Neuroscientists Offer a Reality Check On Facebook's 'Typing By Brain' Project

the_newsbeagle writes: Yesterday, Facebook announced that it's working on a "typing by brain" project, promising a non-invasive technology that can decode signals from the brain's speech center and translate them directly to text (see the video beginning at 1:18:00). What's more, Facebook exec Regina Dugan said, the technology will achieve a typing rate of 100 words per minute. Here, a few neuroscientists are asked: Is such a thing remotely feasible? One neuroscientist points out that his team set the current speed record for brain-typing earlier this year: They enabled a paralyzed man to type 8 words per minute, and that was using an invasive brain implant that could get high-fidelity signals from neurons. To date, all non-invasive methods that read brain signals through the scalp and skull have performed much worse. Thomas Naselaris, an assistant professor at the Medical University of South Carolina, says, "Our understanding of the way the words and their phonological and semantic attributes are encoded in brain activity is actually pretty good currently, but much of this understanding has been enabled by fMRI, which is noninvasive but very slow and not at all portable," he said. "So I think that the bottleneck will be the [optical] imaging technology," which is what Facebook's gear will be using.

Re:

[The Real Dr John]

Considering how much people need to correct what speech recognition software does with your spoken words, this isn't going to work. The electrical signals that get through the skull are weak and noisy. Each person is going to have different brain patterns. But it makes for a great press release and gets people talking nonsense about it.

Re:

[michelcolman]

In other news, rocket scientists offer a reality check on the SpaceX "land rockets on barges" project. If NASA can't do it, nobody can so it's a waste of time and money to even try it.

What's the budget of the Medical University of South Carolina? How much money does FaceBook have? Yep, thought so.

Re:

[The Real Dr John]

Oh sure, rich people are going to save the world. Ayn Rand would love your world view. If a rich person can't do it, it can't be done, right? I'm sure they will put millions of dollars into it before they realize that it won't work. But why not? I saw it in a movie once! And next, Facebook will invent warp drive and teleportation. You know, because they are rich!

I've read some of the primary literature on doing this in people who are disabled. It is not going to be practical for regular use, and will never

Re:

[michelcolman]

Would you prefer them spending their money on private jets, big mansions and wild pool parties?

For crying out loud, if rich people want to spend their money to advance the state of the art, I applaud it. Musk has already achieved things that seemed impossible a decade ago. Landing and reusing orbital class boosters at prices even the Russians can't match, making electric cars that people actually want to buy, etcetera. And he's not the only one. I'm all for rich people spending their money that way instead

Re:

[The Real Dr John]

So what is Musk going to do with his rockets? Send rich people into orbit for a hefty price tag so he can make even more money? Wow, now I'm sold on rich people doing big things. How about maybe he spend his money to cure cancer, and give the treatments away once they are available? Oh, that wouldn't make any money. OK, so he can sell his cancer cure to very rich people who can afford it. Great. Now I'm really sold on rich people doing big things.

Look, rich people are rich because of a broken, oligarchy sys

Re:

[tsqr]

where do you go if you want to continue to grow your company and maintain your omnipotence?

Heh. If your omnipotence needs to be maintained, you're not really omnipotent.

Re:

[Ihlosi]

Is it just me, or isn't my typing this with my fingers "typing by brain"?

No, it's typing with muscular peripherals that are connected to your brain via UPPNCP (universal point-to-point neural connection protocol).

Just imagine what happens if either the peripheral is not working correctly or the UPPNCP connection is faulty or severed. Your brain may still want to type, but it can't.

Re:

[Ihlosi]

I'm just not sure I'd want to record that stream of consciousness for shiny things mmm hungry is my laundry done posterity.

You don't want to, but FB does. Who knows what little gems of data it can extract from that stream and then monetize ...

Re:

[Ihlosi]

Hungry for donuts? Need detergent to get donut grease out of your favorite T-shirt?

And you'll also need coffee, and the latest miracle drug to treat obesity. Hey, can we check your insulin levels while we're at it? You already gave us access to your brain, so looking into something as innocent as the level of a hormone shouldn't be an issue, right?

Re:

[dwillden]

Count seconds to 60. Use the traditional quasi accurate method of "one one thousand two one thousand....". You might come up a little short but that's 180 words a minute, and it's not all that fast.

Re:

[tsqr]

Now multiple that by about 20. That's how many characters you're actually going to generate with your 180 words per minute.

For the purpose of calculating typing speed, a "word" is 5 characters in length, including spaces and punctuation. 180 typed words per minute is 900 characters per minute, or 15 characters per second. That's superhuman typing speed, but not all that impressive for speech.

Re:

[TheRaven64]

Most people talk at around 150 words per minute, so presumably they're doing something approximating thinking at the same rate.

Re:

[avandesande]

with the things I hear people say I would guess it is actually about 1/3 of that ;)

Scientisis

[110010001000]

Those fuddy duddy scientists know nothing. Facebook will make an APP for it. In the Cloud. VR and IoT ready.

