Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Reading minds with deep learning

blog.floydhub.com

Post reply on HN
Print view View original

11–20 of 27 posts

Re: Reading minds with deep learning

#11

Post by carapace »

janeybrad wrote: post #2

I think BCIs are super exciting. Humans have such high bandwidth in terms of input; tons of information is interpreted by our nervous system every second. Yet in terms of output, we are relatively poor; we can only type a few words a minute! This article suggests we could one day work 'like digital octopuses' but I'm naturally dubious, do you think it's possible as CTRL-labs propose that we could entirely rewire our…

> do you think it's possible as CTRL-labs propose that we could entirely rewire our nervous systems' output to let us operate computers as if we had multiple hands?

Yes. You would not quite do it that way though. Your motor cortex is capable of handling that sort of thing[1] but your conscious awareness isn't[2] so you have to develop a kind of expanded awareness, like a dancer or martial artist, but in an abstract space.[3] (I suppose you could just hallucinate having eight arms, now that I think about it.)

You don't have to wait, you can get results with crude hardware. The trick is to use hypnosis. See my other comment in this page[4].

We could have fantastic "magical" UI a minute ago, but people connect a brain to a computer and then focus on programming the computer. It's silly.

(BTW, our output channels are primarily the face and the whole body[5], but there's evidence that sign language predates spoken language. I've heard that pre-verbal babies can learn to communicate with gestures.)

[1] Higher dimensionality or more degrees of freedom

[2] Generally, for most people. However it is possible to train your mind to carry out two or more simultaneous activities, such as writing a sentence with one hand while solving a math equation with the other. But this is sort of a stunt. Most folks would never conceivably need to do something like that.

[3] Obligatory reference to "Cyberspace", and "Minority Report" UI.

[4] https://news.ycombinator.com/item?id=18564989

[5] Gait, posture, overall tension/ease, etc.

Re: Reading minds with deep learning

#12

Post by samlynnevans »

Cyphase wrote: post #10
samlynnevans wrote: post #3

Earlier quoted context omitted.

Great points. Thomas reardon (CTRL-labs founder) says the same in a talk you can see on YouTube (definitely check it out). I share your doubts, and wonder exactly how neuroplastic we are. Perhaps we are simply built to operate in a certain way, and solely tapping in to our nerves responsible for output won't necessarily allow us to output in hugely different ways. Even if possible, it may not be so simple,and instead…

Here's that talk you mentioned: https://www.youtube.com/watch?v=5Z5aZK2C3ew (2018 O'Reilly AI Keynote, Thomas Reardon, CEO, CTRL-labs)

Cheers!

Re: Reading minds with deep learning

#13

Post by samlynnevans »

cwkoss wrote: post #8
emilwallner wrote: post #4

While the results are really cool, they're not perfect, and seemingly even the best companies working with deep learning are still not getting it right every time. It makes me wonder if this technology will ever be useful for these tasks... I mean would you be willing to use BCIs even if they made small errors from time to time?

There is a lot of variance in error rate in biological human nerves as well. Certainly a lot of people on the low-end of the bell curve could benefit. Interesting to consider whether a 'filter' could be made for people with parkinson's that would drop tremor noise while preserving intentional actions. Perhaps with robotic exosleeves tremor could be offset/counteracted - or eventually we could input the filtered signa…

That would be a fantastic application, however it would differ slightly in that it would not be a BCI (as in a traditional brain computer interface), but a device within the brain that impeded certain signals being sent to motor neurons that it detected were Parkinsonian. This would involve directly manipulating the output of the brain to the body itself, and require some serious bio-mechanical hardware!

Re: Reading minds with deep learning

#14

Post by trevyn »

carapace wrote: post #9

If you connect a computer to a brain which element is the more sophisticated processing device? The brain. If you want high-fidelity cheap BCI don't bother with fancy sensors, galvanic skin response is enough. The trick is to use the high-powered processor to do the heavy lifting. You use hypnosis and create a simple feedback loop to calibrate e.g. one channel per finger and one per thumb. That gives you 8 data lines…

That’s cool! Have you actually done this? What clock rate can you achieve?

