"Futile" discussions

We cannot find most solutions through discussions.

I have two criteria on whether a problem can be solved through discussions: 1. convergence test: do the conclusions converge by discussing with more people; 2. singleton test: whether the problem would still exist if I were the only person on earth.

This post will address the converge test. In machine learning, many algorithms look for the global optima, a point where errors are minimized. This is recognised as a solution to the problem. In science, we set up questions, experiments, and debates, hoping we are approaching truth in the process. Through these discussions, we either find one person’s theory closer to the possible truth; or, we seek a unifying theory that explains both observations. In both cases, we reach a local optimum that is closer to the global optimum.

However, many problems in life, like relationships, do not have optima. Your conclusion converges at different points depending on who you are talking to and they are equally valid. There is no point where everyone can agree on. If some topics fail the convergent test, you will find your understanding about this issue keep turning around in a circle and there is no final answer. Therefore, we must realise that when we discuss a non-convergent problem, we are not looking for a solution. We are asking for the other person’s experience with that problem, with a sample size of 1. It is an extreme case of Wittgenstein’s Ruler - we are seeking to understand the person much more than understanding the problem.

P.S. Though this post is about discussions, the same thinking can be applied to learning, if we consider learning as an asynchronous discussion between authors and us.


Personalizing medicine