Precognition and Retrocausality: The Physics Allows It

In 2011, Daryl Bem published a paper in the Journal of Personality and Social Psychology that landed like a grenade in two separate rooms simultaneously. Psychologists were upset because it seemed to prove something impossible. Physicists were less upset because they knew it might not be. The paper was called “Feeling the Future,” and it reported nine experiments across 1,000 participants showing that humans could respond to stimuli before those stimuli were presented. Not metaphorically. Literally. Reaction times to emotional images were faster when the image was about to appear than when it wasn’t, measured before the computer had selected which image to show.

The paper cleared peer review at one of the field’s most rigorous journals. Bem is not a crank: he’s a Cornell social psychologist with a conventional career and a record of careful methodology. The experiments weren’t complex or exotic. They were standard cognitive psych tasks run backward, with the dependent variable preceding the independent variable. The effect sizes were small but consistent. The statistics were sound enough to get through review. And then, predictably, the field went to war.

The Dismissal That Wasn’t Actually a Dismissal

The response from mainstream psychology followed a predictable script. The results must be an artifact of p-hacking, file drawer effects, or some subtle procedural flaw. Several replication attempts failed. Several succeeded. A meta-analysis by Bem himself in 2016 pulled together 90 studies from 33 laboratories across 14 countries and found a mean effect size of 0.09, statistically significant at p = 6.4 x 10^-17. That’s not the kind of number you get from noise. That’s a signal that keeps showing up across independent research groups, different paradigms, and different subject populations, and it keeps being roughly the same size.

The failed replications deserve honest accounting too. Some prominent attempts did fail to find the effect. The literature is mixed. But “mixed literature” is not the same as “no effect.” Mixed literature is what you get when you’re studying something real but small, context-dependent, or difficult to induce reliably. The Bem results have not been definitively debunked. They have been contested. That’s a different thing, and conflating the two is the move that lets people feel they’ve settled a question they’ve actually just dismissed.

What’s more interesting than the replication debate is what happened when skeptics proposed a pre-registered adversarial collaboration with Bem. He agreed. The results of that collaboration, published in 2015, were inconclusive. Neither side could claim vindication. Which is, again, not debunking. Inconclusive data from a single adversarial study doesn’t cancel a decade of replications any more than a decade of replications proves precognition definitively. We’re sitting in genuine uncertainty, and the right response to genuine uncertainty is not to pick the explanation that makes you more comfortable.

Why Physicists Aren’t As Bothered As You’d Expect

Here’s where it gets genuinely interesting. The reason the Bem paper disturbed psychologists but not most physicists is that physicists already know time isn’t what it looks like from inside it. The equations of fundamental physics are time-symmetric. The laws of motion at the quantum scale don’t prefer one direction of time over the other. The arrow of time that we experience is a thermodynamic phenomenon, an emergent property of entropy gradients, not a bedrock feature of reality. This is not fringe physics. This is the standard picture.

From that picture, several serious physicists have developed frameworks in which causal influence can run from the future toward the past. Huw Price at Cambridge has spent decades arguing that retrocausality is the most natural reading of quantum mechanics if you drop the assumption that causation must be forward-directed. His argument isn’t that we’ve observed backward causation; it’s that our refusal to consider it is a philosophical prejudice we’ve smuggled into our interpretation of the data. We assume causation runs forward because it always seems to in our experience, and then we interpret quantum mechanics through that assumption, and then we’re surprised when quantum mechanics is weird. Price thinks we’ve caused some of our own confusion.

The two-state vector formalism (TSVF), developed by Yakir Aharonov and colleagues over several decades, provides a concrete mathematical framework in which quantum systems are described by two wavefunctions: one propagating forward in time from the initial state and one propagating backward in time from the final state. The present state of a system is determined by both. This is not a speculative hypothesis; it’s a fully worked-out formalism that reproduces all the predictions of standard quantum mechanics while making additional predictions about weak measurements that have been experimentally confirmed. The TSVF is not the interpretation everyone uses, but it’s a legitimate, mathematically rigorous framework that serious physicists publish serious papers about.

There’s also the transactional interpretation of quantum mechanics (TIQM), developed by John Cramer, in which quantum transactions are completed by a “handshake” between a forward-moving offer wave and a backward-moving confirmation wave. Again: not everyone’s preferred interpretation, but not crackpottery either. It’s published in peer-reviewed physics journals by a physicist at a major research university.

What all of these frameworks have in common is that they take seriously the possibility that causal influence isn’t strictly unidirectional. None of them predict robust macroscopic precognition of the kind Bem was testing. But none of them forbid it either. And the distinction between “not predicted” and “forbidden” matters enormously when you’re evaluating a body of empirical evidence.

What the Specific Experiments Actually Did

It’s worth being concrete about what Bem tested, because “humans responded to stimuli before those stimuli were presented” is easy to wave off as vague. It isn’t vague. One experiment type, “retroactive facilitation of recall,” showed participants a list of words, asked them to recall as many as possible, then showed them a random subset to practice. The practice session came after the recall test. The finding: people recalled practice words at higher rates than non-practice words, even though the practice hadn’t happened yet at the time of recall. The effect was small, around 2-3 percentage points, but it was consistent across multiple runs. Another experiment type presented erotic images to participants after measuring their arousal responses, and found that physiological arousal had already spiked slightly before the image appeared, but only for erotic images, not neutral ones. Again: small, specific, and not random.

