The Unobserved Is Not the Unobservable: Taleb's Epistemological Warning

"It is a very recent disease to mistake the unobserved for the nonexistent; but some are plagued with the worse disease of mistaking the unobserved for the unobservable."

This two-part aphorism from The Bed of Procrustes contains, in two sentences, the epistemological foundation of Taleb's entire project. Understanding the distinction — and why both errors are errors — is a prerequisite for thinking correctly about risk.

Two Different Errors

Error 1: Mistaking the unobserved for the nonexistent.

You haven't seen something. You conclude it doesn't exist. This is the simple version of the error — the assumption that your sample of observation is comprehensive, that the absence of evidence in what you've seen is evidence of absence everywhere.

The practical consequence: risks that haven't been observed in your historical record are treated as not-real-risks. The financial model that uses historical volatility as a proxy for future risk. The medical study that doesn't observe a side effect in 500 patients and concludes the side effect doesn't exist. The city that hasn't experienced a flood in 50 years and concludes the flood risk is manageable.

Each of these has produced catastrophes when the unobserved materialized.

Error 2: Mistaking the unobserved for the unobservable.

This is Taleb's "worse disease" — not merely assuming the unseen doesn't exist, but assuming it cannot exist. Treating one's current observational framework as exhaustive.

The difference matters enormously. The first error is epistemic modesty failure — you forgot to account for what you haven't seen. The second error is epistemic arrogance — you've concluded that what you haven't seen is categorically beyond the possible.

The doctor who says "we've never seen that interaction in our patient population, so it's unlikely to occur" has made Error 1. The doctor who says "that interaction is not possible given the mechanism we understand" has made Error 2. The first is careless; the second has closed the category of the possible to what the current theory can account for.

Why This Is "A Very Recent Disease"

Taleb's observation that this is recent is not accidental. Pre-modern epistemic traditions actively maintained a category for the unknown and unknowable — what lay on the other side of the veil of opacity. Religious traditions, classical philosophy, and indigenous knowledge systems all preserved an acknowledgment that more lay beyond human observation than within it.

The Enlightenment project — rigorous, immensely productive, genuinely transformative — had an epistemic side effect: it naturalized the assumption that what we haven't yet observed can be brought within the realm of the observable, given sufficient method. This is often true. But it slid into the stronger assumption: that what we haven't observed is either nonexistent or will soon be accounted for.

Modern empiricism, especially in its naïve form, treats absence-of-evidence as evidence-of-absence. This is a logical error. "We have no evidence of X" is not evidence that X doesn't exist — it is evidence that X hasn't been found in the places we've looked. Whether X is absent or merely unobserved depends on how comprehensive the search was, and the search is almost never comprehensive in complex domains.

The Black Swan Connection

This is the epistemological root of Black Swan blindness. A Black Swan is, by definition, an event outside your observational history. If your model is built on the assumption that your observational history is comprehensive — if you have mistaken the unobserved for the nonexistent — your model assigns zero probability to the Black Swan.

Zero probability is not "very low probability." Zero probability means "cannot happen." The model that assigns zero probability to a Black Swan will not hedge against it, will not build robustness to it, will not price it. When it happens, the model is not merely surprised; it is catastrophically wrong in a way that the model's architecture made inevitable.

The 2008 financial crisis is the canonical case. The models used by major financial institutions had been calibrated on historical data from a relatively benign period. That period contained no event like the one that occurred. The models therefore assigned it something close to zero probability. The events of 2008 were not improbable on a longer historical view — they were improbable on the view of the recent data that had been treated as comprehensive.

The Remedy: Keep the Unknown Open

"Something finite but with unknown upper bounds is epistemically equivalent to something infinite. This is epistemic infinity."

The practical remedy is to keep an active category for the unobserved — to model as if there is more than you can see, because there is. This doesn't mean treating every unknown as equally likely. It means never collapsing the unknown into zero.

In portfolio construction: maintain exposure to positive tail events; don't assume the absence of negative tail events in recent history means they won't occur.

In medicine: don't treat the absence of observed side effects as evidence of their absence; treat it as absence of evidence, which is different.

In policy: don't treat the absence of a specific failure mode in your historical record as confirmation that the failure mode doesn't exist; treat it as confirmation that you haven't seen it yet.

The discipline is uncomfortable because it means holding positions open that the brain wants to close. The brain wants to know. The unobserved is uncomfortable. But the discipline of keeping it open is what the Procrustean bed makes impossible — and what the nonsucker has learned to maintain.

For the full framework, read The Bed of Procrustes Explained.