Re: Reading minds with deep learning

#15

Post by kakarot »

carapace wrote: post #9

If you connect a computer to a brain which element is the more sophisticated processing device? The brain. If you want high-fidelity cheap BCI don't bother with fancy sensors, galvanic skin response is enough. The trick is to use the high-powered processor to do the heavy lifting. You use hypnosis and create a simple feedback loop to calibrate e.g. one channel per finger and one per thumb. That gives you 8 data lines…

I'm interested in talking with you about this in length in a more private channel, do you have an email I can reach out to?

Re: Reading minds with deep learning

#16

Post by dmreedy »

carapace wrote: post #9

If you connect a computer to a brain which element is the more sophisticated processing device? The brain. If you want high-fidelity cheap BCI don't bother with fancy sensors, galvanic skin response is enough. The trick is to use the high-powered processor to do the heavy lifting. You use hypnosis and create a simple feedback loop to calibrate e.g. one channel per finger and one per thumb. That gives you 8 data lines…

Because you're an unknown, uncredentialed agent using language usually associated with pseudo-science, and asserting it as if it were plain as day. Unfortunately, most people don't have enough time to give every idea that comes across their mind's desk a full and rigorous examination. So we all use heuristics to filter which ideas are worth engaging with. And by and large ideas that start with the word 'hypnosis' are not.

You may well be on to something, but the bar of attention is high, and it's much higher for you than for someone using more normative ideas and terminology. Most people are not going to have the deep,intuitive connection to the idea that you have. They'll need some mechanism for attaining that intuition for themselves. Some numbers, a demo, a proof-of-concept. Something that helps them overcome their trained skepticism.

Re: Reading minds with deep learning

#17

Post by exo-pla-net »

emilwallner wrote: post #4

While the results are really cool, they're not perfect, and seemingly even the best companies working with deep learning are still not getting it right every time. It makes me wonder if this technology will ever be useful for these tasks... I mean would you be willing to use BCIs even if they made small errors from time to time?

See the progression of voice recognition. The answer is, yes, a BCI is still useful and will still be used, even if it makes errors. It will of course grow more useful and become more embedded into our lives, as it gets more accurate.

Re: Reading minds with deep learning

#18

Post by mrfusion »

janeybrad wrote: post #2

I think BCIs are super exciting. Humans have such high bandwidth in terms of input; tons of information is interpreted by our nervous system every second. Yet in terms of output, we are relatively poor; we can only type a few words a minute! This article suggests we could one day work 'like digital octopuses' but I'm naturally dubious, do you think it's possible as CTRL-labs propose that we could entirely rewire our…

I’m a bit of a contrarian on this issue. If you look at how you type you actually spend more time thinking about what to say rather than typing it out.

I think language is extremely high bandwidth. We’re bottling up extremely complex, abstract thoughts into a few symbols.

Re: Reading minds with deep learning

#19

Post by mrfusion »

carapace wrote: post #9

If you connect a computer to a brain which element is the more sophisticated processing device? The brain. If you want high-fidelity cheap BCI don't bother with fancy sensors, galvanic skin response is enough. The trick is to use the high-powered processor to do the heavy lifting. You use hypnosis and create a simple feedback loop to calibrate e.g. one channel per finger and one per thumb. That gives you 8 data lines…

I don’t see what function the hypnosis is providing?

Re: Reading minds with deep learning

#20

Post by taneq »

mrfusion wrote: post #18
janeybrad wrote: post #2

I think BCIs are super exciting. Humans have such high bandwidth in terms of input; tons of information is interpreted by our nervous system every second. Yet in terms of output, we are relatively poor; we can only type a few words a minute! This article suggests we could one day work 'like digital octopuses' but I'm naturally dubious, do you think it's possible as CTRL-labs propose that we could entirely rewire our…

I’m a bit of a contrarian on this issue. If you look at how you type you actually spend more time thinking about what to say rather than typing it out. I think language is extremely high bandwidth. We’re bottling up extremely complex, abstract thoughts into a few symbols.

Same. We're pretty good at adapting the information density of language to fit what we need to convey into the available bandwidth.
Post reply on HN