Imagine a computer interface capable of actualising the conceptual structures that you are imagining.
Imagine that this interface had a feedback mechanism simulating a virtual eye allowing you to perceive a full three dimensional rendering of that concept in real time.
Now think of an arraylist, think of how it responds to changes, visualise the corner cases.
Can we please not adopt this as a potential use case for this topic though? Real applications look more like this:
Imagine a movement. Follow an intense training protocol many times for each possible movement so that the signal processing component actually gets meaningful data it can differentiate and is certain enough that you want to trigger something. After multiple hours of training it might actually get good enough. Input dimensionality for that method would likely be simple action triggers (at least less than a traditional keyboard) instead of a complex concept abstraction.
Alternative: Stare at a board of visual stimuli, repeat for a few hours in order to train the system to your brain. Stare at board of visual stimuli again in order to actually use the thing. Again, slow paced and limited input dimensionality.
While these invasive models produce better signals than non-invasive BCI, they do not include a feedback mechanism and have far less throughput than using a keyboard. Which is great if you have a medical need for the technology and thus cannot use traditional interfaces, everything else is sci fi. I'm not trying to say that this isn't cool but we should really refrain from painting it as some miracle tech, otherwise we cannot really complain if media adopts that image and misrepresents the whole field again.
According to my 1990s graphics card packaging I have a dragon in my bedroom. I wouldn't expect a Musk company to produce underwhelming marketing claims but marketing departments don't change the state of the art in research.
Making this sci-fi levels of usable and getting it to consumers are both pretty unrealistic claims. I'm pretty sure you still couldn't get invasive research BCI past an ethics board around here if the patients skull isn't opened out of medical necessity anyway, doing so for fun isn't analogous to getting a piercing or tattoo, it's analogous to chopping off a limb to get a robotic arm.
the first electric car was invented in 1828, and the electric motor isn't that complicated of a device. Reusable rockets are a little more complicated, but fantasies about plugging the human brain into a computer are to those inventions what a sandcorn is to the sahara desert.
Musk has the habit of taking some established technology like solarpanels or pneumatic tubes and then repackage them in ways that get futurology type people all exited, but that doesn't explain how he's going to hook us into the matrix.
And yet, until Tesla came about we didn't have that many electric cars running around, nor did we have many companies investing heavily to change the status quo of gas cars. The same goes to reusable rockets. As many other people said in this thread, Musk is a PR nightmare and is a very peculiar public person when it comes to his tweets and such, but I wouldn't put it past him to actually find the talent and bankroll this until they achieve their objective.
There's no doubt that Musk has a knack for high risk entrepreneurship and he definitely very well how to repackage and promote things, but Neuralink wants to invent devices that come straight out of the science fiction realm.
Not only do we not have the AI required for natural language processing of the sort Elon is after, we also don't really have the neuroscience that explains how information is stored in the brain or let alone how we "write" to it. That's a different beast.
Sure, I didn't want to imply that they won't make money on this. Just trying to give a bit of perspective versus their marketing :) No matter what I think of the feasibility, a Musk-backed commercial player in the field will likely stir something up. With the pace of progress in the field over the last decades, that might just be a good thing.
> Sure, it's not like a Musk company ever broke the mold before.
Well to be fair they really haven't much, as far as technological advancements are concerned.
Reusable rockets weren't used before not because we didn't know how to do it, but because it doesn't seem to be economical. The judge is still out on whether SpaceX will actually be able to reuse the first stages in a way that's even remotely useful.
And there's even less innovation involved in Tesla cars. The only new thing they bring to the table is Autopilot, which arguably is a huge accomplishment, but not one that's "breaking the mold" in the hypercompetitive autonomous car space right now, and hasn't really delivered on its promises either.
What Neuralink proposes is so far ahead in terms of what we even glimpse to be technologically possible I have trouble thinking it's not basically his pet project that is guaranteed to receive press and funding just because the topic is sexy and it has Musk's name of it. If we are ever to get as far as NL implies, it will take more than a single consumer electronics company.
People who are not impressed by the innovations of Musk’s companies generally aren’t involved in those fields.
I had a NASA fellow (guidance, navigation, etc, so a relevant field) who literally told me landing on a barge is impossible, just months before they accomplished it.
I personally think this will take decades. And the implications of it for humanity if it works are probably greater than anything else Elon is working on.
Because Neuralink wasn't started in order to develop a medical device, it was started with an explicit goal to develop a BCI to ultimately improve the cognitive capabilities of healthy humans.
This has little to do with how code is written. It's so much more than "visualize an array". The gap from brain reading thread to what you described is like the gap between DeepMind and AGI
We're already close to replacing the eyes as a source of visual data.
We've already used brain waves to directly control input to a computer.
What I was talking about was essentially computer aided design with an integrated computer.
And I know how code is written, on many occasions I've known almost completely the structures that I'm going to spend the next two and a half hours translating and typing out.
The minutiae of the code that emerges as I'm typing is often an exception, but I generally have a fair idea of where and what the corner cases are going to be.
The biggest bottleneck by far comes in translating the internally visualised structure into an English derived programming language.
Imagine a computer interface capable of actualising the conceptual structures that you are imagining.
Imagine that this interface had a feedback mechanism simulating a virtual eye allowing you to perceive a full three dimensional rendering of that concept in real time.
Now think of an arraylist, think of how it responds to changes, visualise the corner cases.
You just wrote the code.