Psychophysical modulation of fringe visibility in a distant double-slit optical system

Radin, D., Michel, L., & Delorme, A. (2016). Psychophysical modulation of fringe visibility in a distant double-slit optical system. Physics Essays, 29(1), 14–22. https://doi.org/10.4006/0836-1398-29.1.014

AI Assessment

A two-year online double-slit experiment (1479 people, 2013 to 2014) testing whether directed attention shifts the visibility of interference fringes, with computer-run robot sessions as controls. The pattern is the whole story: in 2013 fringe visibility significantly decreased (4.78 sigma) and in 2014 it significantly increased (3.41 sigma), the opposite direction. The headline 5.72 sigma result is obtained only by reversing the sign of the entire 2014 dataset, which the authors justify by a feedback coding bug that inverted the 2014 goal and which they present as evidence for an active steering effect rather than passive observation. The control apparatus, optional-stopping tests, outlier-robustness checks, and an independent reanalysis are genuinely strong, but the absolute effect is about a 0.001% change in fringe visibility and the significance rests on a post-hoc sign flip. This audit reports what the paper ran, found, and disclosed; it takes no position on whether consciousness modulates quantum systems.

Provenance

DOI. 10.4006/0836-1398-29.1.014 · Physics Essays 2016, 29(1), 14–22. Received 17 February 2015, accepted 13 December 2015, published online 10 January 2016. Physics Essays is a small, specialist venue rather than a mainstream physics journal.

Study type. An online observational experiment with pre-specified metric and lag, plus a post-hoc analytic decision (the 2014 sign reversal) and several robustness analyses.

Authors. Dean Radin and Leena Michel (Institute of Noetic Sciences) and Arnaud Delorme (University of California, San Diego).

Funding. The Federico and Elvia Faggin Foundation, the Bial Foundation, and Richard and Connie Adams (stated in the Acknowledgments).

Data availability. Summary statistics and figures are in the paper; the raw line-camera data (about 140 GB) were stored on a web server and cloud. The authors state that Dr. Nicolas Trembley independently reanalyzed the raw data and confirmed the reported results.

Source basis. Every figure below is taken from the article’s own Abstract, Methods, Results, and Discussion.

What the paper reports

Following von Neumann’s proposal that an observer’s mind is part of quantum measurement, participants worldwide directed attention toward or away from a sealed, distant double-slit system while a line camera recorded the interference pattern at 4 Hz.¹ The metric was fringe visibility (the contrast between adjacent maxima and minima) at a pre-specified central fringe (“fringe 9”), analyzed at a pre-specified +9 s lag. The design is a two-year replication and extension of the group’s 2012 online experiment.

Due to a serendipitous coding error, the feedback used in 2014 accidentally reflected an increase in double-slit spectral magnitude. This mistake was not discovered until the 2014 data were analyzed.

How it was run

• Apparatus. A 5 mW HeNe laser, neutral-density filters, a 10 µm double slit separated by 200 µm, and a 3000-pixel CCD line camera 16 cm from the slits, in a sealed matte-black housing near the lab’s web server.
• Task. Anyone could register (Captcha plus email verification) and run a roughly 11 min session of 21 alternating epochs, starting with relax; concentrate epochs were 30 s, relax epochs a random 30 to 35 s to decouple from environmental periodicities.
• Metric. Fringe visibility v = (Imax − Imin) / (Imax + Imin), computed per camera frame after Savitzky-Golay smoothing; the analysis focused on the median concentrate-minus-relax difference at fringe 9, lagged +9 s, with a 10% outlier trim, and bootstrap resampling (5000 iterations) for the z-score. False Discovery Rate correction was applied across the 20 fringes; all probabilities two-tailed.
• Controls. A separate computer (Ubuntu Linux, Java) simulated a human participant and ran an automatic control session every hour, indistinguishable to the apparatus, generating thousands of control sessions.
• The 2014 sign reversal. A Flash coordinate-system error inverted the 2014 feedback so that participants were effectively steered toward the opposite goal; this was found only when the 2014 data were analyzed. “All data combined” therefore means the 2013 data combined with the sign-reversed 2014 data.

Results, as reported

Values are reproduced from the article’s Abstract and Results. The decisive interpretive point is that 2013 and 2014 produced significant effects in opposite directions, and the combined 5.72 sigma is obtained only after reversing the sign of the 2014 data.

