“I See Dead People” – Machine Learning and Post Mortems.

Machines help us learn. Machines can avoid failure, assist failures and reverse failures. One component of the “learning revolution” is the ability to employ tivo-like capabilities on events. A common end process in many organizations is the post mortem (Autopsy).

Google famously studies post mortems to grow and evolve their approaches. However, what can we learn from looking at these past events? When we see a sequence of events can we extract value from the decisions made and templatize the past towards a new truth?

Recently, there was a marvelous Medium Post indicating that “Interation is not design.”

View story at Medium.com

In addition to this post, there was a fantastic podcast from Software Engineering Daily where Jason Knight correctly points out the importance of training deep learning models with samples that fall outside the normal patterns of data. These samples help the emergent system to grow, adapt and succeed (i.e. don’t waste time).

Biological Machine Learning with Jason Knight

We also have the classic story of penicillin (https://en.wikipedia.org/wiki/Penicillin#Discovery) where “Fleming noticed a Petri dish containing Staphylococcus that had been mistakenly left open was contaminated by blue-green mould from an open window, which formed a visible growth. There was a halo of inhibited bacterial growth around the mold. Fleming concluded that the mold released a substance that repressed the growth and caused lysing of the bacteria.”

Now, how does that relate to post mortems? Post mortems help us avoid the “Tinkerbell” effect or (https://en.wikipedia.org/wiki/Tinkerbell_effect) tendency for our brain’s imagination to fill in the missing elements. In the world of DevOps and CI/CD, post mortems can provide a very valuable tool to adjust pipelines and process and they can reveal patterns or anti-patterns which hurt the business. Machine learning algorithms can assist businesses with this detection process provided they are given the proper datasets and the “open windows” . Open windows come from looking outside departments and businesses for examples of leadership patterns.

Lastly, the discovery of anti-patterns within post mortems are not to be underestimated. Any algorithm which develops from a positive anti-pattern cannot be minimized. As the philosopher, George Santayana, famously put “Fanaticism consists of redoubling your effort when you have forgotten your aim.”

@benmorris