From The Wall Street Journal:
When magnetic resonance imaging came into common use in the 1980s, it made the human brain visible in ways it had never been before. For the first time, we could see the soft brain tissue of a living subject, at a level of detail that could be observed previously only in autopsies. For doctors trying to help patients whose brains were damaged or diseased, MRI provided an invaluable snapshot of their condition.
By the 1990s, researchers had begun to measure changes in brain regions by using “functional” MRI. The technique detects oxygenated blood flow, revealing brain activity, not just brain structure. For cognitive neuroscientists, who study mental processes, fMRI was a godsend: It made it possible to identify which parts of the brain react to, say, faces, words or smells. It was a window through which to see the brain making sense of the external world. Suddenly we could watch human thought rippling across the rainbow-colored regions of brain scans.
Today, fMRI has been joined by newer tools, some still in development, that would allow scientists to track our mental states with ever greater precision. Researchers are generating enormous quantities of brain scan information, and they are analyzing these sets of “big data” with the latest computational techniques, especially machine learning, a subfield of AI that specializes in finding subtle, hard-to detect patterns.
What does all of this amount to? The start of a revolution. Scientists are beginning to unravel the question of how our material brains form our intangible minds. Though primarily motivated by medical and therapeutic goals, this research may have the greatest practical impact in areas such as product marketing, computer interfaces and criminal justice. Ultimately, it may help to answer fundamental questions about consciousness and free will, or even lead the way to preserving the knowledge and memories of individuals long after their bodies have failed.
. . . .
In fact, sensing what words or word categories you are thinking about is one of the more impressive results of modern cognitive neuroscience. Jack Gallant and his caborators at the University of California, Berkeley, have produced a remarkably detailed map of which sections of the brain react to different words and semantic concepts. In a 2016 paper in the journal Nature, they described an experiment in which seven volunteers listened to two hours of stories from “The Moth Radio Hour,” a popular storytelling podcast, while their heads rested in the custom-formed cradle of an fMRI machine.
. . . .
The researchers recorded changes in blood flow to each of tens of thousands of “voxels”—the units in a three-dimensional grid of locations in the brain. They then grouped the words spoken in the stories into 985 categories, each representing some common semantic dimension. (For example, the words “month” and “week” fall into the same category.) By correlating the brain activity with the words used to tell the stories, they were able to produce a detailed map revealing where these words and concepts were processed in the brain.
. . . .
Looking solely at their brain scans, the researchers were able to correctly identify which of eight such different tasks new subjects were performing about 80% of the time. It appears that how our brains work isn’t as unique to us as individuals as we might like to think.
With improved imaging technology, it may become possible to “eavesdrop” on a person’s internal dialogue, to the extent that they are thinking in words. “It’s a question of when, not if,” Dr. Gallant said. Other researchers are having similar success in determining what you may be looking at, whether you remember visiting a particular place or what decision you have made.
. . . .
Consider lie detection. At least two companies—No Lie MRI and Cephos—have tried to commercialize brain imaging systems that purport to tell whether a person believes he or she is telling the truth, by comparing a subject’s differing reactions to innocuous versus “loaded” questions. Their claims haven’t been independently validated and have received considerable criticism from the research community; so far, courts have declined to accept their results as evidence.
Another approach to assessing a suspect’s guilt or innocence is to determine whether he or she is acquainted with some unique aspect of a crime, such as its location, a particular weapon or the victim’s face. Several studies have shown that the brain’s reaction to familiar stimuli differs in measurable ways from unfamiliar ones. Anthony Wagner and his collaborators at the Stanford Memory Lab found that they could detect whether subjects believed they were familiar with a particular person’s face with 80% or better accuracy, under controlled conditions, though they noted in later research that subjects can intentionally fool the program. So—if the kinks can be worked out—crimes of the future may be solved by a “reverse lineup” to determine if a suspect recognizes the victim.
Though the current expert consensus is that these techniques are not yet reliable enough for use in law enforcement, information of this kind could revolutionize criminal proceedings. We may not be able to play back a defendant’s recollection of a crime as though it were a video, but determining whether they have memories of the crime scene or the victim may play as crucial a role in future trials as DNA evidence does today. Needless to say, the use of such technology would raise a range of ethical and constitutional issues.
