Hacker News Reader

Comment by LeifCarrotson | Hacker News Reader

I've done some testing with ECG sensors adhered to the scalp and know what noisy, limited signals that gives you, but man am I looking forward to the day when people can send data from their brain directly to a computer. I don't mind shaving my head (I'm male in a family with a history of male pattern baldness, if that matters) but I'd 100x prefer a device that lived outside the scalp to one with probes beneath the skull.

The amplifier fidelity and data processing rate of eg. TI ADS1299 chips that I worked with 10 years ago is only going up slowly, but (now outside the field) it feels like the data processing potential to turn extremely noisy and hard to measure waveforms into letters and words could move forwards extremely fast with AI.

View on HN

One thing that concerns me a bit is stacking machine learning onto this. Imagine a 100% paralyzed person in the hospital. We attach some probes to their head and get a very noisy signal. What comes back is a very noisy signal, but we've got a machine learning model that takes in a huge amount of context about interacting with computers and text and such and uses it as part of how it cleans up the signal, just like how phones today use machine learning to de-noise camera input.

Then let's say that our paralyzed person says something very impactful like "please turn off the life support" or "I will all of my money to my kids." Can you imagine the lawsuit over to what degree the machine learning was driving the boat?

Require them to spell it out letter by letter. Which btw is almost certainly how current Brain reading works, you think hard to stop a pointer walking a screen.

Or just ask a lot of questions which prove mindfulness.

How can you design questions that can't be answered by an LLM?

Ask them something specific that only they would know. Classic anti-cold reading technique.

Dumas has fun with this in "the Count of Monte Cristo"

A lot of them are viterbi estimators, you lose predictive power when you do character level only.

Most systems I have seen allow for you to delete characters. This was close to 20 years ago now though so perhaps it has gone through a worse is better phase.

Now, the issues you state are very real, terrifying, and depressing. Still anything that allows the person to communicate directly, rather than through an intermediary holding an eye board, will be a major step toward greater independence and the ability to hold those who abuse the disabled accountable.

in research, typically the behavioral tasks are randomized to avoid this sort of thing.

for practical purposes and legal decision-making, i have always envisioned a protocol where the patient is required to repeat back a random string in order to authenticate the system's output.

that is, the patient is shown or played a recording of the authentication sequence (by a separate device) and then they must express it through the bci before any device output from that session can be used as communication of intent for legal purposes.

I loved what ctrl-labs was doing, where they could read what your fingers wanted to do from an arm sensor. Not only was it supposed to be a great virtual keyboard, but it allowed some users to sprout a third virtual arm.

Unfortunately they were bought by facebook and then... nothing. and if they do come out with something, you know it will be locked into their platform.

at meta connect they showed someone playing a mobile game (maybe temple run or subway surfers) with their mind

Hot take:

The metaverse will take off when bi-directional BCI takes off.

I also think the smart players are trying to stake claims right now with in the software landscape, and keep holding on until HW improves, at least I hope no one actually thinks the current head mounted display system is the way forward! I'd look at HMD VR as the equivalent of black and white resistive screen Palm Pilots that communicate with IR. The foundational ideas are there, but smartphones were the revolution.

Totally agree that we are decades away from either.

Based on the news coming out about them lately, Neuralink seems like a lost opportunity to accelerate progress in this space.

A ton of money was dropped into improving HMDs, and a lot of improvements to surrounding technical fields got made, it is sort of unfortunate that physics fundamentally limits what can be done there.

If a similar investment had been made in BCI over 10 years, I suspect we'd be halfway to success.

However it is important that there are multiple companies dumping money into R&D, so that the best methodology wins. Even in HMDs, different devices learned from each other and everyone benefitted.

Of course implantable BCIs have more ethics concerns, and that likely is why the big tech companies didn't even try.

But the first company to get consumer level bi-directional BCI onto the market is going to become the largest tech company ever, orders of magnitude larger than anything we see today.

