These are cool, however they will always be more limited than BCIs hooked directly into the brain. A few bits/sec might be your throughput with the former.
I think EEG (external, that is) device will be really important in controlling things that have very low entropy. For example, clicking start and choosing an item from the list is something that could be achieved with these. But, for example, writing a response to someone would be infeasible.
At some point, if VR goes mainstream, then people will be accustomed to having to wear something dopey on their heads! This would be a perfect "in" for EEG technology and the kinds of simple examples I mentioned. Maybe a future where we "think" to choose apps to use isn't too far off.
I believe "thinking" to call people will be very trivial with EEG devices if your brain does a sufficiently different thing depending on who you think about -- which I have no reason to doubt.
I bought an OCZ NIA the last time this fad came around in 2008ish if I remember correctly. I threw it out like a year ago having used it for about 10 hours total.
Turns out there's a reason medical EEG's are done in a faraday cage room, because mostly what the NIA picked up was the electrical noise generated from the 8 AC-DC transformer plugpacks charging my laptop, phone etc.
Pretty much the basis of "magic" in the hyper-sci-fi RPG Numenera [0], set a billion years in a post-post-post-apocalyptic future, where "mages" are called "nanos" (you can guess why.)
One thing not mentioned in the article is that we can not currently make implants that work after a few years because of scarification. ECOG devices might work but those have worse properties than implants.
I am working on such a BCI for with my girlfriend who's a graduate student in cognitive neuroscience.
It's actually based on a very simple concept that most people here will have stumbled upon in algorithms 101: binary search.
The subject is put in an fMRI machine and is looking at a screen like this[0] where the middle word of the dictionary is presented.
The subject can then perform either "inner speech" (singing or reciting in your head) if the word s/he wants to spell is before the one currently displayed, or "mental imagery" (imagining drawing some figure in space) otherwise.
As you probably know the time complexity of binary search is O(log(dictionary_length)) instead of a classic letter speller which would be O(alphabet_length*word_length) so this method is much quicker in most cases.
From the description, it sounds like you are not taking the difference in frequency between words into account. I think you might be able to improve the average number of choices by splitting at the median position, such that the correct word is equally likely to be before or after it in the dictionary.
Edit: You could also combine that with a language model to condition the probability of words on the preceding context, similar to how a phone keyboard recommends completions. Maybe the gains aren't big when you're still limited to selection by dictionary order, but it might be worth trying out.
Thanks, that is an interesting hypothesis, I had not considered it. The number of choices you have to go trough though should grow with the inverse of the frequency of the word, that means you could have to go trough an inordinate amount of choices to get to an unlikely word...
luckily with such "stateless" paradigms I can just simulate it for every single word!
> Edit: You could also...
This is exactly the improvement we are currently working on!
I had a similar idea from several years ago, to combine some BCI with Dasher, which predictively adapts the choices it gives you based on the patterns of the language, to make up for the limited interface. I experimented with the OCZ Neural Impulse Actuator but could never reliably control it.
Word of warning: Hardware from Emotiv (e.g. Insight) is just complete garbage, and the company clearly does not care much about anything beyond moving as many devices as possible. The hardware is crap, the app is crap, and for example the electrodes in the hardware have like max a couple months' worth of life in them. You can't even buy replacements for all the electrodes you need. Just complete garbage.
Hey, I apologize but I don't have any recommendations unfortunately. Honestly the $400 I traded for complete trash just put me off the whole industry. I'll try again in another ten years or something. I wish you good luck in your search, though.
Pay walled, but I don't even need to read the article. If you have a working proposal for consuming information faster than my eyes and ears can do it and also a working proposal for controlling a computer faster than I can do it with a keyboard; with the constraint that the method that does this doesn't kill me immediately or eventually, then you have my attention. Until that day happens, I do not care and I will focus my time and attention on things that are actually possible.
17 comments
[ 3.2 ms ] story [ 50.9 ms ] threadThis tech is exciting. Imagine a wizard game where you literally meditate to recover mana.
I think EEG (external, that is) device will be really important in controlling things that have very low entropy. For example, clicking start and choosing an item from the list is something that could be achieved with these. But, for example, writing a response to someone would be infeasible.
At some point, if VR goes mainstream, then people will be accustomed to having to wear something dopey on their heads! This would be a perfect "in" for EEG technology and the kinds of simple examples I mentioned. Maybe a future where we "think" to choose apps to use isn't too far off.
I believe "thinking" to call people will be very trivial with EEG devices if your brain does a sufficiently different thing depending on who you think about -- which I have no reason to doubt.
Turns out there's a reason medical EEG's are done in a faraday cage room, because mostly what the NIA picked up was the electrical noise generated from the 8 AC-DC transformer plugpacks charging my laptop, phone etc.
[0] http://numenera.com
It's actually based on a very simple concept that most people here will have stumbled upon in algorithms 101: binary search.
The subject is put in an fMRI machine and is looking at a screen like this[0] where the middle word of the dictionary is presented.
The subject can then perform either "inner speech" (singing or reciting in your head) if the word s/he wants to spell is before the one currently displayed, or "mental imagery" (imagining drawing some figure in space) otherwise.
As you probably know the time complexity of binary search is O(log(dictionary_length)) instead of a classic letter speller which would be O(alphabet_length*word_length) so this method is much quicker in most cases.
I would love some HN input on this!
This is a demo that you can play around with: https://sophia.science/biword/ [0]
And this is a complete explanation of the paradigm: https://sophia.science/biword/Biword_poster.pdf
Edit: You could also combine that with a language model to condition the probability of words on the preceding context, similar to how a phone keyboard recommends completions. Maybe the gains aren't big when you're still limited to selection by dictionary order, but it might be worth trying out.
Thanks, that is an interesting hypothesis, I had not considered it. The number of choices you have to go trough though should grow with the inverse of the frequency of the word, that means you could have to go trough an inordinate amount of choices to get to an unlikely word... luckily with such "stateless" paradigms I can just simulate it for every single word!
> Edit: You could also...
This is exactly the improvement we are currently working on!
Hope 2016 - Hacking the brain [0]
Semantic Maps showing models used to predict brain activity using "Voxels" [1]
0: https://www.youtube.com/watch?v=-1JPnHtLYCQ&index=75&list=PL...
1: http://gallantlab.org/huth2016/