Statistics doesn't work so well on things with unknown unknowns. Actuarial science works super well on predicting life expectancy of an average person, because

Statistics doesn't work so well on things with unknown unknowns. Actuarial science works super well on predicting life expectancy of an average person, because we have a lot of data on when people die. We might not know when a particular person dies, but over a large number of persons the numbers are known and computable. But we don't know how toxic a substance is, and we really have no idea, and the result hinges on the toxicity of a single substance. In theory we could reduce chances of adverse outcomes by using a mix of unproven chemicals and hope that the bad ones are dilute enough not to cause problems, but I think in practice it would just lead to the worst of all posibilities...