brain works. It's so vastly different from any machine we've ever designed, filled with redundancy and not only electrical signals, but chemical signals that combine to create all kinds of inputs and feedback. Trying to read our stream of consciousness is not an easy task, as it may not be represented in the brain in a way we could recognise with instruments (or even in one discrete location in the brain at all).
Brain research is always amazing, but it also shows just how far humanity is lacking in our own understanding of the world. Advertising/Propaganda, the Stanley Milgram Experiments, conformity experiments all suggest brains may be more deterministic than we care to admit. We have hundreds of thousands of inputs (skin, eyes, internal pain/nerve cells, hearing) and there's no way we could raise twins in a lab and control all those inputs to see if we can get two humans to always do the same thing (ethically).
I wonder if we ever do create machines/AI that have connotative abilities and that can understand self-awareness; if they may end up telling us that they (and in turn us) don't actually have the free well we think we do.
Except that those are not claims made by Musk and Zuckerberg.
Musk-related, the article says:
> He says that within eight to 10 years healthy people could be getting brain implants as new computer interfaces.
While in WBW's article Musk's quote is:
> I think we are about 8 to 10 years away from this being usable by people with no disability … It is important to note that this depends heavily on regulatory approval timing and how well our devices work on people with disabilities.
Notice the second part, which this article totally dismisses. That was his optimistic, "if all goes perfect" claim.
From the article:
> What makes it even remotely okay that Musk and Facebook are promising the public telepathy within a few short years?
...while Zuckerberg's quote is:
> I would be pretty disappointed if in 25 years we hadn’t made some progress towards thinking things to computers.
Minefield. I could point you to an extremely logical, well-founded, rigorous, and ancient argument that free will is nonexistent. Then I can turn around and point you to an equally logical, well-founded, rigorous, and ancient argument that it must exist.
I always say, who cares?
We don't need to know how brains work to have useful interactions with them, even pulling out pictures and words from neural activity.
That sensor breakthrough has already happened- and no dangerous intermediate step was necessary. opnwatr can make 8 million 100micron wide scans 120 times per second with what is basically a slightly modified version of cheap commodity/consumer hardware (using IR LEDs).[1] fMRI is slower and lower resolution and can still be used to extract images from brains in real-time.
In the past few weeks I have seen the number of articles, even on HN, about the brain skyrocket. This has been a trend for the past two or so years, but since the announcement of Neuralink it has exploded.
Braintech is becoming a topic that people are actually talking about. This will hopefully stimulate people to go out and work on related technology instead of letting it sit in obscurity as it has previously.
I suspect Musk agrees with you and that his figures have a 30-50 year margin of error. With Hyperloop, he caused a stir and moved on, and now someone else is working on it. I realize that's not a perfect comparison (he never indicated much intention to work on Hyperloop). He often claims that Tesla's main purpose is to get other car manufacturers to act.
Musk is good at using leverage, but with social capital instead of financial capital.
He realizes that our current societal structures are too prone to corruption and too slow to act before it is too late.
So instead of waiting for politicians to force the use of green energy he makes everyone want to drive an electric car.
If that fails, he makes sure that, at least some of, humanity can escape to another planet.
Now he is tackling the existential problem of AI.
I believe this was already touched upon in the waitbutwhy article, but it has obviously been clear for quite a while now and Musk has been rather open about it.
He proved himself with Tesla, even though there were extremely close moments. Now he is leveraging the Tony Stark image to make people interested in whatever ideas he throws out there.
I also predict this will be the year when we start figuring that out in a generative, systematic way.
But Fmri is extremely expensive and unaccesible, even to researchers. Just think of a few min. mri test costs hundreds/thousands - while for fmri neurofeedback training you probably needs some tens of hours, per person, and a minimum of tens of people, at least, for statistically meaningful research.
But recently, Mary Lou Jepsen has started working on a $100, personal, wearable(?) fmri device. This could be the equivalent, in a few years, to jumping from mainframes to the iPhone.
So if you're in Musk/Zuckerberg's place, if this explodes, it would be kind of nice to be in the field.
- Get Model 3 Tesla in production by mid-summer. (Musk said he'd sleep at the plant to make that happen.)
- Launch Falcon Heavy late this year.
- Launch Dragon spacecraft in crewed configuration, unmanned.
- Catch up on Space-X launch backlog and make some money. (Some customers have canceled because of delays.)
- Finish battery factory in Nevada and make some money.
- Get Brownsville launch facility ready for 2018 launch. (March drone video: [1] Not much but dirt piles.)
- Get self-driving car technology to work. (Vehicles are currently shipping with the hardware, but no software.)
That's just this year. For 2018, he's supposed to send people to the moon.
Not a good time for the CEO to be doing a new startup.
Part of his strategy is using his public personality and track record to lure more investments in. His is using financing as rocket fuel. Either there is a huge explosion or spectacular success.
