Inductive reasoning is special

From The World of Being to a World of Becoming

All of us have an immediate grasp of what “reasoning” feels, a faculty as elemental as sight or touch, and just as readily available to us. When Descartes proclaimed “I think, therefore I am,” he established reasoning as foundational, the one undeniable truth.

There are broadly two kinds of reasoning: deductive and inductive.

Consider the following:

Suppose all swans are white.
Suppose you hear a bird, and it is a swan.
Therefore, you conclude that this bird is white.

This is deduction: from general rules (premises), we derive certain conclusions. If the premises are true, the conclusion must be true.

Formal logic codifies deductive reasoning, with a formal language (symbols, syntax), rules of inference (what transformations preserve truth) and semantics (interpretations assigning truth-values). In the history of evolution of ideas, one can draw a direct line from Aristotle, who first studied formal logic, to Frege to Russell to Wittgenstein to Turing and Von Neumman. The foundation of classical computing, thus, are based on laws of deductive reasoning, laws that are a consequence of the structure of language, laws that govern the science of symbols. It has little to do with the real world, which is dynamic, transient, evolving, uncertain.

Now consider the following:

Suppose you observe many swans, and every swan you have ever seen is white. You conclude: “All swans are white.”

This is induction: from specific observed cases, you infer a general rule - even though there is always the possibility of a black swan you haven’t seen yet.

Inductive reasoning allows us to generalize from experience, forming broader expectations and making predictions about future or unseen cases based on patterns in the data we have observed.

With ML, we have, for the very first time, synthesized a very good inductive reasoner. This is a big deal because an inductive reasoner can observe, think, learn in the real world (the world of senses, the world of empiricism, the world of becoming) in a way a classical computer or a deductive reasoner can never.

Modern ML is based on theory of reasoning that is capable of operating on the actual world:

• With probability, we are able to model the world as a dynamic, uncertain, evolving entity where many possible worlds are possible, where events unfold, and as we observe what happens, alternative realities are destroyed.
• Information theory helps us find laws that govern reasoning about uncertainty and information in this world.
• We make machines learn from experience by constantly updating its world model towards a reduction in surprise about what happens.

19 October 2025