The program has built up a vocabulary of 1,000 words and can parse simple sentences from subjects' brain patterns. Of course, the project needs to deal with the frustrating nuances and ambiguities of language, but it's been surprisingly successful. So far, the program is 90% accurate in predicting what subjects are thinking, according to Pomerleau.
The problem is that there are few computer users who have the desire, or the financial wherewithal, to sit in a $2 million brain scanner to compose a memo to the boss about a new marketing campaign. "A big leap is needed in the sensing technology, to a point where it can be miniaturized," says Pomerleau.
That leap may be at hand. Small electroencephalograph (EEG) sensors that track and interpret brain activity can be built into a headset or cap and may prove be a good stand-in for interpreting fMRI readings.
Primarily used in medical research, such devices are also appearing in everything from "neuromarketing" aids (wireless headsets that register test subjects' responses to marketing and branding) to crude toys that, for instance, let you pretend you're a Jedi knight by controlling a ball's height with mental power. Recently, a group of German engineers operated a specially modified car with one.
[ Related reading with cool videos: A mind-blowing look at today's mind-controlled, mind-reading technologies ]
On the computer interface front, Austria-based G.tec showed its Intendix system at this year's CeBit show in Germany. It uses an EEG cap studded with electrodes in conjunction with software you load on a Windows PC. The interface is laid out like a typical qwerty keyboard, with a few additional symbols for things like printing and sending email. After training the system, all you do is stare at the Intendix screen and think about the letters, numbers or symbols to spell out your message.
At the moment, the system can recognize about five characters a minute -- not exactly speed typing, but it's a start. Because of the system's $12,000 price tag, brain-powered computing will probably first be used for people with limited voluntary muscular control or "locked-in syndrome" diseases including amyotrophic lateral sclerosis (ALS), a.k.a. Lou Gehrig's disease.
A neural interface would open a new world for them, and eventually for the rest of us. But it could be decades before the technologies become advanced enough -- and inexpensive enough -- to make sophisticated brain-computer interfaces mainstream.
"The payoff here could be huge," says Supratik Guha, director of the physical sciences department at IBM. "But there's a lot of work that needs to be done to make this type of interface work."