Ok, now think about it.

[Ihlosi]

IoT ready.

The T in IoT is either a) some thing very close to, or even in, your brain or b) your brain

Re:

[OffaMyLawn]

My CPU is a neural net processor. A learning computer.

Re:

[JoeDuncan]

IoT ready.

The T in IoT is ... your brain

I for one, welcome our new Internet of Brains (IoB) overlords!

Re:

[Rudisaurus]

Oh come on -- what fool modded PP as "Insightful"? It was meant to be FUNNY, people!

Re: Just a few more years . . .

[Lije Baley]

No no those cars have to be coming real soon now so I can leave the driving to them and focus on my hipsterisms.

I have EEG experience and my two cents

[TheOuterLinux]

From what I learned doing research, I also think 100 words per minute is extreme. You wouldn't believe the artifacts you get just from blinking or moving a leg or finger, and I had a 21 channel MITSAR and WinEEG to work with. The only way this is going to work is if Facebook has plans involving AI and quantum computing, a very dangerous combination for privacy. This is because an AI would have to get to know your brain waves on an incredibly intimate level, making encryption a joke when your brain is getting digitally fingerprinted. If what they say works, both the polygraph test and "truth serum" would be a joke. The only actual application for finding brainwave averages between people currently is to add to the "what are artifacts" knowledge and perhaps a quick and dirty diagnoses. Very rarely have I ever been able to use EEG to find correlations in research because everyone is too different and there is always some kind of confounding variable. Biofeedback projects in EEG sort of works for fun little things, but it's not the same thing as "mind reading" or 100 words per minute good. Now, I've played around with Tobii eye tracking, and that would be their best bet. In the the Linux and open source world we have a program called Dasher that may work with eye tracking to get their quota. Besides, doesn't Facebook have enough biometric data or are their government overlords hungry for more? This is why they'll never go bankrupt, as long as they have projects like this. Meanwhile, other more important research gets cut.

Re:

[xvan]

But what about correlations between different samples of the same person. With enough training this may be feasible.

Re: I have EEG experience and my two cents

[Anonymous Coward]

And instead if you'd just lied a little you could've sold to FB for 1000000000$.

Re:

[TheOuterLinux]

HBI (Human Brain Indices) Database is what is used: https://bio-medical.com/hbi-hu... [bio-medical.com]. WinEEG did have this as an add-on option, but the department wouldn't splurge for it, so I was stuck using experience versus whatever WinEEG was nice enough to include for cleanup, to which presents another problem. If you clean up artifacts, you have to use the same processes whether it is the same or another person. And no matter how many artifacts you can remove, time, temperature, clothing, hair style, health, etc. af

Re: I have EEG experience and my two cents

[TheOuterLinux]

It gets worse. We were running Window$ on iMacs. If I could at least try WINE on Linux with it, I would have, but they just bought the equipment and WinEEG needs a USB key to communicate live with the MITSAR. It would of been interesting to find out. I did look into other programs for Linux, but none are as easy to use and the compatible equipment is cheap and archaic in comparison. You can find a lot of biofeedback projects for Linux, but most of those are 2-5 channel and not 21. They do make Bluetooth EEG

Unfortunate

[Nidi62]

I would have had first post, but I used a prototype to type this :(

Tiger Direct

[gatkinso]

Used to sell a brain controlled mouse, IIRC.

Re:

[__aaclcg7560]

When I worked at the Google IT help desk in 2008, a user called in to have a tech take remote control of her desktop and move the mouse on her instructions. So she called a coworker over to demonstrate a voice-controlled mouse. When the user gave instructions to move the mouse, the tech moved the mouse for her. The coworker was so amazed by this until she found out it was a prank.

Not impossible

[grumbel]

100 words per minutes might be a stretch, but it doesn't sound all that impossible given that the speed record was set with hunt&peck typing by moving a cursor across the screen. Some fancy machine learning that could guess whole words at a time or something along the line should have no problem beating that by quite a margin. It wouldn't even need to be perfect, just close enough, to give a drastic speed up (i.e. like Tab-completion).

See this earlier work [youtube.com] that guessed video sequences from brain activit

Re:

[account_deleted]

Comment removed based on user account deletion

first thing I will type

[avandesande]

m y s p a c e

I'm using it right now

[ilsaloving]

I'm using it wiggle fingers right now foot itches squirrel is it lunch time yet and it works perfectly did I leave the oven on well I smell pizza i can't believe I did that 10 years ago.

There appears an obvious solution here.

[SuricouRaven]

Invasive brain implants for everyone!

Brainless

[volodymyrbiryuk]

They should make sure Facebook developers have a brain in the first place. Every piece software developed by FB is pure garbage. Is this some kind of lame attempt to get brainwaves data depending on what FB advertisement users are seing? Besides that BCI input devices have been around for years. He's trying so hard to be Elon Musk.