Like, you can't just tape a metal coat hanger to an iPhone and start getting text messages through it.
But somehow you can put some wires just as few mm into the surface of a brain and extract fine motor signals??
There is nothing stopping you from using a cloth hangar. You just need to make sure it is impedance matched and properly sanitized for subdermal use. The signal will be much worse as a cloth hangar had gauge/diameter that is several magnitudes bigger than that of a neuron. Nothing here is particularly "novel" that it cannot be trivially replicated if you have the budget of one of the FAANGs. If you have 100million in capital, you can easily find poor neuroscience and EE PhDs who will be willing to work for you for <80K a year just for the opportunity to build something like this. The reason it is difficult and out of reach for smaller companies is because small amplifiers and electronic packages that can be subdermally implanted requires significant manufacturing capital as they fall under semiconductors and biomedical devices. Once the industry settle on a design expect the price to drop dramatically as the core chips and parts become commoditized.
Here's a graph that summarize the temporal/spatial tradeoff of the different BMI tech: https://www.researchgate.net/figure/The-various-neural-recor...
*a shortcut by
You can't just read fine motor signals from any neuron, but given enough neurons, since they're deeply connected, you can infer a close approximation of the intended motor signals given a known output. That's why they need a lot of sampling wires in the brain; to makes sure they have enough data to make that inference. Luckily, machine-learning is very good at figuring out patterns in this sort of data.
My guess is that they train an ML model using the inputs and outputs when the joystick is connected. Then they can use the trained model to just generate the output directly from the inputs when the joystick is disconnected.
This will work fine for one particular monkeybrain/interface combo; however, you cannot use the same model on other monkeys. You have to train a new one each time.
You could if we didn't encrypt connections.
On that note, does anyone know how to upgrade libssl across my entire nervous system?
‘Inception’ shows someone who tries to do something like that, but there seems to be a vulnerability.
The game examples were super fascinating, but since I was already essentially familiar with the fundamental principles and technology, it didn't quite impact me as much as the researcher pairing his iPhone with a Bluetooth transceiver in the monkey's brain. That sent weird shivers down my spine.
They are also recording from up to 1000 channels, which is probably overkill for mind pong, tbh. But you'll need that many implanted electrodes to study long term electrode biocompatibility.
Not exactly “Hello, World” but maybe “ToDo List App”.
Granted it moves slower than actual Atari Pong, and takes a few minutes to get the hang of controlling the paddle. But IME it isn't much harder to control the up/down of the paddle with the EEG than it is to play a game like flappy bird, and that's using a couple random scalp electrodes from a consumer device.
So yeah thanks for bringing this up, because I feel some of the comments here are acting like this is much closer to "mind reading" than it actually is. Not that it isn't cool, but the overhyping kinda kills it for me lol. Is this how robotics people feel when Boston Dynamics releases a new video?
It's not totally trivial, but a reasonably skilled person should be able to get a not-too-janky version working with a bit of effort. (Also, where are you that undergrads get to interact with monkeys?!)
For example, a very basic approach is design experimental tasks that have contrasting conditions, or events to be predicted. They don't do that. Even something as simple as pressing the controller left vs right vs not moving at all in donkeykong would have been more interesting.
In any case, brains and traditional computer circuits work on different principles as I noted in my OP. Furthermore, there is quite a lot more redundancy in brains--and partial-redundancy is a bit of how it works; for example, if you have a population of neurons with different tuning curves, then the output in response to a stimulus can be integrated, and a likelihood distribution obtained for the actual value of a stimulus; that's just fundamentally different than how, say, a 5-stage pipelined RISC cpu works.
You can get a 128 key out of most electronic devices in a reasonable amount of time: https://en.wikipedia.org/wiki/Electromagnetic_attack#Known_a...