Hard pass. There is not one good argument why Google or Facebook of all companies should be in charge of this technology.
I am all for advancing technology, but not by all means and not at all costs. Advertising companies (with a bad track record nonetheless) being the first to gain from this kind of technology is reckless at best, and dangerous at worst.
Just to point out (again) that you don't need fancy equipment to do this. Connecting electrodes to the brain is unnecessary. Some simple external sensors† and hypnosis and you're good to go.
The basic idea is to set up simple binary signals between your unconscious mind and your sensors. This is trivial.
Then you can take e.g. eight data plus one clock and gang them together to make a simple parallel port between your brain and the computer. Easy. (The obvious thing to do is use eight fingers plus a thumb.)
The brain is already a hugely sophisticated information processing device. The problem is not to get the raw signal out of it and then decipher it. That's just pushing sideways on the train. Just use simple hardware and let the brain do the pre-processing to get legible signals (with the subjective perception of direct mind-to-screen typing, say.)
†Galvanic skin response, or these days use little IMUs in gloves to detect twitches. The signals you're generating are not subtle. Another thing, the brain already has a high-bandwidth output device: the face. These days it should be trivial to train a NN to read expressions and "micro-tics" from e.g. some HD cameras trained on your face, and co-evolve a protocol between mind and machine.
- - - -
Edit to add, you guys have seen "Forbidden Planet" right? Take heed of the fate of the Krell, design your tools well, raise no "Monsters from the Id"...
>Machine learning algorithms eventually determined how to spot when participants were answering a question and which one of the 24 answers they were choosing
Seems like it might be a bit of an overstatement to say that Facebook is close to anything of the sort. Assuming that neural typing is as simple as running some real-time classification algorithm on brain states, to predict words correctly you'd have to be able to predict about tens of thousands of different possibilities. Not to mention the more difficult problem of individual idiosyncrasies in neural architecture and language embeddings for any given brain. Building a word classifier for a single brain is one thing, but I doubt if any two brains process language in the exact same way, so building some general-purpose classification model that works on all brains still seems intractable with current ML techniques. It's a nice start, but I think we're a long way from fully automatic neural typing.
Appreciate this a lot. Cousin is disabled and can’t speak enough for voice software nor use the computer but is a smart guy. Would open so many doors if this works.
I can already type with my mind. I had my brain internally wired to a pair of biomechanical "limbs" with graspers on the end that just happen to interface with a standard keyboard really well. I trained a neural net to run the setup, and since then it's served my needs very well. Granted, some of the inner workings are mechanical rather than electrical, but I can already type faster than I can think.
16 comments
[ 2.9 ms ] story [ 52.5 ms ] threadIt's funnier when they don't proofread.
I am all for advancing technology, but not by all means and not at all costs. Advertising companies (with a bad track record nonetheless) being the first to gain from this kind of technology is reckless at best, and dangerous at worst.
https://news.ycombinator.com/item?id=20566499
The basic idea is to set up simple binary signals between your unconscious mind and your sensors. This is trivial.
Then you can take e.g. eight data plus one clock and gang them together to make a simple parallel port between your brain and the computer. Easy. (The obvious thing to do is use eight fingers plus a thumb.)
The brain is already a hugely sophisticated information processing device. The problem is not to get the raw signal out of it and then decipher it. That's just pushing sideways on the train. Just use simple hardware and let the brain do the pre-processing to get legible signals (with the subjective perception of direct mind-to-screen typing, say.)
†Galvanic skin response, or these days use little IMUs in gloves to detect twitches. The signals you're generating are not subtle. Another thing, the brain already has a high-bandwidth output device: the face. These days it should be trivial to train a NN to read expressions and "micro-tics" from e.g. some HD cameras trained on your face, and co-evolve a protocol between mind and machine.
- - - -
Edit to add, you guys have seen "Forbidden Planet" right? Take heed of the fate of the Krell, design your tools well, raise no "Monsters from the Id"...
Seems like it might be a bit of an overstatement to say that Facebook is close to anything of the sort. Assuming that neural typing is as simple as running some real-time classification algorithm on brain states, to predict words correctly you'd have to be able to predict about tens of thousands of different possibilities. Not to mention the more difficult problem of individual idiosyncrasies in neural architecture and language embeddings for any given brain. Building a word classifier for a single brain is one thing, but I doubt if any two brains process language in the exact same way, so building some general-purpose classification model that works on all brains still seems intractable with current ML techniques. It's a nice start, but I think we're a long way from fully automatic neural typing.