You’ll certainly not get an argument from me on that Kamil :) Plus many scientific functions tend to be applied to stuff that is regarded as naturally “untestable”. Not because the code is poor, but because the code can often be non-deterministic and/or non-Markovian. This creates what folk in mainstream development consider side-effects, even though this is the key mechanism of operation of areas such as Machine Learning, Bayesian Inference and Genetic Algorithms.
That said, I of course, see where they are coming from in both linear and deterministic computing contexts. I use TDD and BDD almost all the time. The issue is one of education I’d argue, especially since the classical trade-off or abstraction of mathematical literacy for coding productivity gathers apace. I am in support of it and computing’s open nature, but at the same time, I do despair when experiencing behaviour which looks down upon mathematical and scientific literacy, especially since it is everywhere!