Scientists Have Used Brain Scans To Decode Complex Thoughts
If you could have three superpowers, what would they be? We're willing to bet that one of them is the power to read minds. That possibility may be expanding beyond comic books: researchers have figured out how to use brain scans to read people's thoughts. Certain thoughts, anyway.
On A Wavelength
Mind reading is nothing new. We've had technology that allows people to control wheelchairs, toy cars, and even robots. But so far, that tech has been primitive: most of it uses an electroencephalogram (EEG) to read the brain's electrical signals, which only provides so much data. In April 2017, for instance, researchers announced a breakthrough EEG technique that could guess a single-digit number in a subject's mind with 90 percent accuracy. That's certainly a step on the road to mind-controlled typewriters, but it's a far cry from reading thoughts.
Researchers at Carnegie Mellon, however, are using a completely different technology. For a 2017 study published in Human Brain Mapping, they used machine-learning algorithms—a form of artificial intelligence—to analyze brain activity from fMRI brain scans.
Unlike EEG, which monitors electrical signals, fMRI analyzes changes in blood flow to help scientists infer which parts of the brain are experiencing the most activity. That's a heck of a lot more precise, and it's why the researchers were able to achieve such an impressive breakthrough.
Concepts > Words
Here's how they did it: the researchers recruited seven people to read 239 sentences inside an fMRI machine. As their brains were scanned, the machine-learning algorithm looked for patterns, examining how the volunteers' brain activity corresponded to the content of those sentences. For the 240th sentence, the algorithm went in blind: it had to figure out the basic content of the sentence from the brain scan alone. It did just that, with 87 percent accuracy. To make absolutely sure their method worked, the researchers also had the program do the same thing with each sentence—they only provided the fMRI data, and the program had to identify the content of the sentence, which it also successfully achieved.
The achievement alone is cool enough, but what it taught the researchers is pretty groundbreaking: humans don't think in words; we think in concepts, and this technique shows scientists how. Take the sentence, "The witness shouted during the trial," for example. According to the research, that concept uses an "alphabet" of 42 components, like the person involved, the setting, sizes, social interactions, and physical actions. Each component is processed in a different brain system, which is what the researchers' algorithm used to decode the volunteers' thoughts.
This is an exciting step that could lead to a whole variety of discoveries. "A next step might be to decode the general type of topic a person is thinking about, such as geology or skateboarding," says lead author Marcel Just. "We are on the way to making a map of all the types of knowledge in the brain."