British philosopher Bertrand Russell once said, “There are two kinds of reasoning: deductive and bad.” He was referring in large part to the fallacy of induction. A well-known example is that a turkey gets fed every day, thinking that life is good. All of a sudden, the day before Thanksgiving ... whap! The poor bird finds out otherwise. Not enough data, you might claim. Well, what about swans? Millions of datapoints collected across centuries of observations led to the conclusion that, surely, all swans were white. But then explorers observed black swans in the newly discovered, at least to Western scientists, continent of Australia.
This matters more than ever in today’s complex, data-dependent world. Even with the rapid proliferation of AI/machine learning (ML) generating knowledge from extremely large data-sets, we still remain susceptible to the fallacy of induction. Despite huge volumes of data, banks, businesses, and bridges still collapse. People are still plagued by side effects from foods and drugs presumed to be safe.
That wasn’t the intent when AI-based expert systems first appeared on the scene. Built upon proven knowledge in the form of axioms, they applied deductive reasoning to known truths (knowledge) in order to draw conclusions and recommendations. Any uncertainty was clearly indicated in the results. But as anti-induction philosophers like Russell warned, we are now feeling the negative effects of our growing reliance on data—especially when much of that data is devoid of context, or worse, devoid of knowledge.
Getting back on course: The forgotten art and science of abduction
Today we have the opportunity to dust off the old yet oft-neglected notion of “abductive reasoning.” Postulated by Charles Sanders Peirce, an American scientist and philosopher, in the late 19th century, it provides a means for strengthening both induction and deduction by generating and testing hypotheses along with explanatory crumb trails based on human sensemaking and plausible reasoning. The result is a set of axioms backed by underlying laws and proven theoretical principles supported by, as opposed to driven by, context-rich data.
This means we no longer need to blindly accept the output of even the most sophisticated AI/ML platforms. In fact, we should not consider any artifact, whether produced by humans or machines, as valid knowledge unless it contains not only supporting data and analyses, including provenance, but also an explanation of the underlying plausibility.
This leaves every so-called “fact” open to what Karl Popper, the Austrian-British philosopher, academic, and social commentator, refers to as “falsifiability.” This may seem downright frightening to many. That’ s where agility and resilience come into play. And it goes back to another aspect of the early days of KM that is even more important today: lessons learned. This requires a willingness to openly admit mistaken assumptions and decisions, no matter how painful or embarrassing. But it is far better to identify and correct mistakes earlier rather than later.
Opportunity for KM
Too many so-called “facts” are being tossed around, devoid of context and driven by emotion. This can lead to wrong conclusions and, as we’ve seen all too often, serious adverse consequences.
The good news is that we’re at a cross-roads in human history where we can combine the massive computational power of computers with the expertise, insights, intuition, and sensemaking of more than 8 billion human minds. Humans teaching machines and machines teaching humans accomplish more together than either could alone.
Are you data-driven or knowledge-driven? Hopefully, your answer is, “Both.”