Taipei Is One of the Safest Cities on Earth. My Dreams Didn’t Get the Memo.

A while back, for reasons that don't matter much here, I ended up living in Taipei for a stretch. Taipei consistently ranks as one of the safest cities in the world by most global indexes, which on paper meant my day-to-day risk of encountering real danger had dropped about as low as it gets. And yet, during exactly that period, I kept having the same dream over and over: being hunted down by someone. Which made no sense on the surface, since I've never actually been chased by anyone in my life — there was no real incident this could plausibly be a memory of. This is the kind of gap that evolutionary nightmare research exists to explain.

The Assumption That Every Nightmare Needs a Real-Life Villain

The default explanation most people reach for is that a nightmare must be mirroring some specific anxiety or trauma from waking life — as if every bad dream has to correspond to one identifiable bad thing that actually happened. But a lot of the time, what we're really doing is dreaming about danger that never occurred at all. That could still be some abstract anxiety wearing a costume, sure. But I've come to think there's a cleaner explanation for why this particular pattern shows up regardless of how safe your actual life is: threat simulation theory.

A Theory With a Name, Not Just a Vibe

Finnish cognitive scientist Antti Revonsuo proposed the threat simulation theory in 2000, arguing that the evolutionary function of dreaming — nightmares especially — is for the brain to rehearse encountering danger and practicing escape or defense, all from the safety of sleep. You don't need to have lived through the actual threat yourself. The brain already comes pre-loaded with a rehearsal script for being pursued by something dangerous, because that's exactly the kind of threat our ancestors faced constantly and repeatedly, long before any of us were born to inherit the wiring. A 2005 review of the theory backs this up directly, describing dream threat simulation as rehearsing the cognitive mechanisms needed for efficient threat perception and avoidance.

Why This Beats the Alternatives, For Me

I find this a lot more convincing than dreams being disguised wishes, or just random neural firing with no function at all, mostly because it actually explains why being chased, falling, and drowning show up as dream themes across so many cultures and generations without needing a personal trauma behind each instance. It also explains why the script keeps updating rather than staying frozen. The threat used to be a predator in a literal jungle; now it's an unfamiliar stranger with bad intent in a concrete one. None of that is really about individual psychological damage — it's more like a rehearsal drill baked into the species' memory, and the drill just gets re-skinned depending on what danger modern humans are actually more likely to run into.

Where the Theory Gets Tested Against Real Danger

Revonsuo's theory eventually got checked against people actually living under real threat, not just safe city-dwellers having odd dreams. Researchers found that children growing up in war zones and other genuinely dangerous conditions had nightmares containing more frequent and more intense danger content than children living somewhere calmer — which lines up with a comparative study on exactly this question. That's a reasonably strong signal that this isn't just a nice-sounding idea — the rehearsal function seems to actually respond to real environmental danger, scaling up or down with it.

The Rehearsal Doesn’t Care Whether the Threat Ever Shows Up

Which brings me back to Taipei, and to a dream that, statistically, had almost nothing to draw on. The adaptive function of a bad dream doesn't require the danger to have ever been real for you personally — the rehearsal script runs regardless, on a schedule dictated by something much older than your own biography. Living somewhere with a near-zero crime rate didn't turn off the simulation. It just meant the simulation had nothing local to attach itself to, so it ran the generic version instead.