I looked into the Neurosky and Emotiv headsets the other day for a side project. They've been on the market for a while, but most reviews are pretty cautious.
The Neurosky headset appears to suffer from inconsistent readings, and is mostly really good at picking up muscle movement (head turning, blinking) - great if that's what you're after, but terrible if not.
The Emotiv headset has some mechanical issues - the pads are a pain to wet and place and keep falling out. The data also appears to be super encrypted and quite hacker-unfriendly, so I stopped reading there.
I got the Emotiv headset a few years ago when the came out. It suffers the same shortcomings of needing neuromuscular responses. When I was training the model, it was just easier to get it to do things squeezing parts of my face.
That being said, I wonder if this shortcoming is more on the user. Their userbase are hobbyist and dabblers who have grown up with the explicit knowledge that you CAN'T move something with your mind (this is an entirely unsupported thought).
Instead of training just being 'think about moving the box', you could look into techniques like phantom hand therapy. Have the user go through a similar session where the application tries to elicit a response similar to the therapy as a way to get them to start manifesting a better response (maybe mix it with the visualization techniques athletes use) for training. While not optimal, have the user physcially do something X times, then make them visualize doing it.
Like I said, I think the problem is more in the user than the hardware (I've never had a problem w/ the Emotiv). If someone comes along with a better training method, then this would certainly help.
Lastly, I don't think neuromuscular training is entirely a bad thing. It lessens the 'ooo' factor of controlling something with your mind and alienates people who'd actually want to use this (quadriplegics), but as an able-bodied person, I see no reason not to be able to use them.
The stuff you mention about muscle/movement artifact for the NeuroSky device is not entirely accurate.
Check the April edition of the journal Clinical EEG and Neuroscience http://eeg.sagepub.com/ (out soon) it has a paper about the validity of the NeuroSky device.
Happy to talk about this more if you're interested.
I got to try it out last year, it's a novel application of the tech but it takes a lot of practice before you can really feel in control. It was more immediately interesting to just use the demo app that shows the exact parameters being measured. The Zen Bound 2 demo is very cool to see a pro use though.
The Neurosky headset appears to suffer from inconsistent readings, and is mostly really good at picking up muscle movement (head turning, blinking) - great if that's what you're after, but terrible if not.
http://www.colorfulwolf.com/blog/2009/12/13/mind-control-for...
The Emotiv headset has some mechanical issues - the pads are a pain to wet and place and keep falling out. The data also appears to be super encrypted and quite hacker-unfriendly, so I stopped reading there.
http://www.joystiq.com/2010/01/27/review-emotiv-epoc-tough-t...
http://thingist.com/t/item/23089/ (some angry words)