Precognition and Presentiment: What Has Actually Been Tested?

Precognition sounds like prophecy until you look at the experiment

The word carries too much cultural weight.

Precognition suggests visions.

Warnings.

Dreams that come true.

Knowing what will happen before it happens.

Laboratory research usually asks a much smaller question.

Can performance on a task measured now correlate with a target or random event selected later?

That sounds less dramatic.

It is also much harder to explain if the effect is genuine.

This is why the topic has repeatedly pulled psychology into arguments about statistics, time, replication and the limits of experimental design.

First, separate precognition from presentiment

These words are often mixed together.

They should not be.

Precognition

In behavioral experiments, a participant makes a choice or performs a task before a future random event is determined.

Researchers then ask whether present behavior is associated with that future event above chance.

Presentiment

The participant may not consciously predict anything.

Instead, researchers measure physiology before a future stimulus appears.

Skin conductance.

Heart rate.

Pupil size.

EEG.

Other bodily signals.

The question is whether pre-stimulus activity differs depending on what kind of supposedly unpredictable stimulus will appear next.

One claim concerns behavior or cognition.

The other concerns physiology.

A result in one paradigm does not automatically validate the other.

The Bem experiments detonated the modern debate

In 2011, psychologist Daryl Bem published a paper in the Journal of Personality and Social Psychology reporting a series of experiments designed as time-reversed versions of familiar psychological effects.

One example involved retroactive facilitation of recall.

Ordinarily, practicing words helps you remember them later.

Bem reversed the order.

Participants completed a memory test.

Only afterward were some words randomly selected for additional practice.

The controversial question was whether words selected for future practice had already been remembered better before the practice occurred.

Other experiments used different paradigms but the same broad idea:

future events appeared to correlate with earlier responses.

The paper was peer reviewed.

The journal was mainstream.

The claim was extraordinary.

The replication race began almost immediately.

A failure to replicate is not a philosophical footnote

In 2012, Jeff Galak and colleagues reported seven experiments involving 3,289 participants using procedures based on Bem’s retroactive recall experiments.

They did not replicate the effect.

They also combined replication attempts in a meta-analysis and found an average effect that did not differ meaningfully from zero.

This mattered because the debate was no longer:

Bem versus people who refused to test Bem.

Researchers were running the procedure and getting different answers.

That is exactly where science becomes interesting.

Meanwhile, another future-facing literature was growing

Presentiment experiments do not necessarily ask a participant to guess the future.

A typical design may show a randomized sequence of neutral and emotionally arousing images.

The computer is intended to choose the next stimulus unpredictably.

Researchers measure physiology before the image appears.

If the nervous system shows a small but systematic difference before arousing versus neutral images, researchers ask whether the body is somehow anticipating the upcoming event.

In 2012, Julia Mossbridge, Patrizio Tressoldi and Jessica Utts published a meta-analysis of 26 reports.

They reported a small overall anticipatory physiological effect.

Importantly, they used the phrase unexplained anticipatory activity rather than claiming that a supernatural mechanism had been proven.

That wording is worth keeping.

It describes the observation without pretending to know the cause.

But a future-correlated signal has many possible enemies

Suppose physiology differs before two types of future stimulus.

That sounds astonishing.

Before invoking backward causation, the design has to survive several ordinary possibilities.

Expectation structure

Random sequences can still create local patterns.

Participants and physiology may respond to recent trial history.

Baseline choices

Different time windows, filtering decisions and normalization methods can alter tiny pre-stimulus signals.

Physiological autocorrelation

The body is not reset to zero after every trial.

Responses bleed through time.

Randomization

The future stimulus must be genuinely unpredictable relative to the measured data.

Multiple analyses

If many channels, time windows and transformations are examined, false-positive opportunities multiply.

Selective reporting

Experiments that find nothing may be less likely to appear in the literature.

None of these automatically explains every reported effect.