The specificity matters. A generic effect that showed up equally for all stimuli would be easier to attribute to noise. The fact that emotional salience modulated the effect suggests a mechanism with structure. You can argue about what that mechanism is. You cannot easily argue that a structured, stimuli-dependent, size-consistent effect across nine different paradigms is pure artifact without specifying exactly which artifact produced which structured pattern and why it would replicate across independent labs.

There’s also the replication crisis context that the dismissal never quite acknowledges. Psychology’s replication crisis, which became public knowledge around the same time Bem’s paper was landing, revealed that a substantial fraction of the field’s most celebrated results don’t hold up under independent replication. The effect sizes for many canonical social psychology findings (ego depletion, power poses, the Stanford prison experiment’s generalizations) shrank dramatically or disappeared entirely when subjected to serious replication efforts. Against that backdrop, a Bem effect that maintains across 90 studies at a consistent effect size looks more durable, not less, than the mainstream results that survived unquestioned for decades. This is not a vindication of precognition. It’s a calibration problem: the evidentiary standards applied to Bem’s results were sharper than those applied to results that fit everyone’s priors, and that asymmetry should make you skeptical of the skepticism.

The Steel Man for Precognition

The strong version of the precognition case, assembled carefully, looks like this. The effect Bem found is small but persistent. It shows up across labs, paradigms, and populations. The physics of time doesn’t categorically rule out backward causal influence. Several working physical frameworks explicitly incorporate it. The Bem experiments used arousal (erotic or threatening images) as the stimulus, which is interesting: if any physical system in the brain were sensitive to future states, you’d expect it to be most sensitive to emotionally significant ones, since emotional arousal involves cascading physiological changes that might produce a larger signal. The effect was consistently larger for arousing stimuli than neutral ones. That’s the kind of specificity that’s hard to get from random noise.

The steel man doesn’t require you to believe in ESP or psychic hotlines. It only requires you to entertain that consciousness might be a physical process with properties we don’t fully understand, that time isn’t the one-way street our intuitions insist on, and that a persistent small effect across a large literature might be pointing at something real rather than being a massive coordinated methodological failure across independent research groups worldwide.

What the Evidence Actually Shows

The honest accounting: Bem’s effect is real in the sense that it replicates at above-chance rates across the literature. It is not real in the sense of being robustly reliable in any individual study. The signal is weak. The mechanism is unknown. The interpretive leap from “small anomalous effect in cognitive psych tasks” to “humans can perceive the future” is large, and nothing in the data forces it.

What the evidence does show is that we’re dealing with a phenomenon that doesn’t go away cleanly when you look at it carefully. That’s not nothing. Most phenomena that are pure artifacts disappear or shrink dramatically under scrutiny. The Bem effect shrinks under scrutiny but doesn’t disappear. The 2016 meta-analysis effect size (0.09) is smaller than Bem’s original findings, but it’s still there, and it’s still statistically robust.

The physics argument is separate and in some ways more important. You don’t need the Bem data to make the retrocausality case. The physics makes it independently. Retrocausality is not exotic; it’s a natural consequence of taking time-symmetric equations seriously. The reason it seems exotic is that we filter physics through our experience of time, and our experience is shaped by thermodynamics rather than fundamental law. That filter is useful for navigating the world. It is not necessarily a reliable guide to what the world allows at the level of quantum systems.

The combination, though: that’s what makes this episode genuinely interesting. You have a persistent empirical anomaly in the cognitive science literature and a set of physical frameworks that don’t rule out the mechanism that would explain it. Neither piece of evidence is conclusive alone. Together they constitute a case for continued investigation that is much stronger than the mainstream dismissal acknowledges.

The Verdict: Gold

This doesn’t prove precognition. The verdict isn’t “precognition is real.” The verdict is that the dismissal is premature, and premature dismissal of this particular question is a failure of epistemology rather than a triumph of skepticism.

Sophisticated skeptics like to say “extraordinary claims require extraordinary evidence.” Fine. But the claim that retrocausality is impossible is also an extraordinary claim, and the evidence against it (from fundamental physics) is weak. When you put a persistent empirical anomaly next to a set of theoretical frameworks that accommodate it, you don’t get proof of the anomaly’s mechanism. You get a research program that deserves to be taken seriously.

The Bem experiments are the most replicated anomalous result in the history of parapsychology. The physics of time is genuinely ambiguous on the directionality of causation at the quantum scale. Several serious physicists have built working formalisms around retrocausality. The effect is small, the mechanism unknown, the interpretation contested. That’s exactly where you’d be if you were at the early stages of discovering something real and difficult.

What breaks most people’s priors isn’t that the Bem data is irrefutable. It’s that the physics doesn’t give you an easy out. You can’t say “that’s impossible because physics says so,” because physics doesn’t say so. You’re left having to do the hard epistemological work of sitting with uncertainty, which is a lot less comfortable than having a clean debunking. Reality, as usual, is not organized for your comfort.