Using Machine Learning Methods to Automatize Target Pool Design and Judging in Free-Response ESP Experiments: A Proof of Concept
David J. Acunzo & Jeremy Ky
Department of Psychiatry and Neurobehavioral Sciences, School of Medicine,
University of Virginia, Charlottesville, VA, USA
—
Extra-sensory perception (ESP) experiments using a free-response design, such as remote viewing or ganzfeld, require two labor-intensive steps that forced-choice design typically do not involve. The first one is the design of a large target pool, from which a single target will be drawn at random for each trial. The target pool must be sufficiently large to minimize the probability of drawing the same target more than once in an experiment. Targets also need to be grouped by sets of four or five to allow for human judging. This allows the blind judge to rate the correspondence of the percipient’s description of a given trial between the target and a small set of decoys. The targets within a set have to be as different as possible to facilitate the work of the judges and to help maximize the variance of rating scores within a set.
The second step is the judging itself. One or several judges are required per experiment, which adds an additional layer of subjectivity and variability in the experimental procedure (Katz et al., 2021). Efforts by various authors have been attempted to systematize this step by coding the targets and the percipients’ responses (Honorton, 1975; Jahn et al., 1980; May et al., 2014), but these methods have not been widely adopted to date. In addition, while judging is systematized, the encoding of the percipients’ responses is still time-consuming and may be strongly dependent on the encoder’s subjectivity.
These two steps are highly arbitrary, and there are no clear guidelines or common reference on how they should be conducted. There is a need to facilitate free-response ESP research, which is particularly resource-intensive. In addition, setting up easily applicable standards and systems to build pools and evaluate descriptions will help researchers to replicate experiments.
Machine learning tools offer an opportunity to make progress on all these fronts. In a recent commentary, we proposed to leverage natural language processing tools to automate the judging procedure (Acunzo, 2023). The current abstract follows up on and extends this idea to present a method to partially automate (1) the design of a target pool constituted of digital images, and (2) the judging stage. Toy examples are presented to illustrate the potential of these approaches.
—
David J. Acunzo, Ph.D., is Assistant Professor in the Division of Perceptual Studies at the University of Virginia School of Medicine.
After a degree in telecommunications engineering from IMT Atlantique (France), David obtained his PhD in neuroinformatics from the University of Edinburgh (UK) and then occupied research positions at CIMeC – Center for Brain/Mind Sciences, University of Trento (Italy), and the Centre for Human Brain Health, University of Birmingham, UK.
He has worked on a diverse array of cognitive neuroscience topics including vision, attention and semantic processing using various computational and functional neuroimaging techniques including EEG, MEG and fMRI.
His current research interests include abnormal perceptions, response to hypnotic suggestion and extraordinary experiences including mystical and psi experiences.
—
The Parapsychological Association is an international professional organization of scientists and scholars engaged in the study of psi (or ‘psychic’) experiences, such as telepathy, clairvoyance, psychokinesis, psychic healing, and precognition. The primary objective of the PA is to achieve a scientific understanding of these experiences.
Follow the PA: https://linktr.ee/parapsychologicalassociation
Donate to support the mission of the PA: https://donorbox.org/support-the-parapsychological-association
Published on November 11, 2026
