The U.S. military is working on computers than can scan your mind and adapt to what you're thinking.

Since 2000, Darpa, the Pentagon's blue-sky research arm, has spearheaded a far-flung, nearly $70 million effort to build prototype cockpits, missile control stations and infantry trainers that can sense what's occupying their operators' attention, and adjust how they present information, accordingly. Similar technologies are being employed to help intelligence analysts find targets easier by tapping their unconscious reactions. It's all part of a broader Darpa push to radically boost the performance of American troops.

"Computers today, you have to learn how they work," says Navy Commander Dylan Schmorrow, who served as Darpa's first program manager for this Augmented Cognition project. He now works for the Office of Naval Research. "We want the computer to learn you, adapt to you."

So much of what's done today in the military involves staring at a computer screen, parsing an intelligence report, keeping track of fellow soldiers, flying a drone airplane - that it can quickly lead to information overload. Schmorrow and other Augmented Cognition (AugCog) researchers think they can overcome this, though.

The idea, to grossly over-simplify, is that people have more than one kind of working memory, and more than one kind of attention; there are separate slots in the mind for things written, things heard and things seen. By monitoring how taxed those areas of the brain are, it should be possible to change a computer's display, to compensate. If a person's getting too much visual information, send him a text alert. If that person is reading too much at once, present some of the data visually - in a chart or map.

At Boeing Phantom Works, researchers are using AugCog technologies to design tomorrow's cockpits. The military expects its pilots to someday control entire squads of armed robotic planes. But supervising all those drones may be too much for one human mind to handle unassisted.

Boeing's prototype controller uses an fMRI to check just how overloaded a pilot's visual and verbal memories are. Then the system adjusts its interface -- popping the most important radar images up on the middle of the screen, suggesting what targets should be hit next and, eventually, taking over for the human entirely, once his brain becomes completely overwhelmed.

Honeywell took a similar approach in recent trials, helping test subjects navigate through a simulated urban battle zone. They avoided enemy ambushes and evacuated wounded colleagues, all while a stream of messages poured on to their handheld computers. EEG meters, attached to subjects' heads, slowed the messages down when the subjects became overwhelmed. Average medical evacuation times were sped up by more than 300 percent, and ambushes dropped by more than 380 percent, as a result.

Zack Lynch, executive director of the Neurotechnology Industry Organization, says he's a bit suspicious of the claims because the improvements sound almost too dramatic. But "all in all, there are clearly tremendous advances" being made under the AugCog program, he notes in an e-mail. "(That progress) will bring benefits well outside the defense community," he says. "All you have to do is imagine what Wall Street will do when they get their hands on technology that can increase trading performance."

The Boeing and Honeywell teams were two of many groups presenting last October in San Francisco, when 75 or so neuroscientists, human-computer interface specialists and military researchers gathered in a Union Square hotel for the second Augmented Cognition International conference. (Other gadgetry included a morphing, brain-monitoring Tomahawk missile controller, a software assistant for a ship's captain, and a next-gen simulator of Marine squads.)

Schmorrow, a skinny, affable, mile-a-minute Navy pilot with five graduate degrees from computer science to experimental psychology, served as both emcee and the ringmaster, buzzing around the conference room. Schmorrow's vision is "AugCog everywhere" -- alarm clocks that sense where you are in your sleep cycle, Blackberries that don't vibrate when you're in a meeting.

"With technology, we're constantly interrupting people, burdening people," Schmorrow explains. "My phone is ringing, my Blackberry is buzzing, I've gotten 20 e-mails since we started talking. We just want people to be able to focus. Give them a bit of peace."

Early in life, Schmorrow didn't see himself as a military man. He was a long hair, playing in a "techno-industrial" band, until his grandmother, a World War II nurse, convinced him to apply to the Navy as a Christmas present to her. The recruiter told him he could see the world, study what he wanted and fly jets. Away he went.

Almost immediately, Schmorrow went into automation and cockpit design, as well as simulators meant to replicate pulling 10 Gs. . Eventually, he connected with University of Virginia psychology professor Denny Proffitt, and they began to brainstorm.

"We began with the idea that there was too much information out there these days for anyone to comprehend," says Schmorrow. "So how can we present it in a way that people will remember? Proffitt tells me, 'And wouldn't it be even better if we could figure out what people were doing, what they were thinking, so we could present them with the right things?'"

Schmorrow took the idea to Darpa. In late 2000, the agency put him in charge of a new program in Augmented Cognition. Initially, the goal was to figure out how to monitor brain activity while it was happening - and then have that affect a computer's display of information.

By the summer of 2003, in tests at the Navy's Space and Naval Warfare Systems Center near San Diego, they pulled it off. The next phase was even more ambitious: Schmorrow's researchers had to get that adaptive unit to work well enough to boost a user's working memory by as much as 500 percent. That launched more complicated experiments; like Boeing's killer drone cockpit.

Now, more than six years into the program, Darpa's involvement in Augmented Cognition program has mostly wound down. But the other military services, as well as academic and corporate labs, have picked up on the agency's efforts.

The work is far from over. In some scenarios, Boeing's AugCog controller shows only minor performance improvements over a more standard approach. At times, Honeywell's test subjects did their jobs more slowly when rigged up the EEGs. Other AugCog demonstrations I saw were rudimentary, like the Navy first-person shooter that sends more lumbering bad guys in your direction if your heart rate drops. (Not that the game ever gets that challenging.) But the basic building blocks of such a system -- sensors that can monitor brain activity while it's happening, and algorithms to let a computer respond, have now been put into place, because of the Darpa kickstart.

"We had this crazy notion," Schmorrow says, "and now it's real. It may take five years, or 10, to get into the field. But it's real."

Meanwhile, Darpa has started a new program, based largely on the same sensing technologies developed for AugCog: Neurotechnology for Intelligence Analysts. Even the best parsers of satellite imagery often miss the terrorist hideouts or missile silos hidden in the pictures taken from orbit. In tests, the Darpa project is improving these intelligence officers' accuracy as much as 600 percent. The secret is tapping into their unconscious minds.

The brain's visual memory centers fire about 250 to 400 milliseconds after someone spots a target, even if he doesn't realize what he's seen. Using infrared, magnetic and electrical sensors, researchers at Honeywell and the Oregon Health and Science University were able to use those unwitting neural spikes to pick likely "hot spots" in a satellite picture, where targets might be.

In one experiment, image arrays that usually took an experienced analyst an hour to scan were handled in 10 minutes. In another test, a smaller set of images that took about eight-and-a-half minutes to pore over, unaided, was scanned in about 80 seconds.

If these kinds of results can be repeated consistently, it could be a major advance. Satellite surveillance is on the rise -- and there aren't enough analysts to keep up with the work. If a neurotech system like this can do a basic triage of the images first, the chances of finding the pictorial equivalent of a needle in a haystack dramatically increase.

Researchers have been trying all kinds of ways to boost this rate. In the end, it may turn out, as Darpa officials note, that "the human visual system is still the best target detection apparatus" there is.