But every one of them has to be controlled before “the body senses the future” becomes the best explanation.

The modern solution is transparency before data

Psi research has become an unusually useful testing ground for open-science methods.

Why?

Because the claim is so controversial that ordinary trust is not enough.

A result needs to be protected from hindsight.

Researchers should not be able to change the main analysis after seeing the outcome.

The protocol, sample size, exclusions and statistical test should be fixed in advance.

Data should be visible.

Labs should be independent.

And ideally, supporters and critics should agree on the design before collection begins.

This is exactly what the Transparent Psi Project tried to do.

The Transparent Psi Project changed the argument

In 2023, Zoltan Kekecs and a large international team reported a multi-laboratory replication of Bem’s Experiment 1.

The study was co-designed by a consensus panel that included researchers with different views of the original claim.

The project used unusually strong transparency procedures:

• preregistration;
• direct data deposition;
• born-open data;
• laboratory logs;
• video-documented sessions;
• external research auditing.

The result did not replicate Bem’s original effect.

The observed success rate was essentially at chance.

That does not prove no form of precognition can exist.

It does show that one influential result did not survive a deliberately high-credibility replication.

Then the sample became much larger

A 2025 PLOS ONE project pushed the methodology further.

Across three studies, the researchers tested the precognitive effect with 26,483 participants and more than 420,000 critical trials.

Study 1 failed to replicate the predicted effect.

An exploratory analysis produced a small deviation in the opposite direction.

Study 2 reproduced that exploratory deviation.

Study 3 did not.

This sequence is almost a textbook lesson in why replication matters.

A result appears.

A follow-up confirms it.

A further follow-up removes the confidence.

If the story had ended after Study 2, the narrative would have looked very different.

This is why “statistically significant once” is not enough

A p-value is conditional on a model.

A reliable phenomenon has to survive changes in sample, lab, analyst and time.

This is especially important for effects that are:

• small;
• theoretically disruptive;
• sensitive to analysis choices;
• difficult to connect to a known mechanism.

The more extraordinary the interpretation, the less useful a single significant experiment becomes.

The bar is not “impossible to criticize.”

The bar is predictive reliability.

Does the absence of a mechanism invalidate the research?

No.

Science has sometimes observed effects before understanding their mechanisms.

Mechanism is not a prerequisite for looking.

But mechanism matters when choosing between explanations.

If one explanation requires information to relate to a future event in a way current models do not accommodate, while another explanation involves subtle bias in randomization or analysis, the extraordinary interpretation carries a heavier evidential burden.

That is not closed-mindedness.

It is model comparison.

What about spontaneous precognitive experiences?

People report dreams, intuitions and feelings that later seem predictive.

Those experiences can be psychologically powerful.

They are not the same thing as controlled evidence.

A person has thousands of thoughts, dreams and worries.

Most are forgotten.

The few that later resemble an event can become unforgettable.

Memory changes after outcomes are known.

Ambiguous predictions can be matched flexibly.

Events with high base rates will sometimes look prophetic.

This does not prove every spontaneous experience is meaningless.

It means spontaneous experience is a poor tool for estimating how often prediction succeeds versus fails.

Laboratory work exists partly to solve that problem.

The most careful current answer

Precognition and presentiment have both produced published anomalous findings.

Some meta-analyses report effects.

Important replications have failed.

Modern high-transparency projects have not produced a stable, independently replicable precognitive effect.

For presentiment, the literature remains debated and requires strong controls over randomization, physiology and analysis.

So the evidence status is not:

established future sensing.

It is:

a disputed research domain containing statistical claims that have not yet matured into a stable explanatory phenomenon.

That leaves room for curiosity.

It removes the permission to pretend the question is settled.

Continue exploring

Next: Why Psi Research Remains Controversial: Replication, Methods and Interpretation

The final Article steps away from any one experiment and asks a larger question:

Why can intelligent researchers look at overlapping data and still disagree so strongly about what the evidence means?