In Your Head
As brain-scanning technology gets better at reading our minds, sexbot designers are pretty sure they already know what we're thinking.
By November 6, 2018 - last reviewed on November 5, 2018published
How close are we? When reading about an anticipated technological advance, this question is always at the core. When can we order a pill to lock in weight loss? Can we tell our grandkids they'll teleport? In two new books, researchers make a strong case for patience and respecting the process.
The New Mind Readers, by Stanford cognitive neuroscientist Russell Poldrack, is a guide to how far neuroimaging might advance toward true thought detection. His verdict: shockingly far, but perhaps never far enough for those eager to use it to create a world free of secrets.
Once a skeptic of neuroimaging's potential, for whom the thought of entering an MRI tube elicited profound anxiety, Poldrack has become a convert to functional magnetic resonance imaging (fMRI). The process detects "shadows" of brain activity by tracking its effects on oxygen in the blood—activity that seems to make parts of the brain "light up" within the magnetic field generated by the scanner.
"It's amazing that fMRI works at all," Poldrack writes. "Its success relies upon a set of chemical and biological dominoes that all had to fall into place…almost as if nature conspired to help make it just a bit easier for us to understand how the brain works." Starting in September 2012, he subjected himself to 104 scans over an 18-month period to produce "My Connectome," an attempt to glean as much information as possible about connections between brain regions from repeated imaging of one subject. What did he learn about himself from the study? "Depressingly little," he writes, but as a contribution to the field, the results were more significant, showing how a brain changes over even a relatively short time, as well as how caffeine, food, and mood affect its function.
Given the work that he and others have put in, Poldrack is aggravated by breathless reports of scans purported to show "hot zones" of psychological stimulation that reveal which Super Bowl ads people favor or how addicted they are to their cell phones. Such findings are unreliable, he insists, because people fallaciously resort to reverse inference—instead of accounting for the many possible triggers of activation, they focus on the one they're most curious about: "There is no simple one-to-one mapping between psychological states and activity in specific brain areas."
Even with the challenges, Poldrack writes, the field has started to produce findings that approach "legitimate examples of mind reading." For example, in 70 percent of trials, scientists could correctly predict, based on scan results, if subjects had chosen to add or subtract a set of numbers. Another test involved scanning subjects as they viewed 120 photos, then showing those photos to later participants and, based on their scans, predicting which image they had seen. In one condition, predictions were 92 percent accurate.
Still, fMRI scans do not "read" minds in the way that popular imagination perceives the trick. Instead, they help "decode" thoughts, Poldrack explains. Researchers can look at a scan and determine with some reliability what a subject was doing, feeling, or thinking about, but only after having done sufficient testing of similar questions to understand what they are seeing. He offers this analogy: Instead of watching a crowd react to three political candidates and judging which was most popular based on the amount of applause, neuroimaging teams hear only the applause, and then determine which candidate was speaking.
Poldrack doesn't think that the dawn of mind-reading machines is too far off, but he suspects it will emerge not from fMRI, but from something that has yet to be developed. The clash he describes between the potential and the possible can sometimes be poignant: There have already been serious discussions, for example, about whether potentially responsive brain injury patients in a vegetative state could be "asked" via a scanner if they wish to continue living. But Poldrack has a special distaste for Mehmet Oz and others promoting fMRI as a foolproof lie detector. The truth is that the scanners are easily thwarted by maneuvers as simple as changing one's breathing patterns or moving a finger.
Mind reading may remain decades away, but sexbots could be closer at hand. In Turned On: Science, Sex, and Robots, University of London artificial intelligence researcher and "robosexologist" Kate Devlin doesn't skimp on the hard science or the fun stuff as she reports on the race to produce fully functioning, AI-driven artificial sex partners.
No sex doll currently on the market could properly be called a companion, but Devlin takes readers on an international tour of the facilities where the next generation of silicone bedmates are being designed. Along with "a wall of 49 disembodied nipples and aureoles," "a tray of eye-wateringly large penises in various states of completion," and AI interfaces that need a human to charm them before they warm up, she also encounters researchers whose faith in the field has moved them to lobby for skin tones, gender identities, and cultural grounding as diverse as the potential market.
If we're to achieve what has been a popular fantasy since at least ancient Greece, there are daunting engineering and AI challenges to solve. We are not yet at a moment when "the machines can feel and understand us," Devlin writes. In the near term, the most likely role for sex robots may be as support for couples with mismatched desires, or whose relationship has been debilitated by illness, dementia, or death. And the future of the field may not be animatronic studs and sex kittens, but more holistic and sensitive human-technology interfaces. The robots are coming, but probably not to replace our partners or upend our core desire to connect: "Through layers of technology," Devlin writes, "we remain resolutely human."