40 Volunteers Were Tested for Psychic Criminology Using an Amiga 1000
By Eleanor Voss ·
Experts sought "psychic witnesses" to identify unknown criminals using random faces, while the personal psychological data of 40 subjects remained unanalyzed.
An Amiga 1000 computer, equipped with a color monitor and a two-megabyte random-access memory extension, sat at the center of an experiment designed to see if a human could anticipate a face that did not yet exist. The machine used a "mouse" for input and a compiler-language called "The Director" to manage graphics and sound.
According to a document from the CIA FOIA Electronic Reading Room, this project, titled "VISAGES," was conducted by Mario Varvoglis and Michel-Ange Amorim of the Laboratoire de Recherche sur les Interactions Psi. The goal was to explore "face precognition"—the idea that a person might be able to perceive the features of a randomly composed face before the computer actually generated it.
Photo-fit Kits and Amiga 1000s
To build the target faces, the researchers used a subset of the "Photo-fit" kit, a tool provided by the Préfecture de Police of Paris. In a police setting, this kit consists of transparencies showing various male facial elements—eyes, noses, mouths, and jaws—which an interviewer mixes and matches to approximate a witness's memory of a criminal.
Varvoglis and Amorim digitized these images using a surveillance camera and an interface that allowed for video input. They selected 16 distinct instances for four specific facial elements: the mouth, eyes, nose, and facial outline (which included the hair, forehead, and jaw). These images were arranged in 4x4 arrays. The researchers grouped the images by degrees of resemblance, such as "little hair" versus "full hair," to create a scoring system that could track different levels of accuracy, from a vague "half-page hit" to a direct hit on the specific image.
The motivational driver for the study was the potential for "psychic criminology." The authors noted that while police often rely on eyewitnesses to reconstruct a suspect's face, those witnesses may be "unavailable, or may be unreliable." The researchers questioned whether "psychic witnesses" could be used to identify the facial characteristics of an unknown person. They hypothesized that because face recognition is a "hardwired" human function, it might translate into a special sensitivity toward face-precognition.
Scanning vs. Timing
The experiment involved 40 volunteers—35 women and 5 men—ranging in age from 19 to 59. Most of them had come to the laboratory after reading an article in a popular women's magazine.
To account for potential response biases—the idea that people might simply choose faces they find aesthetically pleasing—the researchers employed two different task modalities. In the "Scanning" task, the subject was presented with all 16 possibilities on the screen at once and used the mouse to select one. In the "Timing" task, the images shifted in a rapid sequence, creating a "roulette" effect. The subject had to decide when to click the mouse to stop the sequence.
To prevent the subject from simply timing their reaction to a specific image, the program was designed so that the image actually selected was randomly generated after the mouse input. This ensured that the subject could not consciously select a particular target through speed or reflex.
To add a layer of "meaningfulness" to the fictional faces, the researchers associated each target with a randomly selected name from a file of 80 common French names. In half of the trials, they also attached a biography drawn from a pool of 200 statements covering themes such as "paranormal experiences," "social life," and "health."
A p-value of .006
The results were evaluated by comparing the distribution of hits against an expected distribution of chance. According to the record, the test yielded a "significant chi-square for experimental condition (p=.015)," while simulation studies and chance results did not show the same effect.
When the researchers broke the data down by the two modalities, they found a more pronounced result in the Timing task, which yielded a p-value of .006. The authors suggested that this might indicate a capacity for receptive psi, though they acknowledged the limitations of a laboratory environment compared to the "motivational dynamics of real-life psychic criminology."
The technical side of the randomness was managed by the Amiga's pseudo-random function, which was reseeded every cycle by the Amiga clock in microseconds and scrambled by a "Cyclic Redundancy Check" scheme. The creator of "The Director" language informed the researchers that the function yielded the expected range and frequency of values.
While the results of the face-prediction tasks were recorded, other data remained untouched. The researchers had collected personal and psychological data on all 40 subjects using French versions of the Personal Inventory Form (PIF) and the Myers-Briggs-Type-Inventory (MBTI), but the document states that these data had not yet been analyzed.