Comment by swordsmith | Hacker News Reader

Comment by swordsmith | Hacker News Reader

Parent Full thread

This whole "our existing i/o bandwidth is too limited" story started by Musk and co is not very well thought out.

Most people cannot think faster than they type or speak, at least coherently. The rapid information transfer with a computer via "thoughts" ignores the fact that a concept of a "car", for example, exists as different neural activity patterns, distributed over different parts of the brain for different people and changes with time. So that would imply the BCI would need to catalog how different all possible concepts are cataloged within an individual, to achieve such "direct thought communication", which would require whole-brain scale recording, something definitely NOT achievable with ephys (as a thought experiment, you'd have to introduce wires whose total volume become significant compare to brain volume).

Further, the act of speaking and typing is also a way to articulate and organize your (abstract) thoughts. If one skips that stage, would the output be meaningful at all?

The limiting communication bandwidth idea is a very fancy marketing message easy for people to hype up on, in typical Elon fashion.

>>Most people cannot think faster than they type or speak, at least coherently.

Likely going to sound quite pretentious - but I definitely can and have been able to do so for as long as I can remember. It's honestly quite annoying as I grow older. I like to think it may directly correlate with the fact I was able to type 160+ WPM on a shitty membrane keyboard at school in the 5th grade quite easily.

But yes, as you said and what I've come to believe... is that most can't.

> But yes, as you said and what I've come to believe... is that most can't.

I am also like you, in the sense that I can think orders of magnitude faster than I can type or speak.

It is bizarre to me that others wouldn't be able to. But now that I think about it, yeah, I think you two might be right.

(disclaimer: I have no knowledge on the subject and purely reason based on naive grounds)

I agree with you that language helps formalize and structure thoughts in an intelligible manner, and I do also have reserves on the hypothesis that one can think much faster than it speaks.

Though, is it completely wrong to assume that a neurolink interface would act like a language?

I mean, you mentioned that it would not be possible to probe the whole brain, which makes sense, but I could imagine that to communicate easily with a neurolink interface, you would somehow need to convert your thoughts to _something_ that is probed and intelligible to the interface, just like you do when you articulate sounds to speak.

For instance, if I just vaguely think of moving my hand, that's a pretty blurry and noisy thought. But when I focus, I can almost physically feel that I am at the very edge of moving it, my sensations in the hand increase, my muscles don't move but I feel like they are ready to at any time.

Surely, thoughts that are the result of a specific focus like this could be more easily probed. In fact, it could also be that your brain would need to "learn" how to properly focus thoughts that work with the interface.

That makes for a much less appealing interface though, because it's not "effortless" to articulate thoughts in a specific manner that an interface can probe.

You are touching on the subject of attention, which is actually extremely important and has a whole field of research (see for example, the hillarious invisible monkey experiment: http://www.theinvisiblegorilla.com/gorilla_experiment.html)

There are a lot of things that we ignore or don't register consciously. It is possible to decode items in one's working memory in controlled situations (see for example: https://pubmed.ncbi.nlm.nih.gov/28155212/).

So I suppose one possible approach would be to hold things in your working memory and have the Neuralink actively decode those. But in that case, as you've pointed out, requires a lot of attention and most likely would rely more heavily on contextual AI to narrow the decoder search space.

The basic premise of my skepticism with the "we need higher bandwidth I/O" statement still stays though -- the bottleneck of human communication is how quickly coherent thoughts can be formed in a way someone else can understand it (output), and how quickly information can be understood (input). For the latter example, you can scan an entire page in less than second. With immense speed reading practice, you can get a "gist" of a page in around that time. But understanding the details in that time is not possible.

Maybe you’re happy with thumbs. I type with both hands on a keyboard at something upwards of 120wpm and my hands are still the bottleneck.

Reply on news.ycombinator.com