Interesting - I wonder what the limiting factor is for speed? It sounds like there is a bit of a technique to typing with the system, but I wonder if there couldn't be a faster way to filter the signal coming off the EEG.
I think they should take that cellphone metaphor one step further. Modern cellphones allow you to type a sequence of numbers and then using a basic dictionary look-up to determine which word you intended to type. Since we're down to nine characters now instead of 26 that's a start. Place the cursor in the middle let the person choose one of four directions, or maybe put it in the lower left, let them choose between up and right and put the most popular letters in the lower left of the grid. Now we're at at most two or four moves to typing a character. Still only recognizing on and off on the EEG. If we, through practice, can train the computer to recognize two different EEG states, now the patients can think "up, up, right" we'd start really cooking. And ultimately if it could recognize ten states, then the brain-locked could probably type a character a second, which would not be bad at all. At that point you're fast enough to chat and no-one would know the difference.
I guess what I'm saying is maybe an open source project doing character classification using an off the shelf EEG system might not be a bad idea.
I’ve seen people do up to eight characters per minute
17.5 minutes per tweet. Yikes. Hopefully they can improve that with UI tricks -- prioritize the letters, choose a letter group and then a letter (much faster than choosing a letter if you have to go through options sequentially), predict based on what was previously typed, use a dictionary and allow autoexpansion, etc.
They completely missed the point of Twitter, but you have to give them credit for making history. Nobody has ever posted to Twitter using a brain before. This could change everything.
5 comments
[ 6.7 ms ] story [ 15.4 ms ] threadI guess what I'm saying is maybe an open source project doing character classification using an off the shelf EEG system might not be a bad idea.
17.5 minutes per tweet. Yikes. Hopefully they can improve that with UI tricks -- prioritize the letters, choose a letter group and then a letter (much faster than choosing a letter if you have to go through options sequentially), predict based on what was previously typed, use a dictionary and allow autoexpansion, etc.