Computers Learning to Read Human Intentions

Computers can now be trained better than ever before to be able to predict what a human being plans to do, according to a new study published Thursday.

Using a combination of technologies including pattern recognition software and functional magnetic resonance imaging (fMRI), the authors of the study investigated whether a computer could predict ahead of time which of two set tasks a subject had planned to do. What they found was that the computer was accurate about 70 percent of the time.

The researchers, led by Dr. John-Dylan Haynes of the Max Planck Institute for Human Cognitive and Brain Sciences, asked subjects to choose whether they wanted to perform an addition problem or a subtraction problem, and to keep their decision to themselves for variable amounts of time.

The fMRI technology monitored each subject's brain responses during the time they were thinking about their decision. The software, which had to be trained ahead of time to "learn" each subject's mind, then compared those brain scans with the patterns it had learned were associated with different thought processes, and made its prediction.

A Mental Fingerprint

"In the last few years, it has slowly emerged that spatial patterns are like a fingerprint of a thought in the brain," Dr. Haynes told TechNewsWorld. "Each thought has a different fingerprint, so if you can read the fingerprint, you can read the thought."

It used to be that scientists focused on entire areas of the brain when examining tasks, looking for which brain areas were associated with vision, for example. This advance in thinking, however, focusing instead on spatial patterns, has allowed researchers to go beyond finding the general brain area, and to zero in on the particular thought or image a person is thinking about, Haynes explained.

The computer's ability to predict intentions is currently limited to contexts in which there are just a limited set of options, however. For that reason, Haynes believes the most likely early applications of his team's discovery is probably lie detection, because lie detectors typically focus on a limited set of choices -- "yes" or "no," for example.

No Mind-Reading Yet

"Computers and brain scanners can read out certain thoughts, but we don't know how to read out arbitrary thoughts," Haynes declared. "The main limitation is that we can't generalize to new thoughts we haven't recorded before."

So, while it's tempting to give in to wild speculation about computers that can read our minds, that scenario is still in the distant future, he added.

Indeed, while similar research based on electroencephalograms (EEG) is being used to help paralyzed people communicate, for example, comparable applications of this fMRI-based research may be further off.

"I think the primary import of this is its contribution to our understanding of how the brain works, and what areas of the brain are doing in the course of performing an activity," Jonathan Wolpaw, chief of the New York State Department of Health's Wadsworth Center Laboratory of Nervous System Disorders, told TechNewsWorld. "fMRI is a relatively new technique that provides important new information, but applications to communication are in the future."

More Advances Needed

"We need the brain scanners to get cheaper and smaller," Haynes agreed, "and that won't happen in the next 100 years."

Still, the area is rich for potential applications, and much is happening already, James Cavuoto, editor and publisher of Neurotech Reports, told TechNewsWorld.

"A lot of exciting things are going on, some of them even commercially viable," Cavuoto said, noting brain-computer interfaces by companies such as Cyberkinetics that can help "locked" patients such as quadriplegics and victims of Lou Gehrig's disease communicate with the outside world.

"That's the goal here -- to re-establish communication for those who have lost the ability to speak or control their body, and give them a little more control over the outside world," he concluded.