-->

NEW EVENT: KM & AI Summit 2025, March 17 - 19 in beautiful Scottsdale, Arizona. Register Now! 

The Dethroning of Deduction

Article Featured Image

For the “Well, technically ...” crowd— and thank you for your service—machine learning also uses abduction, often called “inference” to the best explanation, which identifies the most plausible cause for an observed effect. Like induction, abduction works outside of the rigid framework of deduction. Like induction, it is probabilistic and usually grounded in data.

The trade-off for the uncertainty of machine learning’s responses is that its answers can be much more precise when dealing with real-world issues, what with all its inter-relational messiness. That’s why vastly improved our ability to predict the weather or the best route to take to avoid traffic. AI’s favoring of induction over deduction is the root of its power, for it lets it deal with the specifics that bedevil the application of broad major premises.

Real life—as opposed to games—is complex, which is why it’s no accident that the standard examples of deductive logic use very simple premises, such as “Socrates is mortal.” But try using deductive logic to predict Socrates’ ideas about the nature of bravery or the value of poetry, as he discusses in his famous dialogues.

Machine learning’s success paints a picture of the world that is structurally more like the humanities, in which deductive logic struggles because of the abundance of interrelated, nonquantifiable variables. So much of disciplines such as literature and history is about the multitudinous contingent forces that shape unexpected outcomes, from great novels to savage wars. Still, researchers in physics and other “hard” sciences are also leveraging machine learning to analyze complex phenomena such as the faint, noisy signals of gravitational waves, thus helping to enhance detection and refine models of these ripples in spacetime.

Machine learning is showing us the world as composed of the interrelationships of millions of particulars, each governed by the deductive application of principles, but each interacting in ways too complex to be deduced. Together, they give rise to probabilistic truths that can be more useful and valuable than deductions from eternal laws.

KMWorld Covers
Free
for qualified subscribers
Subscribe Now Current Issue Past Issues