Skip to main content

Verified by Psychology Today

Artificial Intelligence

A New Tool to Assess Possible Dream-Related Psi Phenomena

We may now be able to objectively test for psi events.

Even though there have been several meta-analytic studies on psi-related dream effects (e.g. see International Journal of Dream Research Volume 10, No. 2 (2017)) which have concluded that these effects are reliable, substantive, and large, the larger scientific community seems unaware or unconvinced of the reality of psi. I, therefore, propose a new test of dream-related psi effects which could give us a new tool to assess the reality of these effects.

With the advent of fMRI-related multi-pattern voxel analysis (MPVA) used in concert with machine learning algorithms trained to identify and classify fMRI brain activity patterns as certain categories of images, we could potentially bypass subjective reports of dreamers to literally look and see what "receivers" in psi experiments were dreaming about.

It is possible to use AI (artificial intelligence) machine learning tools to learn how to accurately classify what given brain activity patterns (defined via MPVA) are mediating or depicting. First, you use AI to learn to accurately classify brain patterns associated with several image categories (e.g. faces, cars, houses, animals, etc.) generated while participants view thousands of pictures of instances of these categories while under the fMRI magnet.

After analyzing or “looking at” brain activity patterns of thousands of these fMRI signals associated with viewing instances of each of these categories, the AI machine learns to very accurately classify characteristic brain activity patterns associated with each of these categories and specific instances of these categories. So in effect, when it “sees” brain activity pattern X, it classifies that pattern as an image of face number 1, and so on. We therefore can look at brain activity patterns of a person who is dreaming and, in theory, accurately identify what that person is dreaming about.

Now, in a standard psi experiment, a sender “sends” an image to a receiver who is asleep in REM in some other location. The receiver “receives” the image in a dream and is awakened by the experimenter who asks for a dream report. The receiver and sender do not know one another and the experimenter and judges (see below) are all blind to what the image is that the sender is sending. Independent judges then score the dream report in terms of similarity of the images in the dream report vs. what the sender was supposed to send. If the sender was supposed to send an image of President Obama, for example, and the dreamer reports a dream of President Obama, then the judges would score it as a hit.

Neither skeptics nor proponents of psi particularly like one aspect of this basic psi experimental paradigm. The one loophole in the experimental set-up has always been the subjective nature of the dream report. The dream report is subject to biased recall and to other uncontrollable factors associated with the fallibility and motivational states of the dreamer. How do we really know that the receiver really had a dream image of Obama?

With the above innovations in AI-linked MPVA fMRI analyses, we can bypass the subjective reports of the dreamer altogether. To know objectively whether or not the dreamer “received” the image the sender was sending, we would simply look at what the brain activity patterns tell us.

So for example, if we ran a double-blinded psi experimental paradigm like the one described above and we found that the brain activity pattern results consistently indicated very high hit rates while the subjective dream reports were consistently ambiguous with the judges returning lower hit rates, I would personally conclude that psi has been objectively verified, as that set of results would imply that hit rates were vastly underestimated in past experiments due to the cloudy reports of the dreamer who may or may not have good recall of what he actually dreamed. Suppose further that the dream reports and brain activity patterns matched one-another up and down the line... then once again psi will have been verified. But if both dream reports and brain activity patterns consistently return low or chance hit rates, then I think the evidence would argue against psi.

More from Patrick McNamara Ph.D.
More from Psychology Today