OTOH, if something does come out of this, it could have extremely far-reaching consequences. Forget interfacing with AI. Think - locked-in syndrome, or people with various disabilities (blindness, etc).
This is the kind of moonshot that ought to be encouraged.
We had tried external BCI for locked-in patients based on visually evoked potentials -- thought being that even if you lose the ability to move your eyes you can still steer an on-screen keyboard. EEG is so noisy, though, that it doesn't work unless the patient's looking at the target directly; ie, they can move their eyes. In which case, you should use an eye-tracker because it's $90 and 10 times faster.