Verification Is Not Knowledge: Science needs more than answers

The central argument is that a verified scientific answer is not automatically knowledge. Debates about AI often frame human understanding as an optional extra: if a system proves a theorem or produces a successful treatment, why care whether people understand how? But the examples reveal practical risks. An AI-designed cancer treatment might pass a trial by exploiting a flaw in the trial, while authors unable to explain a paper’s claims leave no one able to answer challenges or investigate failures. Understanding, criticism, and accountability are therefore not sentimental obstacles to progress; they help science detect errors and make results usable.

This output-first view reflects the “value-free ideal,” strengthened after World War II, that science can reach truth through impartial procedures while excluding social and ethical judgment. The essay argues that this ideal is mistaken: values shape which questions matter, how evidence is weighed, and which findings become part of shared knowledge. Mathematics illustrates the point. Automated systems may verify proofs, but people still choose which problems to pursue and decide what results are significant. Fermat’s Last Theorem gained importance through its history and its links to other mathematical ideas, not correctness alone. Scientific objectivity likewise depends on communities whose members share language, draw on others’ expertise, and criticize claims. Replacing human judgment with automated procedures risks generating answers without meaningful questions, weakening the expertise and collaboration needed to guide science toward human interests.