The evidence problem: blinding a brain implant
Psychiatric neuromodulation is unusually hard to evaluate, and it is worth being precise about why. Mood is expectation-sensitive, so the placebo and expectation effects of undergoing brain surgery are enormous. Symptoms fluctuate and often improve on their own. Patients and clinicians can sometimes tell whether the stimulator is on. And the numbers are small — these are rare, severely ill people. Together these make sham control and blinding both essential and genuinely difficult, which is exactly why the open-label enthusiasm of the first generation did not survive contact with rigorous trials.
The N-of-1 biomarker trap
The personalised biomarker that makes closed-loop psychiatry possible is also its statistical soft underbelly. When you search a large number of candidate features — many sites, many bands, many connectivity measures — against a short, sparse record of one patient's mood, some feature will correlate with the symptom trajectory by chance alone. Reporting that feature as 'the patient's mood biomarker' without correcting for the size of the search is how the field fools itself.
Why personalised biomarkers need multiplicity control. If you test K independent candidate features at level α under the null of no real relationship, the chance that at least one clears the bar spuriously grows roughly as Kα. With hundreds of site-band candidates, an uncorrected 'significant' biomarker is expected even when none is real.
The remedy is the same discipline good motor-BCI already uses, translated to this regime: pre-register the search, cross-validate against held-out symptom variation, and above all prospectively validate — show the biomarker predicts future symptoms it was not fit on before you let it drive stimulation. This is the concrete meaning of regulatory science for adaptive decoders: how do you approve a device whose decision rule was learned on, and personalised to, a single brain?
The ethics only this field faces
A decoder that moves a cursor and a stimulator that moves a mood raise very different questions. Modulating affect can change what a person wants, decides and feels — so the ethics of mood modulation reaches into the sense of agency and personal identity in ways cursor control never does. Patients on psychiatric DBS have described feeling more like themselves — and, in other cases, less; some report stimulation-induced states such as unexpected mirth or hypomania that are clearly the device, not them.
Closing the loop sharpens this. An autonomous controller that adjusts your mood-state without a moment-by-moment decision by you blurs authorship: when the device nudges an affective state, whose action is it? This is why mental integrity and cognitive liberty appear in emerging neurorights frameworks specifically around therapeutic mood modulation, and why the obligation to maintain and not abandon an implant that has become part of a person's felt self is treated as an ethical duty, not an afterthought.
An honest reading of where this goes
So where does an honest reading land? The near-term wins are likely to be modest and specific: adaptive stimulation that reduces side effects and battery drain at matched efficacy; better individualised targeting via connectomics; and a few carefully-validated biomarkers for the best-characterised circuits. The grand vision — a device that reads any mood and titrates it away — remains aspirational, gated less by hardware than by the biomarker, the trial designs, and the ethics. The right posture is neither dismissal nor hype: this is a real frontier with a handful of real demonstrations, advancing one honestly-validated patient at a time.