Who Is Speaking? Machine Testimony, Authorless Scripture, and the Unity of Inquiry

Inquiry is a testimonial through and through. A physicist believes the calibration report she did not produce, a historian believes the archivist, and both believe the output of a statistical package that neither wrote. This paper argues that the question of what could unify diverse forms of inquiry rests on a prior question: what makes an utterance a source of knowledge for someone who did not produce it? Whatever unifies inquiries must travel between them by testimony; therefore, a theory of epistemic integration presupposes a theory of testimonial warrants. This method is comparative. Classical Indian categories are used as analytical tools rather than being expounded historically, the contemporary Western epistemology of testimony is engaged on the same footing, and machine speech serves as a shared test case by which each tradition is measured against the other. 

First, the Nyāya account is presented. Verbal testimony is āptopadeśa, the statement of an āpta, one who knows the matter and wishes to communicate it without deception (Nyāyasūtra 1.1.7). Three conditions govern the verbal cognition of a hearer: expectancy, semantic fitness, and contiguity. The Naiyāyikas add a fourth, tātparya, the meaning intended by the speaker [1]. A large language model satisfies the first three as well as any speaker, and it has no tātparya.

Second, contemporary Western epistemology converges on this point from the opposite direction, which matters because it is ordinarily considered the more permissive tradition. Lackey denies that a testifier must believe or know what she asserts, yet still requires that a speaker intends to express communicable content, and she withholds the status of testimony from branches that fall so as to read “HELP ME” to a passing pilot [2]. Sentence-shaped output produced without any intention is not defective testimony. It is not testimony.

Third, the Mīmāṃsā alternative is examined, since it is the one developed account of warranted speech with no speaker. Mīmāṃsā denies the intention condition, recovers the purport of a passage from six text-internal marks alone [1], and holds the Veda to be apauruṣeya, composed by no one, not even by a god [3]. Matilal and Chakrabarti deliberately excluded authorless texts from Knowing from Words as too theological for a secular epistemology of testimony [4]. Machine speech makes that exclusion untenable.

Fourth, it is argued that machine speech cannot occupy the apauruṣeya position. Kumārila allows exactly two grounds for defeating a verbal cognition: a deficiency in the means and contradiction by a subsequent cognition. The Veda escapes both, being uncomposed and supersensible in subject matter [5,6]. Machine speech escapes neither. It is composed, out of authored speech, by a datable process, which makes it author-effaced rather than authorless, and effacement conceals defects rather than preventing them. Its subject matter is the ordinary checkable world, where contradicting cognitions are cheap.

Fifth, a dilemma follows that bears on the theme of the conference. He and Yang propose that warrant rest on a system robustly performing propensities of veracity and caution, whatever its inner structure [7]. This relocates warrant from persons to processes, and processes individuate along with their objects, which is the Buddhist restriction theory (vyavasthā) set against the Nyāya mixture theory (samplava), on which one object may be established through more than one way of knowing [8]. Instruments individuated in this way do not converge, so integration fails by construction. Retaining the person condition avoids this, but it leaves much of contemporary inquiry uncredentialed at its source and recasts disciplinary fragmentation as a fragmentation of trust rather than of method.

Finally, understanding shows what is lost. Understanding why p requires abilities that knowing why p does not, and these are ordinarily acquired by practice rather than by being told [9]. Transmission of any epistemic good turns on uptake conditions already in place in the recipient [10]. As inquiry grows more instrumented, it grows more testimonial; therefore, knowledge scales and understanding do not. Epistemic defragmentation is therefore not a problem of communication. It is a question about the conditions under which one inquirer can stand as āpta for another across a disciplinary boundary.

References

1.         Kunjunni Raja, K. Indian Theories of Meaning; Adyar Library and Research Centre: Madras, 1963.

2.         Lackey, J. Learning from Words: Testimony as a Source of Knowledge; Oxford University Press: Oxford, 2008.

3.         Matilal, B. K. The Word and the World: India's Contribution to the Study of Language; Oxford University Press: Delhi, 1990. 

4.         Matilal, B. K.; Chakrabarti, A., Eds. Knowing from Words: Western and Indian Philosophical Analysis of Understanding and Testimony; Kluwer: Dordrecht, 1994.

5.         Kumārila Bhaṭṭa. Ślokavārttika, codanāsūtra section, vv. 95–101; translated by G. Jha.

6.         Taber, J. A Hindu Critique of Buddhist Epistemology: Kumārila on Perception; RoutledgeCurzon: Abingdon, 2005.

7.         He, J.; Yang, C. Testimony by LLMs. AI & Society 2025, 40, 6201–6213.

8.         Matilal, B. K. Perception: An Essay on Classical Indian Theories of Knowledge; Clarendon Press: Oxford, 1986.

9.         Hills, A. Moral Testimony and Moral Epistemology. Ethics 2009, 120, 94–127.

10.        Grimm, S. R. Transmitting Understanding and Know-How. In What the Ancients Offer to Contemporary Epistemology; Hetherington, S., Smith, N. D., Eds.; Routledge: New York, 2020; pp. 124–140.