What is admirable in him is that his risk taking is directed towards making new things that provide value even if his business totally collapses and he loses it all personally.
Let's be optimistic and call it a factor 9/2 ~ 4. Using the same factor for Neuralink, it will have its first medical device out in 15 years and its first general consumer device in 30 years.
The one big objection I have to this article is the emphasis on how "opening a person’s skull is not a trivial procedure". I would be willing to bet that Neuralink is shooting for something which won't require that, along the lines of injectable "neural dust" or stents:
Ahgg, leave this poor notation alone. What has it ever done to you?
Expanding to "of order 10" in spoken language.
If the objection is about context dependence, why is nobody going "Would you mind elaborating on the O(10) and what you mean by that? O(10) is the orthogonal group in ten dimensions".
The stent is placed in a blood vessel resting on top of the cortex, this will never be able to give you the resolution required for what we would consider a neural lace.
If Neuralink could somehow make "neural dust" that crosses the blood brain barrier and transmits very local recordings then they might be on to something.
The dream scenario would be tiny robots that cross the BBB and keep going until they latch onto a neuron.
There are 100 billion neurons in the brain. Even if you decide on a low-bandwidth limited system, you're still looking at somewhere between 1 and 10 billion neurons.
The bandwidth is fairly low, but you still need to work out a way to address the individual dust units for both reading and writing. You have to be able to place them fairly accurately. And you also have to make them maintenance-free, or replaceable if they break.
None of this is even remotely possible today. It may be possible fifty to a hundred years from now, but by then some other technology may have appeared.
I'd guess this approach is going to seem impossibly crude - like trying to build an i7 core out of valves.
We're just not very good at it. But placing a more sensitive EEG inside the head and then using stimuli mapping to reverse enough of the signal-information map to read intentional output seems a fine research path.
(And we already know you can get some output woth that method.)
And that article seems to me like a direct response to an article in which Tim dedicated thousands of words trying to explain how intrusive the current methods of scanning the brains are in order to support the claim that Musk is planning on coming up with something little less intrusive than opening a skull.
In fact, here's a direct quote from the Wait But Why article:
> This is a major topic at Neuralink. I think the word “non-invasive” or “non-invasively” came out of someone’s mouth like 42 times in my discussions with the team.
I have a strong feeling that the author of this article had a strong feeling about the subject before reading the Wait But Why article and dismissed some of the claims because of his previous opinion. Otherwise, I have a difficulty believing that this was his opinion after spending a considerable amount of time actually consuming the article and its claims.
For example:
* new way to insert implants into the brain without the risks of seizure, infection, stroke, swelling and eroding wires.
* New nonmetallic electrode array implants and wires that don't require removal before using MRI.
---
If I had few millions of extra money. I would spend it to figure out how the brain of Caenorhabditis elegans (roundworm) works in detail. The brain of roundworm has been mapped exactly (connectome is known) and we know it's 302 neurons and 8000 synapses well but we still can't understand how it brain works.
http://blogs.sciencemag.org/pipeline/archives/2015/10/30/sim...
Perhaps there's more involved than a connectome and a simple neural model.
There are non-spiking neurons in humans as well. They may have significant role in neuroplasticity.
Neuralink has none of that. However, I am very happy that someone is putting in the effort because the results so far have been real and impactful.
Of course it is true that Neuralink is a much larger leap from the current state of the art than mass production electric vehicles or reusable rocket stages. But is it a larger leap than a million people living on Mars? These ventures have many valuable milestones along the way, even if the ultimate goal proves unreachable.
Not to mention that sometimes you want to just take a break from your spouse's stream of grumbling, "telepathy device" would make that even harder.
To my mind the far more reasonable mechanism to interact with living neural tissue is given just a cursory 'in the future' remark in the mentioned essay, optogenetics. If you use light for the transmission of signals you get rid of many of the above issues. But, unlike Facebook's tech, you would actually need direct line-of-sight to the cells. The trouble with optogenetics currently is there is the same delivery problems associated with using CRISPR in a live organism. On the other hand, if you solve that delivery problem, you not only get all the medical benefits of being able to fix mutated genes, but you also get the (from first principles) useful access to optical I/O in the brain.
And the fact is that optogentics actually has some significant capabilities even today. It's much more of a bottom-up approach, but that is traditionally the direction Musk has used for his success:
http://web.stanford.edu/group/dlab/optogenetics/sequence_inf...
https://www.youtube.com/results?search_query=mouse+optogenet...
Rockets and electric cars were invented decades, if not a century before Musk came in and 'redid' the technologies from first principles. In this case he seems to have approached this like a rich man, not a principled engineer - rather than from first principles, more simply, "let just hire lots of smart people". And I don't think the result will be without utility - lots of good things have come from such research groups. But the result of this experiment will look nothing like the commercial success of Tesla or SpaceX. A best-case scenario is a bell-labs investment into a multi-decade long exploration of technology.