40 Volunteers Try to Precognize Random Faces Using Amiga 1000
By June Hargrove ·
Forty volunteers were subjected to "image roulette" to see if the human brain can precognize a face before it is randomly generated in a study found in CIA FOIA files.
A subject sits before an Amiga 1000, watching a sequence of noses, mouths, and eyes shift rapidly on a color monitor. This is "image roulette," a mechanism designed to remove conscious choice from the act of facial recognition. The goal is not to identify a known person, but to guess a face that the computer has not yet created.
According to a document from the CIA FOIA Electronic Reading Room titled "'VISAGES': A COMPUTER-BASED TEST OF FACE PRECOGNITION," researchers Mario Varvoglis and Michel-Ange Amorim of the Laboratoire de Recherche sur les Interactions Psi sought to determine if subjects could precognize the features of a randomly composed face. The study focused on whether the human brain's inherent "hardwired" function for face-recognition could translate into a sensitivity for precognition.
Photo-fit Kits and Digitization
The experiment relied on a subset of the "Photo-fit" kit provided by the central police department of Paris. In standard police work, these kits use transparencies of male facial elements—eyes, noses, mouths, and jaws—which can be combined to approximate the face of a criminal or missing person. For the Visages test, the researchers selected 16 different instances for four specific categories: eyes, nose, mouth, and facial outline (including hair, forehead, and jaw).
These physical images were passed into the Amiga 1000 via a surveillance camera and an interface for "digitization." The researchers used computer graphic tools to maximize the clarity of the resulting bit-map screens. The 16 instances of each feature were then arranged in a 4 x 4 array, grouped by resemblance. This layout allowed for different degrees of accuracy in the subjects' guesses, ranging from a "half-page" hit (identifying a general characteristic, such as "lots of hair") to a direct hit on the specific image.
Scanning and Timing
The study involved 40 volunteers—35 women and 5 men—ranging in age from 19 to 59. Most were recruited via an article in a popular women's magazine. Before beginning the Visages test, subjects underwent a different test called "Volition" on an Apple-based computer.
Once at the Amiga, subjects were asked to use their intuition to guess which four features the computer would randomly select to build a face. The researchers employed two different task-modalities to test these guesses. In the "Scanning psi task," the 16 possibilities for a feature were displayed simultaneously in the 4 x 4 array, and the subject chose one using the mouse.
In the "Timing psi task," the images appeared in a rapidly shifting sequence. The subject clicked the mouse to stop the "image roulette." To ensure the subject could not consciously time their click to hit a specific image, the program was designed so that the actually selected image was generated randomly immediately after the mouse input.
Timing Task Results
The hardware supporting the experiment included a color monitor and a 2-megabyte random-access memory extension. The software was written in a compiler-language called "The Director," which was explicitly oriented toward graphics and sound. The researchers used the Director language's pseudo-random function, reseeded by the Amiga clock in microseconds, to select the target face.
To add a layer of "meaningfulness" to the targets, the program originally assigned each randomly generated face a name from a file of 80 common French names and a biography composed of six statements drawn from 20 categories, such as "mood and temperament" or "paranormal experiences." However, the researchers noted that subjects found the biographies "incongruent" with the nature of the task. This led the experimenters to drop the assessment of the biography factor during the study.
When the results were evaluated, the global test for the experimental condition yielded a significant chi-square (p=.013). However, further analysis showed that this significance was driven specifically by the Timing task modality, which produced a chi-square of p=.006.
To verify these findings, the researchers conducted a simulation study using a modified version of the program where the subjects' guesses were replaced by random numbers. This process continued until the number of simulation runs equaled the total number of experimental runs (212). The simulation study yielded chance results.