Based on your job, I'm guessing you are also sad that we've reached a local maxima regarding input devices! If machines were designed from scratch now days, meaning everyone had no prior expectations of keyboards/mice/touch/etc, I suspect we'd do everything very differently.

Neuralink wasn't a lost opportunity because it wasn't a possible opportunity in the first place. We are so far off from BCI input to the brain it's actually laughable. I'm willing to bet well over 60-80 years away

Output is also a joke but a far easier problem to solve.

> We are so far off from BCI input to the brain it's actually laughable. I'm willing to bet well over 60-80 years away

Progress depends on discoveries in surrounding fields having already been made, and on resources invested. To an extent, you can help spur surrounding fields with lots of $, but it seemingly isn't possible to force brilliant theoretical breakthroughs.

We have direct input for audio (cochlear implants) that needs iterating on, what needs tons of investment is visual.

Figuring out how do to research in a humane fashion, that'll be the tricky part, and it isn't something corporations have a good history of.

That's going to happen only when you have a million nanowires worming themselves throughout your brain. Of course you can't insert that many, so it will need to be a device capable of self-assembly.

You've just predicted a sci-fi dystopia featuring actual brain worms.

Thanks, I hate it.

Don't worry, nobody would want the massive extra complexity to make them self-sustaining. They'll all lead back to the same implanted surface and if anything starts to go wrong you can unplug it. And it would be the opposite of contagious.

You need a feed into the visual cortex and then some form of input.

Full immersion? Yeah, centuries away maybe, but that isn't needed for a revolution. Scientists never perfected smell o vision, but cinema still changed the world!

I think this as well - putting the proverbial 'money where my mouth is' by starting a company around it [0]. The signals one can get from a 'BCI' are quite magical. BCI is in quotes, because I, like CTRL-Labs, base the term loosely around non-obvious control to bystanders around you (no vocalization, no obvious hand movements, etc).

Extending the idea here: but much of the world is context based. Kitchens house a few main functions across cultures - where I would bet money that on a 'invocation count' metric for GoogleHome/Siri, the highest ranked function is the simple timer. If BCI companies finally got this around their heads (yes), then the skies clear. That's what we think, and what we prove with some pretty insane results. The common refrain from experts in the field is 'well your metric is only one bit' - and to that I say, 'well, yes, but it's the key to acting within the context.'

There's a lot of ink I could spill about this; but I'm excited by everyone getting it so wrong. It just proves with time we will get this right. The only sad part is I have no idea how it truly turns into the empire that is the personal computer distribution model through AppStores. But honestly, I don't care - the user stories are gobsmacking.

And to add to your hot take, here's a bit more: AR won't be glasses or contacts. It will be DynamicLands + Interactive Projection Mappings of the world. Wearing your compute will be seen like mainframes with mini wheels - oh cute, but I think you're missing the point.

[0] An old demo of the first embodiments of this idea: https://www.youtube.com/watch?v=VUbJl_xiDFU

I also tried working with EEG data from an OpenBCI device. My goal was to get an accurate enough signal to move a cursor on an X/Y axis and use that to control a predictive text input tool like Dasher.

https://en.m.wikipedia.org/wiki/Dasher_(software)

The X/Y BCI signal goal seems fairly practical and could open up critical pathways for interaction and communication.

If anyone is interested in pursuing this idea, please send me a message. My previous attempts were derailed due to technical difficulties with the BCI hardware/software link, but I'd like to see this idea come to light.

> but I'd 100x prefer a device that lived outside the scalp to one with probes beneath the skull.

So would everyone, but it's just too limited. It's going to be implants.

Would you mind giving me an idea of how I might go about procuring some sensors like this for myself? I’m at a loss for how to even search for one especially that gives signals I can parse myself rather than through some special software or proprietary machine. I’d love to be able to feed this stuff into code and mess around with it!

Try searching for OpenBCI, you can also look on AliExpress.

Reply on news.ycombinator.com