The problem with EEG-based BCIs is their lack in precision and speed. EEG waves average the activity of millions of nerve cells and thus tend to be slow and long, which is especially true for the SCPs, but also for p300-systems and mu-based approaches.

If you go down the ladder, i.e. look at smaller groups of neurons, this should improve, as you come closer to the organisational levels used by the brain itself for real-time control of movement. Although a lot is still out in neuroscience, it seems as if groups of several hundreds of neurons act as the principal "units" in movement control, mirroring each a particular movement or a particular direction of a complex movement in which they are the most active. With all other movements, "fire" activity is lower, which give the brain the possibility to go for a particular movement dpedning on how activity patterns are distributed throughout the motor cortex. Mathematicians call this a vector summation, but you may also think of it as "vote".

Given this and the fact that nerve cells enjoy a considerable degree of plasticity, i.e. they are able to "learn" or change, it should be possible to make predictions about intended movement with a reasonably small number of electrodes distributed throughout the brain. Once it is possible to make these predictions in real time, building a (quick) neuroprosthesis should also be manageable.