Eleven-dimension audit

Pre-registration

Not preregistered. The fringe of interest (fringe 9) and the +9 s lag were pre-specified from the group’s earlier 2013 experiment, which constrains some flexibility. But the pivotal analytic move, reversing the sign of the 2014 data, was decided after the data were in and the coding error was found, and the authors themselves recommend that future work preregister the analysis.

Randomization

Relax-epoch lengths were randomized (30 to 35 s) to decouple the attention schedule from environmental periodicities, and conditions were counterbalanced. Participants were self-selected online volunteers rather than a random sample. The statistical null was built by bootstrap resampling.

Sensory leakage

Strong. The apparatus was sealed and physically isolated from participants, who interacted only over the Internet and received only a summary feedback number, never the raw camera data. The computer-run control sessions were indistinguishable to the apparatus from human sessions, which is the study’s central artifact defense.

Blinding

The outcome is instrumental, so there is no human rater. The robot control sessions and the time-matched experimental-versus-control comparison serve the blinding role, and they came out null while the human sessions did not.

Optional stopping

A genuine strength: the authors anticipated the critique and tested two forms directly. A within-session correlation between fringe visibility and session length was null (r = 0.0004), and an across-session correlation between effect size and number of sessions contributed was null (r = 0.09). Neither supports an optional-stopping artifact.

Outcome measure

Fringe visibility at a pre-specified fringe and lag, with a 10% outlier trim and FDR correction across the 20 fringes. The metric is principled (it normalizes for illumination drift), and its robustness to outlier-trim choice (tested from 0% to 50%) is shown. The one heavy analytic choice outside the pre-specified plan is the 2014 sign reversal.

Effect size

Very small in absolute terms: about a 0.001% change in fringe visibility, with per-session effect sizes around 0.08 to 0.10. The large sigma reflects the roughly 3000-session sample, not a large per-session effect. The authors are explicit that this is a subtle modulation, not wavefunction collapse.

Multiple comparisons

FDR correction was applied across the 20 fringes and all tests are two-tailed, which is appropriate. Residual latitude lies in the outlier-trim setting and, most importantly, in the post-hoc sign-reversal decision, which is not the kind of choice a multiple-comparison correction addresses.

Internal replication

Qualified. The study was designed to replicate the 2012 online experiment across two further years, but the two years went in opposite directions. They only combine into a single significant effect after the 2014 sign is reversed, so the internal “replication” is contingent on the steering interpretation rather than being a straightforward same-direction repeat.

External replication

The paper situates itself in a body of optical psychophysical experiments by, in its account, four principal investigators across three laboratories (including Ibison and Jeffers 1998, where Ibison’s attempt succeeded and Jeffers’ did not). An independent reanalysis of the raw data by Dr. Nicolas Trembley confirmed the reported numbers, but that is a reanalysis, not an independent replication, and the authors call for independent, preregistered replication with skeptically inclined co-investigators.

Transparency

High on disclosure. The coding bug and the resulting sign reversal are reported openly, the tiny absolute effect size is stated plainly, optional stopping is tested, the independent reanalysis is named, and preregistration is recommended. Two points a reader should note: the English abstract reports 5.72 sigma with a 0.17 sigma control while the French résumé reports 5.60 sigma with a 0.04 sigma control, and the abstract’s session counts (2985 experimental, 5738 control) differ slightly from the Results text (2984 experimental, 5723 control).

The adversarial record

• The headline depends on a sign flip. 2013 and 2014 produced significant effects in opposite directions, and the combined 5.72 sigma exists only after reversing the entire 2014 dataset. Even granting the disclosed coding bug, it is the active-steering interpretation that licenses combining opposite-direction years into one effect.
• The absolute effect is minuscule. An average 0.001% change in fringe visibility reaches high sigma only because roughly 3000 sessions were pooled. Statistical significance here is a function of sample size, not effect magnitude.
• No preregistration, self-selected sample. The participants are online volunteers and the analysis was not registered in advance, which the authors acknowledge and recommend changing.
• What the authors do right. This is one of the better-controlled studies of its kind: indistinguishable robot controls that came out null, time-matched experimental-versus-control comparisons, two explicit optional-stopping tests, outlier-trim robustness curves, an independent reanalysis of the raw data, candid disclosure of the coding bug, and an honest statement of the effect’s tiny absolute size. That candor is what lets a reader weigh the sign-reversal issue for themselves.