>When the electrodes are inside the brain and thus in the cranium, they are effectively protected from outside EM activity since they're in an effective Faraday cage, so your signal has much higher fidelity.
Pretty much this on a different and somehow related note I would still try to check if the algorithm does not just trick us into thinking that it is doing what the monkey wants it would be nice if they post some more in depth material of how the code or whatever they use looks at the end just to be sure we are not in front of a half machine clever hans (https://en.wikipedia.org/wiki/Clever_Hans)
TO clarify: we dont have the ability to know if the monkey made an error he wont say hey I didnt want to get that one right or win that pong movement, have your milkshake back please. And we kind of need to know that in fact when doing mathematical modeling and AI having a way to quantify that is a requirement and one of the first things they teach you about.
True, the post-processing algorithm could just be doing a randomish walk towards the answer, but I highly doubt that is what's happening. 2D cursor movement is well understood to be within the realm of possibility for EEG, so I imagine it is actually much easier with an in-vivo device. From what I understand, this is entirely supervised learning. The monkey is "telling" the device what neural pattern corresponds to moving the cursor. If I had to guess, then this chip is planted extremely close to the hippocampus, which is known to have cells that encode spatial information. The algorithm is probably doing a basic supervised learning on the cells that fire for a particular location of the cursor.
I'm just guessing on this, but the "training" session probably involved having the monkey stare at the screen for a while, while the cursor moved, which allowed the device to capture a spatial heatmap of which cells fired at which locations. There's probably some online optimization happening as the monkey then continues the "training" process by completing the task. Overall, this task is completely doable with the current technology, so I would not assume any foul play here. If the monkey was writing Shakespeare, I would definitely doubt it.
the chip is implanted in the motor cortex, I believe, so the training consisted of rewarding the monkey for moving the cursor to a target.
After the monkey has played the game a bunch, you can correlate the neuron firings generating motor signals to the arm with the motion of the cursor on the screen. Later, the monkey's arm need not actually move the joystick.
The hippocampus is pretty deep in the brain, and the Neuralink surgery can't go deeper than a centimeter or two.
How would it know which direction to move in if the position of the goal was not part of the input to whatever control algorithm they're using? The part of the code that detects the overlap of the cursor and the goal should be completely separate from the part that is translating the implant inputs into x,y movement so I don't see how it could be driven by anything other than input from the monkey.
Pretty much this on a different and somehow related note I would still try to check if the algorithm does not just trick us into thinking that it is doing what the monkey wants it would be nice if they post some more in depth material of how the code or whatever they use looks at the end just to be sure we are not in front of a half machine clever hans (https://en.wikipedia.org/wiki/Clever_Hans)
TO clarify: we dont have the ability to know if the monkey made an error he wont say hey I didnt want to get that one right or win that pong movement, have your milkshake back please. And we kind of need to know that in fact when doing mathematical modeling and AI having a way to quantify that is a requirement and one of the first things they teach you about.