[originally at jlottosen.wordpress.com, Aug 2021 ]
Recently I was in a network call on the use of automation and machine learning in detection of skin issues . Similarly I was reading about automation in the legal space. Both these stories align with the struggles we see in the discussions around how much we can automate. We can model it on this simple continuum between hunches and hard truths:
- Hunches: All the implicit ideas, smells or feelings
- Quantity: Things and observations we can spot and count
- Qualify: Items we can observe and measure
- Hard truths: All the explicits and facts
Notice the alignment to the Evolutionary Characteristics Cheat Sheet by Simon Wardley as also discussed in Darlings, Pets, Cattle and GUID’s. The continuum between weak signals and commodity solutions.
While we can automate obvious signs of skin issues, automate the authoring of legal briefs and confirm that systems uphold specific highly testable requirements. We cannot automate exploratory testing based on hunches and edge cases. Yes we might be able to provide tool support for some of the quantifiable things – but not automate everything across the board. Especially not ambiguity.
Any prediction of future behaviour based on past patterns is at best a statistical probability.
The Impossibility of Automating Ambiguity, Artificial Life (2021) 27 (1): 44–61.
As with Cynefin – consider, what is your problem space but also consider the goal of our activities