Liberty, Equality, and Fraternity, After All?
It’s too early to say.
In 1972, when Chinese Prime Minister Zhou Enlai was asked about the impact of the French Revolution in 1789, he replied, “Too early to say.” The fact that he misunderstood the question does not make the answer less ingenious. This is especially true when we dance with path-dependent processes.
Ouch! We have a bug in our digital payment function. Customers get stuck at credit card verification. This issue might have happened because Robin, one of our developers, accidentally caused a dozen other defects this year alone. Perhaps the reason is that we switched to a new cloud vendor last month for the third time in two years. Hey, didn’t Alex, our product owner, also mention a big refactoring last Friday?
Abductive reasoning occurs when we notice the outcome (bug in the payment function) and investigate our repertoire of patterns (Robin, cloud vendor, Alex) that may have led to that outcome. To identify the reason for this outcome, we must act as detectives. However, don’t expect a single event to be the root cause. All suspected patterns are candidates, often never-seen-before combinations. That is why the most effective way is to decompose and investigate them all — in parallel. The lawyer and the historian are masters of abductive reasoning. Their profession requires them to respect the uniqueness of every event and individual.
This is very different from inductive reasoning, in which we draw general conclusions from the collected data on the relationship between cause and effect. For example, in a planning meeting, our Kanban team committed to deliver ten prioritized tasks. Historical data showed us — at a 95 percent confidence level — that one task demands between one and three days to complete. It is likely that we have completed all ten tasks in ten to thirty days. Do we dare to promise the stakeholders to deliver within fifteen to twenty-five days?
Induction is a tool used by actuaries when they determine future insurance premiums, a problem in which there are variations within limits or at least long-lasting trends exist. Variation often depends on the current state, such as the length of a queue. We may refer to processes that depend on their current state, but can’t recall prior states — Markovian. Regrettably, reality is usually not very eager to behave in this nice way, especially when there are many unknown unknowns.
We face the risk of confusing cause with effect: B causes A, not the other way around. Or is it C that causes both A and B? In addition, all swans are white until someone spots a black swan. That last fallacy is especially common when we have only a small sample. For example, a pilot experiment showed that one team found ten-week iterations a great idea. The overgeneralized rule then became that all teams in our enterprise must stick to ten-week iterations.
After all, we would achieve the best predictability if we had a general rule that was always true. Based on that rule, we can calculate (deduce) in advance what effect we would get from a more specific action. We call this approach deductive reasoning. Here is an example: The 𝚜𝚞𝚍𝚘 𝚛𝚖 -𝚏𝚛 <𝚏𝚘𝚕𝚍𝚎𝚛> command not only deletes all French files in <𝚏𝚘𝚕𝚍𝚎𝚛>, but it also deletes all other files and folders there! From this rule, we can easily deduce that the more specific 𝚜𝚞𝚍𝚘 𝚛𝚖 -𝚏𝚛 ~ deletes all files in our home directory. Deductive reasoning is sometimes handy when we draw conclusions from deterministic models like timetables, economic order quantity models, and accounting.
However, the weakest part of deductive reasoning is, of course, that the general rule must be universally true, which rarely happens when humans are involved. Wouldn’t it be terrific if all processes in the enterprise were like coin flipping: a known probability for each possible outcome and no current state to monitor? A fair coin gives us an equal chance of heads and tails — eternally. We call this phenomenon Bernoulli processes.
Sadly, probabilities tend to change. History doesn’t repeat itself as often as we would like — neither inside the office nor outside where buyers meet sellers. The world seems to return to similar situations, but on closer inspection, we inevitably learn that the old and the new differ greatly. One reason is path-dependency.
No matter if events occurred recently or 200 years ago–like the French Revolution and liberté, égalité, fraternité — they still affect us, and after they happen only once, nothing is the same. Processes irrevocably bound to events in the past are sometimes called path-dependent. When we deal with these processes, abductive reasoning may come to the rescue.
First, we must accept that the outcome we are experiencing is a unique situation. Then, we must decompose well-known patterns, which reasonably could lead to this outcome. Finally, we conduct small parallel — possibly mutually contradictory — experiments to learn hopefully which of these decomposed ideas may have influenced the perceived outcome.
- Bernoulli and Markov processes are independent of history; the latter is affected by a current state, though. Yet, when all history matters, we deal with a path-dependent process. It will most certainly never return to an old state.
- We may deduce the output from a general rule and a specific input. And a general rule can be induced when we have several observations. However, when humans are involved, most processes are path dependent. This is why we should start with an abductive approach.
Staffan Nöteberg is the author of The Pomodoro Technique Illustrated, published by The Pragmatic Bookshelf.
To save 35 percent on the ebook version of The Pomodoro Technique Illustrated, enter promo code innovation_2022 when you check out at The Pragmatic Bookshelf. The promo code is valid through October 1, 2022, and is not valid on prior purchases. You can also read The Pomodoro Technique Illustrated on Medium.