Patterns, Similarity and Relevance
also Relevance of Patterns
The Popperian position that learning is not extrapolation of patterns from repeated observation, because which patterns, similarities and relevances count is decided by prior critical ideas, not read off the data.
“Patterns,” “similarity,” and “relevance” are three words for one hard problem: judging when an observation fits a pattern well enough, when two things are similar enough, or when one thing is relevant enough to another. CF treats these as the unsolved core of inductive epistemology. Inductivists say the future will resemble the past, but that demands a prior account of which resemblances count, and CF argues no such account has ever been given.
The deeper trouble is logical: for any finite data, infinitely many patterns fit perfectly, and no matter what happens some patterns continue while others break. So data alone cannot pick the right pattern. Strict exact matching avoids the “good enough” judgment but is too brittle for messy real data; building fuzz into laws, or into data via margins of error, just relocates the same “how much?” question. Worse, choosing among the infinitely many perfect-fitting patterns requires prior preferences — abstract reasoning or inborn intuition — which contradicts induction’s own promise to let the data guide us. This is continuous with CF’s theory-laden view of perception.
CF’s resolution, following Popper and Darwin, is to stop founding epistemology on patterns at all and ground it in evolution instead: knowledge grows by conjecture and refutation — guesses plus error-correcting criticism. Patterns and evidence become ordinary objects of conjecture and criticism, not the engine of thought. A pattern is used if critical argument finds nothing wrong with it, never because it was “supported” by instances. This makes the concept a specific application of pattern matching under a criticism-first method, not a rival measure of inductive strength.