Lie still. A symbol appears on the screen above your face, and it tells you what is coming: a hot touch that will be mild, or a hot touch that will not be. Sometimes the symbol tells you nothing at all, and in that gap, with nothing to go on, your brain does what brains do. It guesses. Irina Strigo and colleagues at the San Francisco VA and the University of California built their whole study around that guess, and what they found, published in The Journal of Pain on 12 February 2026, is that the guess has an architecture. It is not one thing people do better or worse. It is two separate things, running at once.
They scanned 43 adults living with non-specific chronic low back pain (12 men, 31 women, mean age 46.7 years) during certain and uncertain anticipation, then let a machine learning classifier sort them by which brain regions actually carried the information. The first split, which the authors call the 'How' phenotype, divided people by insula subregion: 29 leaned on the dorsal anterior insula, the part that behaves like an appraiser, coding precise predictions about what is coming; 14 leaned on the ventral anterior insula, the part that behaves like an alarm, amplifying anticipatory distress when the signal is ambiguous. The second split, the 'What' phenotype, sorted people by the content of the guess under uncertainty: optimistic, pessimistic, or unbiased. The two splits are conceptually independent, and that is the elegant part, but they converged in the data. About 90 percent of the optimists and about 80 percent of the pessimists sat in the dorsal group. Ninety percent of the unbiased sat in the ventral group. People whose brains evaluate the threat form a view about it. People whose brains react to it do not bother forming one.
The phenotypes also showed up in the body. On quantitative sensory testing of the system by which the body damps its own pain, the dorsal group came out more analgesic (mean balance score -18.2) and the ventral group less so (mean +0.9; p = 0.02). The dorsal group had stronger brainstem connectivity, the ventral group stronger amygdala connectivity. On a questionnaire measuring habitual distraction from bodily sensation, the dorsal group scored as the more practised distractors. None of this was assigned, randomised or manipulated. It was observed, in one group of people, at one point in time, and observation of this kind describes a pattern; it does not establish a cause.
Then came the part that gives the paper its edge. Twenty-nine of the 43 went on to eight weeks of smartphone-based interoceptive attention training, a mind-body approach that asks you to turn toward the sensation rather than away from it, and were scanned and assessed again. Seventeen of those 29 were from the dorsal, cognitive-evaluative group, and it was that group that improved: significantly on pain outcomes, and significantly on the non-distraction measure (change 0.598, p = 0.018, Cohen's d = 0.86), while the ventral group managed only a marginal trend. Treatment response had been pre-defined as a 30 percent or greater improvement on the PEG scale and compared between the groups with Fisher's exact test, the right test for numbers this small, which is itself a reminder of how small these numbers are. The authors' reading, offered as interpretation rather than proof, is that a therapy built on non-judgmental evaluation may need an evaluating brain to work with.
Nobody in this study had dementia. The participants were in their forties, the condition was back pain, and the paper makes no claim about any dementia. We report it here because of what it does to a sentence families say to each other all the time: he just will not engage with it. The therapy that helped the person in the next chair, the breathing exercise, the body scan, the app with the gentle voice, may ask for a kind of cognitive appraisal that a given brain, for reasons written into its insula rather than into its character, is not currently doing. Whether that logic carries into conditions that damage the insula and the interoceptive network is an open question, and an untested one. It is a question worth asking out loud, and nothing in these 43 scans answers it.