Abstracting out the essence of this article: It's arguing about the danger of ubiquitous "big data" in certain contexts.
The human population around 8 billion, which is 2^33. 33 is roughly the number of pairs out of 9 objects. Heuristically, this would mean that if we test for more than 9 boolean variables (high/low), purely by randomness we will start seeing correlations due to a "limited" population size of 8 billion. If we start treating people just based on those observations, we would seriously wreck human health.
Diagnosis (decision making) based on correlations (rather than causality) is very tricky.
For the record, we _are_ treating people based on exact same observations albeit correlated on much less number of people. However, we make use of some probability, other observational and subjective evidence in order to determine right course of action. I agree with another comment made above regarding "less data is never the answer". We may make some mistakes with small amount of data but the only way we will ever learn is by making a hypotheses and acting on it. We should tread carefully but not too careful to never step forward.
The human population around 8 billion, which is 2^33. 33 is roughly the number of pairs out of 9 objects. Heuristically, this would mean that if we test for more than 9 boolean variables (high/low), purely by randomness we will start seeing correlations due to a "limited" population size of 8 billion. If we start treating people just based on those observations, we would seriously wreck human health.
Diagnosis (decision making) based on correlations (rather than causality) is very tricky.