Whose Voice Counts as Knowing

Most of what a person thinks they know, they never checked themselves. It arrived secondhand, from someone else's mouth, and got accepted or rejected based on a much quieter judgment: whether that particular someone seemed worth believing.

Few traditions took that quiet judgment as seriously, or built as rigorous a system around it, as early Islamic scholarship did with the isnād, the chain of transmission attached to every reported saying of the Prophet. A hadith was never accepted on the strength of its content alone. It arrived with a list: I heard this from A, who heard it from B, who heard it from C, tracing back, link by link, to the original source. Scholars developed an entire discipline, ʿilm al-rijāl, the science of men, dedicated purely to evaluating each name in that chain: was this person known for accuracy, for honesty, for a good memory, for actually having met the person they claimed to have heard from. A chain was graded by its weakest link. One unreliable name, however strong the rest, capped the reliability of everything that followed it.

Classical Indian epistemology arrived at a structurally similar concern from a different direction. The Nyāya school listed testimony, śabda, as one of only four legitimate sources of knowledge, alongside perception, inference, and comparison, but testimony earned that status only when it came from a reliable, trustworthy speaker. An unreliable source reporting something true did not, in this framework, generate real knowledge. The content could be accidentally correct. The source still failed the test.

Contemporary philosophy of testimony draws a related distinction that neither older system quite phrased this way: the difference between expertise and authority. A person can be an expert across an entire field, broadly competent, while having authority, in the strict sense, only over specific individual claims within it. Someone can be a reliable chemist and still be wrong about one particular compound. The isnād system effectively anticipated this by grading narrators individually, chain by chain, rather than granting one blanket trust to a person for everything they ever said.

None of these systems solved the underlying problem, because it may not be solvable: a thought accepted on someone else's word is never as secure as one checked firsthand, and most thoughts, most of the time, are accepted exactly that way. What they did instead was refuse to let that acceptance happen silently. A chain had to be named. A source had to be evaluated, by title, by track record, by something more specific than how confident they sounded saying it.