Hacker News Reader

Comment by mebassett | Hacker News Reader

What is unclear (at least to me) is how much of this tech actually exists, and how much FB is simply stating their aspirations.

There is such optimism in tech that we take claims like being able to "type at 100 words per minute" without invasive surgery at face value. And there is such cynicism that we are (or at least I am) terrified of the sort of society this would create. But can anyone actively researching this field comment on how feasible this is with invasive surgery, never mind without?

View on HN

I have been working in BCI for 6 years and to me this is a total moonshot (the writing part). There is currently no affordable technology that can scan the brain at the necessary temporal and spatial resolution to even approach the precision of implants. The main problem is that bones and cerebrospinal fluid dilute the currents a lot. I suppose they want to make something like wearable IRM but that is impossible with our current tech. State of the art typing BCIs have speeds in letters per minute. What we need is a breakthrough in sensor technology, if Facebook can manage that then the field will be able to make a huge leap.

I guess they want to use NIR, similar to this-

http://www.opnwatr.io

If you read through the article they don't seem to be talking about BCIs at all, just a camera that reads subvocalisations. I'm surprised they can be detected visually, but once you do that it's functionally audio processing.

Not to underplay the significance if it works, it'd be an enormous achievement if they could get it working outside of lab conditions, but a BCI it is not.

In article they mention to use brain imaging to detect the speech patterns in the brain. I have limited knowledge of subvocalisation but would not that be detected on the throat? There were experiments with implants using phoneme detection and they worked but that gives you only vowels. I will post the article if I can find it.

I worked on similar technologies from 2007 to 2015. With tactile stimulations, using the regular alphabet, a user could read a 5 letters word under 15s after few hours of training. After 20h It was taking me <5s. The learning curve with Braille will be stiffer, I would be surprised if they can reach 100 words per second. They may start with basic common words, such as "ok" or "lol". What happens in the brain is quite interesting, you start to "see" what you feel and associate pattern with words. You don't have to recognize all the letters to get the word.

The technologie is called sensory substitution, I think it started in the 80s. Skin-reading is a funny terminology.

https://www.researchgate.net/profile/Guy_Boy/publication/266...

Is there a link to what you describe and some way of training the brain to produce synesthesia? That is an interesting trait I'd LOVE to experience. I've always envied the idea that one could look at numbers and just instantly and innately visualize some characteristic about them, or be able to see if music were "right" or "wrong" with some aspect of it.

It won't take long, it will probably happen within our lifetime. But I fully believe our grandchildren will take this technology for granted.

https://www.youtube.com/watch?v=jroQCyWwEgE

I just got done writing a post that also directly quotes a television series without direct acknowledgement. Actually you linked to the exact scene, so that's better than what I did.

Are we pulling down the benchmark for what qualifies as a worthwhile comment on Hacker News?

...Actually, yeah, probably.

100 words perminute seems aspirational to me. You as a consumer, could with ~100$ in off the shelf tech make a system that, given user practice, could do 30 though.

Not with subvocalization though; just with basic T9 style typing, just with your thoughts instead of fingers.

IMO best I can tell subvocalization is a holy grail of BCI and thus far even DARPA hasn't made much headway. I don't see FB doing better; especially given they have next to no expertise in hardware.

Well, there is a technology called FNIRS, which uses infrared light to detect blood flow in the brain kind of like fMRI, but without the giant expensive equipment. It's pretty new, and the signal is quite noisy. I'm skeptical of how good the quality of the interface could be when there is an unavoidable 2-4 second latency due to blood flow.

In the keynote it is asserted that the optical technology will not use blood flow but some other (more direct) measure of neural activity. It sounds to me like a near infrared form of intrinsic imaging. Very intriguing if it works.

Were there any details on what they use if not blood flow?

With a public announcement and a team of 60, seems like FB is at least convinced it's feasible.

I fear where we are headed with AR and BCI funded by corporations.

Check out the future portrayed by Keiichi Matsuda in HYPER-REALITY: https://vimeo.com/166807261

Reply on news.ycombinator.com