11.10.20

Hypotheses are bad?

Posted in Hacking, programmer productivity at 11:55 am by ducky

As mentioned here and here, there is evidence that generating lots of hypotheses helps find the right answer faster.

Hypotheses are a Liability shows that having a hypothesis can make you less able to discover interesting facts about a data set. Very interesting!

2 Comments

  1. Jeff Powell said,

    November 10, 2020 at 6:07 pm

    I’ve seen this – or something very much like it – while debugging. If I (or others) have a hypothesis about what causes a bug in a software system before we start actually working on finding the problem, we can sometimes be mislead and take longer to fix the bug than if we’d let the erroneous behaviour lead us from the start. I suspect we design tests to prove what we hypothesize, rather than to elicit more detail on what is actually going wrong.

    If it happens that the hypothesized problem is actually the bug in the code/system, it will help and the process will be sped up. But if the problem lies elsewhere it may take substantially longer to find and fix it.

  2. David Pablo Cohn said,

    November 11, 2020 at 8:46 am

    Oh man (or rather, Oh Gorilla), I love that data experiment!