On the Precautionary Principle

Summary. Decision principles that place a strong asymmetric burden of proof on showing either that something is or is not harmful can lead to extreme conclusions. Symmetric approaches based on expected values work better.

This essay critiques the basic notion behind the precautionary principle. I acknowledge that the precise definition of the principle may indeed consider the points that I make; if that is the case, then this criticism is not directed against technical application of the actual principle but, rather, against popularized and oversimplified conceptions of it.

In looking through Google's list of definitions for "precautionary principle," I found a few that state the idea in the way in which I have always heard it:

"Better safe than sorry" attitude. The idea that, in the face of uncertainty, society should assume that potential problems are real and address them accordingly. (source)

Assumption of the worst-case scenario with respect to actions whose outcomes are uncertain. (source)

In other words, when one does not know how harmful something will be, one assumes the worst until it is proven otherwise.

This idea may sound very nice initially, but consider a few cases that one might encounter in practice.

Example 1. Suppose that Monsanto has just developed two new pesticides, Pesticide A and Pesticide B. Each has so far been subjected to only one preliminary study, so their actual toxicities remain uncertain. Yet, comparing only these initial results, it appears that A is very likely to be carcinogenic, while B shows no such signs. Following the simple precautionary approach given above, we must assume that both chemicals are seriously carcinogenic, because that is the worst-case scenario and their actual deleteriousness is not yet certain. But this would mean that regulatory agencies ought to spend just as many resources controlling and limiting the use of B as they do A--which is clearly not what actually ought to be done.^[1]

Example 2. In 1985, EPA characterized the chemical dioxin as "the most potent carcinogen ever tested in laboratory animals" (source). One source of dioxin is the bleaching of paper with chlorine. It is true that confirmedly safe alternatives for bleaching paper have been discovered, but imagine that they had not. Suppose someone develops a new chemical that proves as effective and inexpensive in bleaching paper as chlorine. This chemical, too, creates bleaching byproducts, but they are entirely different from dioxins. Before any studies have been performed, the precautionary principle requires assuming the worst possible toxicity--perhaps as bad as the toxicity of dioxin. But in all probability, the new chemical byproducts will not be this baleful. If there is no cost for paper mills to replace chlorine with the new bleaching agent, then they clearly ought to do so, even if it will take years for the new chemical to be studied extensively. Yet the precautionary principle as outlined above rejects this action--or at least makes no recommendation about it.^[2]

As these examples show, the problem with the simplified precautionary principle lies in its inordinately strong presumption of harmfulness, even when such harmfulness is not likely. Certainly we should not wait until adverse consequences are proven to take action--to that extent, the precautionary principle is right. But it can be just as shortsighted to assume great harm as it is to assume none at all.

The best approach, therefore, is to calculate an approximate expected value of deleteriousness (see Why Maximize Expected Value?). Granted, coming up with probability values is an inexact and somewhat arbitrary process (indeed, the frequentist school of probability rejects assignment of such values even in principle), but what is the alternative? It is even more arbitrary to assume one particular result, as the simplified precautionary principle does. It is ultimately more helpful to say that A has a 0.9 chance of being carcinogenic while B has a 0.4 chance than to assume that they both have a 1.0 chance until proven otherwise.

[1] I'm ignoring other (more important) considerations about pesticide use, like the effect on insects themselves. It may be that pesticides are net beneficial to insect welfare if they reduce wild insect populations over the long term?

[2] Once again, these comments concern themselves only with impacts on human health. However, it's possible that dioxin reduces wild-animal suffering if its inhibitory effects on reproduction outweigh the acute injury that exposure causes to marine organisms. This topic deserves further study.