. . . .
The new technologies may render moot the debate over torture and its supposed efficacy. “Enhanced interrogation” would become a thing of the past if investigators could directly query a suspected terrorist’s mind to reveal co-conspirators and targets. The world will have to decide whether such methods meet human-rights standards, especially since authoritarian governments would almost certainly use them to try to identify subversive thoughts or exposure to prohibited ideas or materials.
. . . .
Brain monitoring could also become more routine in employment. Selected high-speed train drivers and other workers in China already wear brain monitoring devices while on duty to detect fatigue and distraction. The South China Morning Post reports that some employees and government workers in China are required to wear sensors concealed in safety helmets or uniforms to detect depression, anxiety or rage. One manager at a logistics company stated that “It has significantly reduced the number of mistakes made by our workers.”
Link to the rest at The Wall Street Journal
What could go wrong?
No mention in the article of how fMRI lost a great deal of credibility by detecting brain function in a dead fish.
Also no mention of neuroplasticity: the same concepts are not necessarily mapped to the same locations in different people’s brains. There was one celebrated case of a boy with life-threatening seizures, a condition which could only be cured by totally removing the left hemisphere of the cerebrum. After the surgery, the right hemisphere began rewiring itself to take over the missing functions, and in fact, the boy learned to speak and the new speech centre of his brain was found on the right side. This is an extreme example, but it demonstrates the problem: trying to read mental activity by voxels is only one step more sophisticated than phrenology.
The dead fish scan took place about ten years ago. fMRI has progressed quite a bit since then. Note also, that the scans that showed activity were statistically uncorrected, a practice that was already questionable in 2009. I too am somewhat skeptical of the WSJ article, but you aren’t quite playing fair by citing an old article. (I’ve made the same mistake myself often enough.)
This article, only 3 years old, offers a much more complete view of the strengths and weaknesses of fMRI: https://www.vox.com/2016/9/8/12189784/fmri-studies-explained
fMRI measures brain blood flow in small chunks of the brain (each about the size of a grain of kosher salt). That may seem fine-grained, but on the scale of the brain, each chunk can contain thousands of neural structures, so it’s scale is closer to looking at chapters rather than words in a book. And the brain is a book whose subject we only vaguely know and written in a language we don’t understand. What are the odds that you would miss the point and project your own aspirations into a book written in an unknown language pulled off the shelf in an unfamiliar library?
I agree that the WSJ article was giddy. All that being said, lie detection using fMRI seems more credible than polygraph technology that has not changed that much from when it was invented close to a century ago when dial telephones were cutting edge. Today, polygraphs are usually not allowed in court, is discredited as often as not, and yet is still a big industry and used all the time.
I see. So, you are saying that this piece of junk science is not on a par with the junk science I suggested, but is on a par with a different junk science that you suggest instead? Junk is junk.
‘Better than polygraphs’ is a mighty low bar to set. And even the most fevered supporters of polygraphs never claimed that they conferred telepathy upon the operator.
With tremendous rudeness, the comment editor timed me out just as I was finishing an edit on the comment above. It was roughly to this effect:
‘Statistically corrected’ data are weaker, not stronger, than ‘uncorrected’ data, since (as any statistics nerd knows) you can, by choosing your algorithms carefully, ‘correct’ any complex data set to support pretty much any conclusion your heart desires. Unless the specific method of correction is itself published along with the raw data and the adjusted findings, the entire ‘corrected’ publication is merely an extended exercise in ‘because I say so’. And many researchers (not just in neurobiology, either) strongly resist any attempt to make them publish either their raw data or the method they used to arrive at their adjustments.
What could go wrong? Nothing of course, since our minds are simply memory cores like in a computer, recording just the facts and all such recordings are isolated from any previous or following recordings. All such recordings are permanent and never get altered by time, dreams, feelings, or future failings. All this is obvious, which is why there are never any disagreements in witness statements or when recalling some past memory of that time Uncle Roger made a fool of himself at Cousin Ema’s wedding… or did he?