The Precautionary Principle: When Expected Value Doesn't Apply

The precautionary principle is often dismissed as intellectually weak — the position of people who don't understand risk, who are so afraid of negative outcomes that they want to prevent everything new. "Don't try anything until it's proven perfectly safe."

Nassim Taleb's formulation is different. It's technically grounded in the ergodicity argument and applies specifically to a class of risks where expected-value reasoning doesn't just fail to be optimal — it's the wrong framework entirely.

The Technical Argument

The precautionary principle applies when two conditions hold simultaneously:

1. The failure mode is irreversible (absorbing). The system cannot recover. Once the bad event occurs, there is no return to the prior state, no opportunity to update and try again.

2. The outcome is systemic rather than local. The failure doesn't affect a bounded population. It affects the entire system — or a sufficiently large portion of it that recovery is impossible.

When both conditions hold, standard expected-value analysis produces the wrong answer because it treats the negative outcome as one data point in a distribution rather than as an absorbing state.

Expected value says: if there's a 1% chance of harm and a 99% chance of benefit, and the expected benefit outweighs the expected harm, proceed.

Precautionary principle says: if the harm is absorbing (permanent, systemic, irreversible), there is no probability threshold below which it's safe to proceed. The 1% chance of permanent systemic collapse is not commensurable with the 99% chance of some benefit.

What Makes a Risk Precautionary

Nuclear accidents. Chernobyl and Fukushima produced contamination of agricultural and inhabited land that will persist for decades to centuries. The area within the Chernobyl exclusion zone is still uninhabited 40 years later. This is absorbing on a human timescale. No amount of expected benefit from nuclear power eliminates this risk class — it can only be managed.

Pandemic release of engineered pathogens. A pathogen released into the global population is, in a worst case, an absorbing event. You cannot recall the pathogen. The feedback that would allow updating comes too late. The irreversibility is physical and the scale is systemic.

Global ecosystem disruptions. The interaction between complex ecosystems and new interventions (GMOs at global scale, novel chemicals, atmospheric changes) operates on non-linear dynamics that are not well-modeled. The failure modes include absorbing states — collapse of pollinator populations, disruption of ocean chemistry, permanent loss of soil viability at scale. These are recoverable over geological timescales; they are not recoverable on human timescales.

Novel financial instruments with systemic interconnections. 2008 was a near-absorbing event for the global financial system. The failure modes of highly interconnected, highly leveraged financial instruments operating at global scale are absorbing in a specific sense: the positive feedback loops of cascade failures can exceed the capacity of any institutional response.

What Doesn't Trigger the Precautionary Principle

The precautionary principle applies to absorbing, systemic risks. It does not apply to recoverable, localized risks — and applying it to those is indeed intellectually confused.

Starting a business carries risk. The failure mode is local (affects the founder and investors) and recoverable (you can start another business, update the approach, try again). Expected-value reasoning applies.

Medical interventions for serious illness: the risk is often local to the patient, and many adverse outcomes, while serious, are recoverable or at least survivable. Expected-value analysis is appropriate. The precautionary principle doesn't apply to individual medical decisions.

Novel technologies with localized failure modes: a software system that fails has recoverable failure modes (restore from backup, rebuild, try again). Expected-value analysis applies.

The error that anti-technology activists make: they apply the precautionary principle to recoverable, localized risks where expected-value reasoning is appropriate. The error that pro-technology advocates make: they apply expected-value reasoning to absorbing, systemic risks where the precautionary principle is appropriate.

The Complementary Risk

Taleb notes a crucial symmetry: there are also risks you cannot afford not to take.

Paralysis — refusing all risk because some risks are absorbing — is itself a form of ruin. Stagnation carries costs. Missed medical interventions kill people. Avoided technological development leaves problems unaddressed.

The goal is not to apply the precautionary principle universally. The goal is to correctly classify risks:

Recoverable + local = expected value reasoning applies. Optimize within the risk. Accept reasonable bets. Update when wrong.

Absorbing + systemic = precautionary principle applies. Don't play the game at all, or play it in controlled, isolated conditions that prevent systemic effects.

The intellectual work is in the classification, not in the application of either rule. Both rules are simple. Knowing which applies is hard.

The Skin in the Game Connection

The precautionary principle is strongest when the people making decisions about absorbing risks are not the ones who will bear the absorbing consequences.

Nuclear power plants are sited in communities that don't make the siting decision. GMO crops are introduced into ecosystems whose complexity is not well-understood by the people running the trials. Engineered pathogens are created in labs that won't bear the consequences of an accidental release.

When the decision-makers have no skin in the game — when they capture the upside but the downside is absorbed by others or by future generations — they will systematically underweight absorbing risks. The precautionary principle and skin in the game reinforce each other: the person who bears the absorbing risk should have a voice in decisions about whether to accept it.

For the full framework, read Ergodicity, Ruin, and Rational Risk-Taking.