Comment by j7ake

If you generate a process using a Cauchy distribution, you can observe finite error bars, but actually the variance of the generating process is infinite.

No matter how hard you model you won’t be able to predict processes that are fundamentally unpredictable. And you would get fooled because you only observe finite amount of data.

It is surprising how complicated the math gets even if you try to model very simple processes (eg think of the n-body problem and how complexity increases with every addition of a body). It is not a given that complicated models mean you’re modelling a complicated process.