Lorne Semrau was a Tennessee psychologist charged with health care fraud, accused of billing Medicare and Medicaid under improper codes. His defense was that the billing errors were genuine misunderstandings of a notoriously complex coding system rather than deliberate deception — which made his state of mind the entire case.
So his lawyers hired Cephos Corporation to scan his brain.
Cephos was one of two companies commercializing functional magnetic resonance imaging as a lie detector in the late 2000s. The underlying research is genuine. In laboratory studies where subjects are instructed to lie, deception is associated with greater activation in prefrontal and anterior cingulate regions relative to truth-telling — consistent with deception requiring executive control, working memory and inhibition of the true response. The effect is robust at the group level and has been replicated many times.
Semrau was scanned three times. The first two sessions were reported as indicating truthfulness. The third, on a different set of questions, indicated deception. The examiner attributed that result to fatigue and re-ran a portion.
In May 2010, Magistrate Judge Tu Pham held a Daubert hearing and excluded the testimony. His opinion is the first substantial American judicial analysis of fMRI deception evidence and it is worth reading, because the reasoning is not hostile to neuroscience.
The court accepted that the technique had been tested and had been subject to peer review. It found the known error rate unestablished for real-world individual use, found no controlling standards for the technique's operation, and found that while the research had general acceptance as research, its application to individual credibility determinations did not. It also excluded under Rule 403, noting that credibility is the exclusive province of the jury, and that the third test's inconsistent result — explained away by the expert — undermined reliability.
The core problem is the one the court identified and it is not fixable by better scanners.
Group-average differences do not license individual diagnosis. A study showing that a group of people instructed to lie shows more prefrontal activation than the same group telling the truth establishes an average difference across many trials and many subjects. It does not establish that a given activation pattern in one person on one occasion indicates deception, and the distributions overlap substantially.
Beyond that: laboratory subjects are instructed to lie about trivial matters with no consequences and no motivation to defeat the test. A defendant facing prison has both. Countermeasure studies on fMRI deception protocols have shown that simple mental strategies reduce accuracy considerably.
The generalizable finding — and it is the closing argument of Part 4 — is that a real neural correlate of cognitive effort is not a forensic test of truthfulness. That sentence applies equally to the polygraph, to thermal imaging, to pupillometry, to voice stress analysis, and to whatever arrives next.
Cross-ref: polygraph validity chapters