I believe pong has less entropy than the initial task of putting the cursor into the box. Pong is a 1D game with one degree of freedom, where you are effectively moving the cursor along the Y-axis. The previous task involved moving the cursor on the XY-plane, which involves two degrees of freedom.
Therefore, pong is less impressive than the previous task.
This is difficult to say because pong involves moving components while the box stays still. In order to play pong the monkey must first estimate where the ball will go and then give the command to move the cursor. This indicates that Neuralink may be able to handle complex multitasking, i.e. you may be able to walk or drive and the system will still be able to decode you correctly.
That's a good point, it is nuanced for sure. My line of reasoning is that ultimately the monkey only has to control one dimension of movement of some object, so whatever complex reasoning is required to arrive at the decision to move up or down (one dimensional movement), it would still be easier for a supervised learning algorithm to learn spatial-encoding neurons over a single dimension.
I think this gets deep into the difference between premeditation and action and what that means at a neural level. I would guess that Neuralink is not processing the "logical thinking" as much as it is on the neurons that dumbly encode an association with ping-pong racket location.
As long as some set of neurons uniquely fire approximately when the pong moves along its degree of freedom, different electrodes will pick up different voltages, and that can be learned by an algorithm. So we don't need to know how the monkey arrived at a decision, but rather just be able to index the possible outcomes of the decision. Which is why I thought that the entropy of the decision space mattered. It's easier to learn a smaller space.
I would tend to think this is a dumb pattern matching cluster (e.g. see visual encoding[0]) that's being picked up by Neuralink, and not the actual "logical decision making". Although, there may be more subtle neurons that fire in a sort of "premonition" which have learned the correlation of the racket position with the ball over time.
I can link to more papers on that -- it's not just visual encoding, but rats can "plan" where they are going to go, a sort of premonition. Humans also do this. It's like an echo that happens before you yell. That's what the brain does when you plan to go to the grocery store.
This is all really interesting stuff, but I'm not an expert!
If you watch closely, you can see the monkey is still "playing" pong with his right hand/paw. (He's pushing and pulling on the 'smoothie straw' in sync with the brain signals / on screen actions)
Therefore, pong is less impressive than the previous task.