Abhinav Saxena
Glossary

Instrument reading

Also called: instrument-based belief, Instrument-Reading Thesis, calibrated reliance

Definition

Coming to know by competently reading a reliable device, where warrant is generated by the reader’s use rather than transmitted from a speaker.

By · Updated · 1 min read

Instrument reading is coming to know something by competently reading a reliable device, where the warrant is generated by the reader’s use rather than transmitted from a speaker. A user reading a barometer comes to know the pressure when conditions are right: the instrument is well calibrated, she is competent with it, the conditions of use fall within those for which the calibration is valid, and she has reason to take all this to be so. No entitlement of interlocution, no assurance, no norm of assertion is in play. The epistemic load lies with her.

My positive proposal about language models is the Instrument-Reading Thesis. When a user comes to a warranted true belief that p on the basis of a model’s output, her standing is that of a competent reader of an instrument: a function of the instrument’s reliability in the conditions of use, her competence in operating and interpreting it, and the absence of defeaters. No warrant is transmitted from the system to the user; her warrant, when present, is generated by her use.

“Instrument” classifies the structure of warrant and is not a demotion. Instruments are genuine sources of knowledge, and perception itself has this structure. Nor does anything turn on the label: Freiman’s technology-based beliefs, Duede’s category between instruments and agents, and Alvarado’s epistemic technologies are live alternatives, and the claim about warrant can be accepted under any of them.

Two differences from testimony follow. A hearer may defer, because her warrant runs through the speaker; an instrument reader cannot, and her appropriate attitude is calibrated reliance. And when an informant misleads, the hearer has social remedies (challenge, a demand for grounds, appeal to others), whereas when an instrument misleads the remedies are engineering ones.

Language models force a refinement: they are instruments without transparent calibration, and degenerate ones. A working barometer satisfies the Tracking Minimum; these systems, on the best current evidence, do not. The record a user projects is all the connection there is (see epistemic opacity).

My essays and papers on it

Paper · Episteme 2026Assertion Without a Speaker: Testimony, Tracking, and Large Language ModelsEssayYour Chatbot Is Not a Witness

Questions people also ask

What is instrument reading in epistemology?

It is coming to know something by competently reading a reliable device. When you read a barometer or a thermometer, your knowledge rests on the instrument’s reliability and on your skill in using it, not on anyone vouching for the reading.

How is reading an instrument different from accepting testimony?

When someone tells you something, you can rely on their word and hold them answerable if they mislead you. An instrument vouches for nothing. When you know through it, the warrant comes from your own competent use, so the checking, and the responsibility for when to check, is yours.

What is an example of instrument reading?

Reading the outside temperature from a well-calibrated thermometer, a pressure from a barometer, or a weight from a scale. Each is a case of knowing through a device whose readings vary with the facts.

Is using ChatGPT a kind of instrument reading?

On my account, yes, in the structure of the warrant it gives you. Relying on a language model is like reading an instrument whose calibration you cannot see, and one that, unlike a working barometer, is not tied to the facts through its own operation. So it calls for checking in proportion to the stakes.

Related terms

Epistemic opacityThe condition of a process whose epistemically relevant elements an agent cannot all know, as with a complex simulation or a trained neural network.Computational reliabilismThe view that reliance on opaque computational processes can be warranted by verification, validation, robustness analysis and a track record, without insight into the process.AI testimonyThe contested thesis that outputs of conversational AI systems can be testimony, transmitting knowledge to users as a competent human informant does.Frequency reliabilityReliability as a high rate of true outputs over a fixed distribution of cases, whatever explains that rate.Cognitive offloadingThe use of physical action or external tools, such as writing a note, setting a reminder or asking a device, to reduce the mental work a task would otherwise require.

Questions that turn on it

QuestionCan AI testify, or is a language model's answer something else?QuestionWho is responsible for a belief that reached you through a machine?QuestionIs an AI chatbot an informant or an instrument?QuestionShould you trust what ChatGPT tells you?QuestionCan you cite ChatGPT as a source?QuestionDoes using AI make you worse at thinking?

In the library

David HumeHume on testimony: 'Of Miracles' (Enquiry X)

Referred to by

KnowledgeTechnologyJustificationTrustThe Tracking MinimumTestimonial chainAutomation biasEpistemic dependence

Further reading

  1. Saxena, A. (2026). Assertion without a speaker: Testimony, tracking, and large language models. Episteme, 1–22.doi.org/10.1017/epi.2026.10160
  2. Goldberg, S. C. (2012). Epistemic extendedness, testimony, and the epistemology of instrument-based belief. Philosophical Explorations 15(2): 181–197.doi.org/10.1080/13869795.2012.670719
  3. Freiman, O. (2024). Analysis of beliefs acquired from a conversational AI: Instruments-based beliefs, testimony-based beliefs, and technology-based beliefs. Episteme 21(3): 1031–1047.doi.org/10.1017/epi.2023.12
  4. Duede, E. (2022). Instruments, agents, and artificial intelligence: Novel epistemic categories of reliability. Synthese 200(6): 491.doi.org/10.1007/s11229-022-03975-6